Proceedings of the 2nd International Conference on Emerging Technologies and Intelligent Systems: ICETIS 2022, Volume 2 3031204301, 9783031204302

Table of contents :
Preface
Committes
Contents
Artificial Intelligence and Data Science
ECG Signal Classification Based on Neural Network
1 Introduction
2 Background
3 Methodology
3.1 ECG Dataset
3.2 ECG Dataset Pre-processing
3.3 ECG Classification
4 Results and Discussion
4.1 Classification Evaluation Metrics
4.2 Confusion Matrix
4.3 Performance Evaluation
5 Conclusion
References
A Review of Long Short-Term Memory Approach for Time Series Analysis and Forecasting
1 Introduction
2 Background Study
2.1 Long Short-Term Memory Approach
2.2 Time Series
3 Related Works
4 Discussion
5 Conclusion
References
Applying Modified LBP for 2D ECG Images Classification
1 Introduction
2 Literature of Related Works
3 Methodology
3.1 Preprocessing
3.2 Extracting the Features
3.3 Modified LBP (MLBP)
4 MIT-BIH and PTB Diagnostic ECG Datasets
5 Information on Spatial Distribution (SDI) of Dominate Patterns
6 Experiential Findings
7 Conclusion
References
Integrating Ontology with Imaging and Artificial Vision for a High-Level Semantic: A Review
1 Introduction
2 Review Papers
3 Works Paper
3.1 Integration of Ontology with Human Behavior Analysis
3.2 Integration of Ontology with Image Recognition/Classification
3.3 Integration of Ontology with Image Retrieval
3.4 Integration of Ontology with Risk Prediction
4 Conclusion
References
Agent-Based Simulations for Aircraft Boarding: A Critical Review
1 Introduction
2 Background Information
2.1 Boarding Strategies
3 Literature Review
3.1 Studies Evaluating Various Boarding Strategies
3.2 Studies Evaluating Infrastructural Changes or Proposing New Methods
3.3 Studies Evaluating Boarding with Restrictions – COVID-19
4 Conclusion
References
Adaptive Systems and Materials Technologies in Civil Engineering: A Review
1 Introduction
1.1 What Are the Benefits of Smart Structures?
2 Research Objectives
3 Smart Structure System (Adaptive System)
3.1 Smart Structure Technology for Civil Engineering
3.2 Type of Smart Structural System of Civil Engineering
3.3 Passive System
3.4 Semi-active System
3.5 System of Active Control
3.6 Hybrid Control System
4 Smart Materials
4.1 Characteristics of Smart Materials
4.2 Application of Smart Materials in Civil Engineering
5 Conclusion
References
A New Method for EEG Signals Classification Based on RBF NN
1 Introduction
2 The Description of Dataset
3 Literature Related Works
4 Changing the Features of the Domain
5 Findings and Discussion
6 Conclusion
References
Intelligent Fault Detection of Rotating Machinery Using Long-Short-Term Memory (LSTM) Network
1 Introduction
2 An Overview of the LSTM Network
3 Experimental Setup
4 The Proposed Method
5 Results and Discussion
6 Conclusions
References
What People Post During the Movement Control Order (MCO): A Content Analysis of Intagram’s Top Posts
1 Introduction
2 Literature Review
3 Research Method
4 Discussion
5 Conclusion
References
A Comparative Study on the Recognition of English and Arabic Handwritten Digits Based on the Combination of Transfer Learning and Classifier
1 Introduction
2 Related Work
3 Data-Set
4 Method
4.1 Dataset Usage
4.2 The Stages of the Workflow
5 Implementations
6 Results and Discussion
7 Conclusion
References
Predictive Analytics for Oil and Gas Asset Maintenance Using XGBoost Algorithm
1 Introduction
2 Literature Review
2.1 Big Data in the Oil and Gas Industry
2.2 Predictive Analytics in Oil and Gas Industry
2.3 Machine Learning (ML)
3 Methodology
3.1 Data Acquisition
3.2 Data Exploration and Pre-processing
3.3 Data Splitting
3.4 Data Training
3.5 Data Testing and Evaluation
4 Conclusion and Recommendations
References
Event Detection and Information Extraction Strategies from Text: A Preliminary Study Using GENIA Corpus
1 Introduction
2 Literature Review
2.1 Event Detection and Extraction
2.2 TEES 2.1 Extraction Model
2.3 EVEX Text Mining Resource in ST’13
2.4 DeepEventMine Extraction Model
3 Methodology
4 Result and Discussion
5 Conclusion and Future Recommendation
References
Human Evacuation Movement Simulation Model: Concepts and Techniques
1 Introduction
2 Related Works
2.1 Evacuation and Human Movements
2.2 Modelling and Simulation (M&S)
2.3 Techniques: Agent-Based and Social Force
2.4 Existing Conceptual Models
3 Evacuation Simulation Model
4 Discussion, Conclusion and Future Works
References
A Data Mining Approach to Determine Prospective Debtor of Unsecured Credit (Case Study: Bank XYZ in Indonesia)
1 Introduction
2 Method
3 Result and Discussion
3.1 Data Collection
3.2 Data Pre-processing
3.3 C4.5 Implementation
3.4 Model Validation
3.5 Result Analysis
4 Conclusion
References
Date Palm Leaves Discoloration Detection System Using Deep Transfer Learning
1 Introduction
2 Related Works
3 Method
4 Results and Discussion
5 Conclusion
References
Solving Drinking-Water Challenges: Supply and Temperature in a Smart Poultry Monitoring System Using IoT System
1 Introduction
2 Proposed System Overview
2.1 Proposed Hardware
2.2 Proposed Software
2.3 Implementation
3 Results
4 Conclusion
References
Offline Marker-Less Augmented Reality Application for Exploring Threatened Historical Places
1 Introduction
2 Related Work
3 Methodology
3.1 Data Collection
3.2 Software Analysis
3.3 Software Design
3.4 User Interface Designing
4 Implementation
5 Testing
6 Conclusion and Future Work
References
A Robust Tuned K-Nearest Neighbours Classifier for Software Defect Prediction
1 Introduction
2 The k-Nearest Neighbors Algorithm
3 Related Works
4 Robust Tuned K-Nearest Neighbors Classifier for Software Defect Prediction
5 Results and Discussion
6 Conclusion
References
Smart Virtual Robot Automation (SVRA)-Improving Supplier Transactional Processes in Enterprise Resource Planning (ERP) System: A Conceptual Framework
1 Introduction
2 Literature Review
2.1 Methodology Fit
2.2 Technology Fit
3 Methodology
3.1 Requirement Development
3.2 Design and Development of RPA
3.3 Validation and Confirmation Process and Development of RPA
4 Limitations, Conclusion and Future Work
References
Recent Trends in Software Engineering
Design and Implementation of Modified Vedic Multiplier Using Modified Decoder-Based Adder
1 Introduction
2 Conventional Architecture of Vedic Multiplier
2.1 Vedic Multiplier
2.2 Conventional 2 × 2 Bit, 4 × 4 Bit Vedic Multiplier
3 Proposed Vedic Multiplier
3.1 Modified Architecture of Vedic Multiplier
4 Result and Analysis
5 Conclusions
References
Design and FPGA Implementation of Matrix Multiplier Using DEMUX-RCA-Based Vedic Multiplier
1 Introduction
2 Conventional Architecture of Vedic Multiplier and Full Adder
3 Proposed Vedic Multiplier
4 Result and Analysis
5 Conclusion
References
Analysis and Modeling of Brushless DC Motor PWM Control Technique Using PSIM Software
1 Introduction
2 Control Techniques of PMBLDC Motor
3 Proposed Model
4 Structure and Building of the Proposed Model
4.1 Mathematical Model of BLDC Motor
4.2 Speed Control of BLDC Motor
5 Simulation Results and Discussion
6 Conclusions
References
Single-Bit Architecture for Low Power IoT Applications
1 Introduction
1.1 Background of 6TSRAMC
1.2 Production Charges
2 Literature Review
3 Low Power Reduction Techniques
3.1 Low Power Sleep Transistor Technique (LPSTT)
3.2 Low Power Forced Stack Technique (LPFST)
3.3 Low Power Sleep Stack Technique (LPSST)
3.4 Low Power Dual Sleep Technique (LPDST)
4 Proposed Single Bit Architecture
4.1 Circuit of Write Driver (CoWD) Working and Schematic
4.2 6TSRAMC Working and Schematic
4.3 Sense Amplifier (SA)
5 Result Analysis
6 Conclusion
References
Hybrid Fuzzy Logic Active Force Control for Trajectory Tracking of a Quadrotor System
1 Introduction
2 Literature Review/Related Work
3 Methodology
3.1 Quadrotor Modeling
3.2 Proposed Fuzzy Logic Active Force Control with (FLAFC)
3.3 Simulation
4 Results and Discussion
5 Conclusion
References
Semantic Analysis of Moving Objects in Video Sequences
1 Introduction
2 Related Works
3 Background of the Proposal
3.1 Semantic Analysis Methods
3.2 Features of Video
3.3 Extraction of Video Features
3.4 Machine Learning at Video Processing
3.5 Diagnosing Moving Object Status
4 Practical Part
4.1 Input Video
4.2 Extraction Feature
4.3 Object Detection
4.4 Object Diagnosing
4.5 Machine Learning
5 Conclusions
References
Improved Automatic License Plate Recognition System in Iraq for Surveillance System Using OCR
1 Introduction
2 Related Works
3 Iraqi License Plate
4 Methodology
5 Results
6 Conclusions
References
Do We Use the Right Elements for Assurance Case Development?
1 Introduction
2 Related Work
3 Results and Discussion
4 Conclusion
References
Development of a Mobile Application for Scheduling Electric Vehicle Charging in Wind Energy Powered Facility
1 Introduction
2 Literature Review and Related Works
3 Method and Materials
4 Results and Discussion
4.1 Application Architecture
5 Conclusion
References
Evaluating Websites Audit Tools: A Case Study of the Amazon Website
1 Introduction
2 Literature Review
3 Methodological Approach
3.1 Website Grader Tool (WGT)
3.2 SEOptimer Tool
3.3 Qualidator Tool
4 Conclusion
References
Blockchain and Internet of Things (IoT): A Disruptive Integration
1 Introduction
2 Blockchain Structure and Features
2.1 Blockchain Description
2.2 Consensus Mechanism
2.3 Smart Contract
2.4 Blockchain Feature
2.5 Blockchain Classification
3 Blockchain Effect on IoT
4 Blockchain Applications in IoT
5 Conclusion
References
Emerging Technologies in Education
Implementing UX Model at Dijlah University College
1 Introduction
2 Our Methodology
3 Experiments and Rsults
3.1 Design UX Model
3.2 User Experience Application
3.3 User Experience Tools
3.4 Evaluation Tools
4 Analysis and Test
4.1 The Results of SortSite Comparisons
4.2 Classification and Descriptive
4.3 Data Distribution
5 Discussion
6 Conclusion
References
Integration of Face-to-Face Instruction and CALL in Computer-Assisted Cooperative Learning
1 Introduction
2 Literature Review
3 Theoretical Framework: Integration of Face-to-Face Instruction and CALL
4 Methodology
4.1 Research Design
4.2 Participants
4.3 Process
4.4 Data Collection
4.5 Data Analysis
4.6 Findings
4.7 Mode
5 Model of Integration
6 Distribution of Learning Content and Objective and Assignment of Purpose
7 Teaching Method
8 Involvement of Learning Objects
9 Location
10 Discussion
11 Conclusion and Implications
References
Smart Techniques for Moroccan Students’ Orientation
1 Introduction
2 Related Work
3 Background and Methods
3.1 Unsupervised Learning
3.2 Reinforcement Learning
3.3 Supervised Learning
4 Implementation and Results
5 Conclusion and Future Work
References
A General and Theoretical Background of English Academic Writing with Reference to Saudi EFL Context
1 Introduction
2 Academic Writing
2.1 Definition of Academic Writing in Higher Education
2.2 English Writing Teaching Approach in the Saudi Context
2.3 The Nature of English Writing
2.4 The Complexity of Writing for EFL Students
2.5 Differences in Writing Between English and Arabic Language
2.6 Possible Factors Influencing English Academic Writing
3 Theoretical Background of English EFL Writing
3.1 Contrastive Rhetoric Theory
3.2 Social Constructionism Theory
3.3 Connectivism Theory
4 Issues Faced by EFL Students in Academic Writing
4.1 Linguistic Issues
4.2 Cultural and Psychological Issues
4.3 Instructions’ Strategies and Practices Issues
4.4 Teaching and Learning Methods
4.5 Learning Environment Issues
4.6 Adaption of Technology in the Higher Education Learning Environment
4.7 Technology-Based Environment
4.8 Academic Writing and Technology-Based Environment
5 Conclusion
References
Measuring Educator Satisfaction of Learning Analytics for Online Learning Systems in Malaysia
1 Introduction
2 Literature Review
2.1 Online Learning Systems
2.2 Learning Analytics
2.3 Information System (IS) Success Model
3 Research Hypotheses and Methodology
3.1 Hypotheses
3.2 Methodology
4 Data Analyses
5 Discussion
6 Conclusion
References
Digital Support, Teacher Support, and Blended Learning Performance: Investigating the Moderating Effect of Gender Using Multigroup PLS-SEM Analysis
1 Introduction
2 Methodology
2.1 Measurements
2.2 Sample and Data
3 Data Analysis
4 Conclusion and Discussion
4.1 Practical Implications
5 Conclusion
References
The Research on Design and Application of Dynamic Mathematics Integrable Ware Design Model in Junior High School
1 Introduction
2 Research on Design Model of Dynamic Mathematical Integrable Ware
3 Multiple Representation Strategy
4 Method and Research Framework
5 Result and Discussion
5.1 The Results and Analysis of the Pre-test on Student Achievement
5.2 The Results and Analysis of the Post-test
5.3 The Concept of Multiple Representation of Accumulates a Teaching Investigation
6 Conclusion
6.1 Research Conclusions
6.2 Research Reflection
6.3 Research Prospects
References
Exploring the Accuracy of Mathematics Students on the Final Semester Assessment Based on Racsh Model Analysis in Timor-Leste
1 Introduction
2 Student Accuracy
3 Methodology
3.1 Research Subject
3.2 Learning Implementation and Instrument
3.3 Research Framework
3.4 Data Analysis Techniques
4 Result and Finding
4.1 Wright Map (Person-Item)
4.2 Scalogram
5 Conclusion
References
ZOOM-ing into a New Pedagogy: Permanent Adoption of Online Teaching and Learning in Private Higher Education Institution in Malaysia
1 Introduction
2 Literature Review
2.1 Hypotheses Development
3 Research Method and Data Analysis
4 Discussion and Conclusion
References
Analytical Review and Study on Student Performance Prediction: A Challenging Overview
1 Introduction
2 Literature Review
2.1 Classification of Prediction Techniques
3 Identified Research Gaps
4 Analysis and Discussion
4.1 Analysis Based on Various Techiques
4.2 Analysis by Means of Toolset
4.3 Analysis Based on Publication Year
4.4 Analysis in Terms of Employed Datasets
4.5 Analysis Based on Evaluation Measures
4.6 Values of Performance Metrics Analysis
5 Conclusion
References
Artificial Intelligence Applications in Education
1 Introduction
2 The Concept of Skills of Using Artificial Intelligence Applications in Education
3 Smart Methods in Artificial Intelligence
4 Research Methodology
5 Research Tools
6 Statistical Analysis of the Scale Items
6.1 Items Recognition Power
6.2 Reliability
7 The Final Form of the Scale
8 Statistical Means
9 The Results
10 Discussion
11 Recommendation
12 Suggestion
References
Setting Up a Dedicated Virtual Reality Application for Learning Critical Thinking and Problem-Solving Skills
1 Introduction
2 Related Works
3 Case Study, Problem-Solving Skills and Critical Thinking
3.1 Problem-Solving Skills and Critical Thinking
3.2 Case Study and Relationship Between Problem-Solving Skills and Critical Thinking
4 Design
4.1 Unity 3D Game Engine
4.2 Game Objects
4.3 Application
5 Experiment
5.1 Headset
5.2 Implementation
6 Challenges and Discussion
7 Conclusion
References
Online Learning Readiness and Satisfaction Among Undergraduate Students
1 Introduction
2 Literature Review
3 Methodology
4 Results
5 Discussion and Recommendation
6 Conclusion
References
User Acceptance of Augmented Reality in Education: An Analysis Based on the TAM Model
1 Introduction
2 Literature Review
2.1 Technology Acceptance Model–TAM
2.2 Hypotheses Development
3 Methodology
3.1 Questionnaire and Research Sample
3.2 Data Collection
4 Results
4.1 Measurement Model
4.2 Outer Model
5 Discussion
6 Conclusion
References
UTAUT2 Model to Explain the Adoption of Augmented Reality Technology in Education: An Empirical Study in Morocco
1 Introduction
2 Literature Review
2.1 UTAUT 2
2.2 Hypotheses Development
3 Methodology
3.1 Questionnaire and Data Collection
3.2 Respondent Demographics
4 Results
4.1 Analysis of Construct Validity
4.2 Analysis of the Structural Equation Model
5 Discussion
6 Conclusion
References
Question Guru: An Automated Multiple-Choice Question Generation System
1 Introduction
2 Automated Multiple-Choice Questions (MCQs) Generation Systems
3 System Design and Development
4 Implementation
4.1 Stem Extraction
4.2 Ranked Sentences Generation
4.3 Keyword Extraction
4.4 Key Identifiers Extraction
4.5 Answer Generation for Each Stem
4.6 Distractor Generation
4.7 Word Vector for Answer
4.8 Generating Distractors
5 Results and Discussion
6 Conclusion and Future Work
References
Importance and Implications of Theory of Bloom's Taxonomy in Different Fields of Education
1 Introduction
2 The Learning Domains of the Bloom’s Taxonomy
3 Importance and Applications in Different Fields of Education of This Theory
4 Conclusion
References
Learning Chemistry with Interactive Simulations: Augmented Reality as Teaching Aid
1 Introduction
2 Related Works
3 Vuforia Package for Unity
3.1 Image
3.2 Object
3.3 Vumarks
4 Unity Game Engine
4.1 Overall Design
4.2 Material Contents
5 Case Study: XR Chemistry LAB
5.1 Interface and Structure
5.2 Image Target
6 Conclusion
References
Upskilling Educators for Pandemic Teaching: Using Video Technology in Higher Education
1 Introduction
2 Literature Review
3 Methods
4 Results and Findings
5 Discussion and Conclusion
References
Analysing English for Science and Technology Reading Texts Using Flesch Reading Ease Online Formula: The Preparation for Academic Reading
1 Introduction
2 Literature Review
2.1 Types of Texts
2.2 Readability
3 Research Methodology
3.1 Research Design
3.2 Theoretical Framework
3.3 Data Analysis
4 Findings
4.1 Types of Reading Texts
4.2 Readability Level
4.3 Length of Reading Texts
5 Discussion
6 Conclusion and Recommendation
References
Research on Continued Intention to Adopt E-Learning in Beijing University During Covid-19 Epidemic in China
1 Introduction
2 Literature Review
2.1 Continued Intention to Adopt
2.2 Perceived Usefulness
2.3 Perceived Ease of Use
2.4 Perceived Enjoyment
2.5 Perceived Convenience
2.6 Hypotheses Development
3 Research Method
4 Discussion and Conclusion
4.1 The Reliability Test
4.2 Pearson’s Correlation Analysis
4.3 Multiple Regression Analysis
References
Intelligent Health Informatics
Real-Time Healthcare Surveillance System Based on Cloud Computing and IoT
1 Introduction
2 Related Works
3 The Planned System Model
4 Dataset
5 System Methodology
5.1 ECG Signals Filtering
5.2 Extraction of Features
6 System Design and Implementing
6.1 Process of ECG Data Utilizing MobiDevs
7 Designing and Implementation of CAS
8 Processing of Alarms
9 Experimentation Based Findings
10 Conclusion
References
A Review of Agent-Based Model Simulation for Covid 19 Spread
1 Introduction
2 Literature Review
2.1 Contain the Virus Through Non-pharmaceutical Interventions (NPI)
2.2 Virus Spread’s Impact on the Economy
2.3 Contain the Virus Through Pharmaceutical Interventions
3 Conclusion
Appendix A
References
Dengue in Bangladesh: Strategic Assessment Considering the Future Outbreak and Hospital Scenario
1 Introduction
2 Dengue Epidemiology
2.1 Global Situtation of Dengue
2.2 Dengue in Bangladesh
2.3 Factors Influencing Transmission
3 Forecasting Dengue
4 Hospital Scenario
4.1 Current Situation
4.2 Future Scenario (Prediction)
5 Instructions to the Government
5.1 Enrichment of Data
5.2 Regularity of Data
5.3 Taking Initiatives to Face the Upcoming Outbreak
5.4 Establishment of New Unit Under DGHS
5.5 Development of a User-Friendly Application
6 Conclusion
References
Covid-19 Vaccine Public Opinion Analysis on Twitter Using Naive Bayes
1 Introduction
2 Related Work
3 Methodology and Methods
3.1 Data Collection
3.2 Data Annotation
3.3 Data Preprocessing
3.4 Data Processing
4 Results and Discussions
5 Conclusion
References
Medical Application of Deep Learning-Based Detection on Malignant Melanoma
1 Introduction
2 Related Works
3 Methodology
3.1 Dataset
3.2 Networks Fine-Tuning
3.3 Evaluation Metrics
4 Result and Discussion
5 Conclusion
References
Privacy-Preserving Data Aggregation Scheme for E-Health
1 Introduction
2 Related Work
3 System Models and Design Goals
3.1 Network Model
3.2 Attack Model
3.3 Design Goals
4 Proposed Scheme
4.1 Overview
4.2 Key Distribution
4.3 Data Encryption and Submission
4.4 Aggregation on Patient Data by the Server
4.5 Sending Aggregation Request and Data Decryption
5 Privacy and Security Analysis
6 Performance Evaluation
7 Conclusion
References
Classification of Skeletal Muscle Fiber Types Using Image Segmentation
1 Introduction
2 Materials and Methods
2.1 Slice Preparation
2.2 Image Segmentation
2.3 Image Analysis
3 Results and Discussion
3.1 Slice Preparation
3.2 Image Segmentation
3.3 Image Analysis
4 Conclusions
5 Recommendations
References
Modeling the Intention to Use AI Healthcare Chabot’s in the Indian Context
1 Introduction
2 Theoretical Background
2.1 TAM Model
2.2 Anthropomorphism Theory
2.3 Social Presence Theory
3 Methodology
4 Developing Conceptual Framework
4.1 Awareness and Perception of Chatbots
4.2 Trust & Mistrust in Chatbots
4.3 Trust & Mistrust in Doctors and the Healthcare System
4.4 Health Seeking Behavior and CAM
4.5 Intention to Use Chatbots
5 Developing the Propositions
6 Conclusion
References
Exploring the Technology Acceptance of Wearable Medical Devices Among the Younger Generation in Malaysia: The Role of Cognitive and Social Factors
1 Introduction
2 Problem Statement
3 Conceptual Framework and Hypotheses Development
3.1 Perceived Usefulness and Intention to Use
4 Research Methodology
4.1 Participants and Data Collection
4.2 Data Analysis Procedure
4.3 Descriptive Analysis
4.4 Analysis of Measurement Items
4.5 Structural Equation Modeling
4.6 Measurement Invariance of Composite Model (MICOM)
4.7 Multigroup Analysis
5 Discussion
6 Conclusion and Limitations
References
Can LSTM Model Predict the Moroccan GDP Growth Using Health Expenditure Features?
1 Introduction
2 Literature Review
3 Methodology
3.1 Data Description and Preprocessing
3.2 Model
4 Result
4.1 Experiment
4.2 Results and Discussion
5 Conclusion
References
Author Index

Citation preview

Lecture Notes in Networks and Systems 573

Mohammed A. Al-Sharafi Mostafa Al-Emran Mohammed Naji Al-Kabi Khaled Shaalan   Editors

Proceedings of the 2nd International Conference on Emerging Technologies and Intelligent Systems ICETIS 2022, Volume 2

Lecture Notes in Networks and Systems Volume 573

Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Fernando Gomide, Department of Computer Engineering and Automation—DCA, School of Electrical and Computer Engineering—FEEC, University of Campinas—UNICAMP, São Paulo, Brazil Okyay Kaynak, Department of Electrical and Electronic Engineering, Bogazici University, Istanbul, Turkey Derong Liu, Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, USA Institute of Automation, Chinese Academy of Sciences, Beijing, China Witold Pedrycz, Department of Electrical and Computer Engineering, University of Alberta, Alberta, Canada Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Marios M. Polycarpou, Department of Electrical and Computer Engineering, KIOS Research Center for Intelligent Systems and Networks, University of Cyprus, Nicosia, Cyprus Imre J. Rudas, Óbuda University, Budapest, Hungary Jun Wang, Department of Computer Science, City University of Hong Kong, Kowloon, Hong Kong

The series “Lecture Notes in Networks and Systems” publishes the latest developments in Networks and Systems—quickly, informally and with high quality. Original research reported in proceedings and post-proceedings represents the core of LNNS. Volumes published in LNNS embrace all aspects and subfields of, as well as new challenges in, Networks and Systems. The series contains proceedings and edited volumes in systems and networks, spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor Networks, Control Systems, Energy Systems, Automotive Systems, Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems, Robotics, Social Systems, Economic Systems and other. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution and exposure which enable both a wide and rapid dissemination of research output. The series covers the theory, applications, and perspectives on the state of the art and future developments relevant to systems and networks, decision making, control, complex processes and related areas, as embedded in the fields of interdisciplinary and applied sciences, engineering, computer science, physics, economics, social, and life sciences, as well as the paradigms and methodologies behind them. Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago. All books published in the series are submitted for consideration in Web of Science. For proposals from Asia please contact Aninda Bose ([email protected]).

Mohammed A. Al-Sharafi · Mostafa Al-Emran · Mohammed Naji Al-Kabi · Khaled Shaalan Editors

Proceedings of the 2nd International Conference on Emerging Technologies and Intelligent Systems ICETIS 2022, Volume 2

Editors Mohammed A. Al-Sharafi Department of Business Analytics Sunway University Subang Jaya, Selangor, Malaysia Mohammed Naji Al-Kabi Al Buraimi University College Al Buraimi, Oman

Mostafa Al-Emran The British University in Dubai Dubai, United Arab Emirates Khaled Shaalan The British University in Dubai Dubai, United Arab Emirates

ISSN 2367-3370 ISSN 2367-3389 (electronic) Lecture Notes in Networks and Systems ISBN 978-3-031-20428-9 ISBN 978-3-031-20429-6 (eBook) https://doi.org/10.1007/978-3-031-20429-6 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

Over the past ten years, no one has disputed the contribution that intelligent systems and emerging technologies have made to developing digital societies and transforming the knowledge-based economy. The high number of practical new technologies is causing a rapid increase in experimental and theoretical outcomes. These technologies have played a crucial part in many industries, such as healthcare, education, tourism, and marketing. The main aim of the 2nd International Conference on Emerging Technologies and Intelligent Systems (ICETIS 2022) is to provide a forum for academics, researchers, and developers from both academia and industry to share and exchange their latest research contributions and identify practical implications of emerging technologies to advance the wheel of these solutions for global impact. In line with the fourth industrial revolution goals and its impact on sustainable development, ICETIS 2022 is devoted to increase the understanding and impact of emerging technologies on individuals, organizations, and societies, and how intelligent systems have recently reshaped these entities. ICETIS 2022 focuses on the recent innovations in Artificial Intelligence (AI) and Data Science, Advances in Information Security and Networking, Intelligent Health Informatics, Management Information Systems, Educational Technologies, and recent trends in Software Engineering. The ICETIS 2022 was able to attract 200 submissions from 33 different countries across the globe. From the 200 submissions, we accepted 117 submissions, which represents an acceptance rate of 58.5%. Out of the 117 accepted submissions, 61 were selected to be published in this volume. The accepted papers in this volume were categorized into four main themes: Artificial Intelligence and Data Science, Software Engineering, Emerging Technologies in Education, and Intelligent Health Informatics. Each submission is reviewed by at least two reviewers, who are considered experts in the related submitted paper. The evaluation criteria include several issues, such as correctness, originality, technical strength, significance, quality of presentation, interest, and relevance to the conference scope. The conference proceedings are published in Lecture Notes in Networks and Systems Series by Springer, which has a high SJR impact. We acknowledge all those who contributed to the success of ICETIS 2022. We would also like to express our gratitude to the reviewers for their valuable feedback v

vi

Preface

and suggestions. Without them, it was impossible to maintain the high quality and success of ICETIS 2022. As gratitude for their efforts, ICETIS 2022 is partnered with Publons to recognize the reviewers’ contribution to peer review officially. This partnership means that reviewers can opt-in to have their reviews added to their Publons profile.

Subang Jaya, Malaysia Dubai, United Arab Emirates Al Buraimi, Oman Dubai, United Arab Emirates

Volume Editors Mohammed A. Al-Sharafi Mostafa Al-Emran Mohammed Naji Al-Kabi Khaled Shaalan

Committes

Conference General Chairs Dr. Mostafa Al-Emran, The British University in Dubai, UAE. Prof. Khaled Shaalan, The British University in Dubai, UAE.

Conference Organizing Chair Dr. Mohammed A. Al-Sharafi, Sunway University, Malaysia.

Program Committee Chair Dr. Mohammed N. Al-Kabi, Al Buraimi University College, Oman.

Publication Committee Chairs Dr. Mohammed A. Al-Sharafi, Sunway University, Malaysia. Dr. Mostafa Al-Emran, The British University in Dubai, UAE.

vii

viii

Committes

Conference Tracks Chairs Prof. Garry Tan Wei Han, UCSI University, Malaysia. Prof. Vitaliy Mezhuyev, FH JOANNEUM University of Applied Sciences, Austria. Dr. Abdullah B Nasser, University of Vaasa, Finland. Dr. Ahmed Shihab Albahri, University of Information Technology and Communications, Iraq. Dr. Amr Yassin, Ibb University, Yemen. Dr. Baraq Ghaleb, Edinburgh Napier University, UK. Dr. Ibrahim Arpaci, Bandirma Onyedi Eylul University, Turkey.

Members of Scientific Committee Abdallah Namoun, Islamic University of Medina, Saudi Arabia. Abdullah Nasser, University of Vaasa, Finland. Abdulmajid Mohammed Aldaba, International Islamic University Malaysia, Malaysia. AbdulRahman Al-Sewari, Universiti Malaysia Pahang, Malaysia. Ahmed M. Mutahar, Management and Science University, Malaysia. Aisyah Ibrahim, Universiti Malaysia Pahang, Malaysia. Akhyari Nasir, University College TATI, Kemaman, Terengganu, Malaysia. Alaa A. D. Taha, University of Mosul, Iraq. Ali Nasser Ali AL-Tahitah, Universiti Sains Islam Malaysia, Malaysia. Ali Qasem Saleh Al-Shetwi, Fahad Bin Sultan University, Saudi Arabia. Ameen A. Ba Homaid, Universiti Malaysia Pahang, Malaysia. Amir A. Abdulmuhsin, University of Mosul, Iraq. Amr Abdullatif Yassin, Ibb University, Yemen. Baraq Ghaleb, Edinburgh Napier University, UK. Basheer Mohammed Al-haimi, Hebei University, Boading, China. Bokolo Anthony Jnr, Norwegian University of Science and Technology, Norway. Dalal Abdulmohsin Hammood, Middle Technical University, Iraq. Eissa M. Alshari, Ibb University, Yemen. Fadi A.T. Herzallah, Palestine Technical University—Kadoorie, Palestine. Fathey Mohammed, Universiti Utara Malaysia, Malaysia. Garry Wei Han Tan, UCSI University, Malaysia. Gonçalo Marques, Universidade da Beira Interior, Portugal. Hasan Sari, Universiti Tenaga Nasional, Malaysia. Heider A. M. Wahsheh, King Faisal University, Saudi Arabia. Hussam S. Alhadawi, Dijlah university college, Iraq. Hussein Mohammed Esmail Abu Al-Rejal, University Utara Malaysia, Malaysia. Ibrahim Arpaci, Gaziosmanpasa University, Turkey. Joseph Ng, UCSI University, Malaysia.

Committes

ix

Joshua A. Abolarinwa, Namibia University of Science and Technology, Namibia. Kamal Mohammed Alhendawi, Al-Quds Open University. Faculty of Management, Palestine. Kamal Karkonasasi, Universiti Malaysia Kelantan, Malaysia. Khaled Shaalan, The British University in Dubai, UAE. Marwah Alian, Hashemite University, Jordan. Marwan Saeed Saif Moqbel, Ibb University, Yemen Mikkay Wong Ei Leen, Sunway University, Malaysia. Mohamed Elwakil, University of Cincinnati, USA. Mohammed A. Al-Sharafi, Sunway University, Malaysia. Mohammed A.Alsaih, University Putra Malaysia, Malaysia. Mohammed Ahmed Talab, Almaarif university college, Iraq. Mohammed Adam Kunna Azrag, Universiti Teknologi MARA (UiTM), Malaysia. Mohammed N. Al-Kabi, Al Buraimi University College, Oman. Mostafa Al-Emran, The British University in Dubai, UAE. Mukhtar A. Kassem, Universiti Teknologi Malaysia, Malaysia. Nejood Hashim Al-Walidi, Sanaa University, Yemen. Noor Akma Abu Bakar, Tunku Abdul Rahman University College (TARC), Malaysia. Noor Al-Qaysi, Universiti Pendidikan Sultan Idris, Malaysia. Noor Suhana Sulaiman, University College TATI, Kemaman, Terengganu, Malaysia. Osama Mohammad Aljarrah, University of Massachusetts Dartmouth, USA. Osamah A. M. Ghaleb, Mustaqbal University, Saudi Arabia. Qasim AlAjmi, A’ Sharqiyah university- Oman. Samer Ali Alshami, Universiti Teknikal Malaysia Melaka, Malaysia. Taha Sadeq, Universiti Tunku Abdul Rahman, Malaysia Tang Tiong Yew, Sunway University, Malaysia. Vitaliy Mezhuyev, FH JOANNEUM University of Applied Sciences, Austria.

Publicity & Public Relations Committee Hasan Sari, Universiti Tenaga Nasional, Malaysia. Noor Akma Abu Bakar, Universiti Malaysia Pahang, Kuantan, Malaysia.

Finance Chair Taha Sadeq, Universiti Tunku Abdul Rahman, Malaysia.

Contents

Artificial Intelligence and Data Science ECG Signal Classification Based on Neural Network . . . . . . . . . . . . . . . . . . Bashar Al-Saffar, Yaseen Hadi Ali, Ali M. Muslim, and Haider Abdullah Ali A Review of Long Short-Term Memory Approach for Time Series Analysis and Forecasting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nur Izzati Ab Kader, Umi Kalsom Yusof, Mohd Nor Akmal Khalid, and Nik Rosmawati Nik Husain Applying Modified LBP for 2D ECG Images Classification . . . . . . . . . . . . Anfal Hamid Hammad and Azmi Shawkat Abdulbaqi Integrating Ontology with Imaging and Artificial Vision for a High-Level Semantic: A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Malak Belkebir, Toufik Messaoud Maarouk, and Brahim Nini Agent-Based Simulations for Aircraft Boarding: A Critical Review . . . . Thaeer Kobbaey and Ghazala Bilquise Adaptive Systems and Materials Technologies in Civil Engineering: A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohammed A. Ahmed A New Method for EEG Signals Classification Based on RBF NN . . . . . . Shokhan M. Al-Barzinji, Mohanad A. Al-Askari, and Azmi Shawkat Abdulbaqi Intelligent Fault Detection of Rotating Machinery Using Long-Short-Term Memory (LSTM) Network . . . . . . . . . . . . . . . . . . . . . . . . A. Anwarsha and T. Narendiranath Babu

3

12

22

32 42

53 66

76

xi

xii

Contents

What People Post During the Movement Control Order (MCO): A Content Analysis of Intagram’s Top Posts . . . . . . . . . . . . . . . . . . . . . . . . . . Hong Lip Goh, Wen Hui Foo, Tat Huei Cham, Bee Chuan Sia, and Way Zhe Yap A Comparative Study on the Recognition of English and Arabic Handwritten Digits Based on the Combination of Transfer Learning and Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bashar Al-Saffar, Amjed R. Al-Abbas, and Selma Ay¸se Özel

84

95

Predictive Analytics for Oil and Gas Asset Maintenance Using XGBoost Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Norshakirah Aziz, Mohd Hafizul Afifi Abdullah, Nurul Aida Osman, Muhamad Nabil Musa, and Emelia Akashah Patah Akhir Event Detection and Information Extraction Strategies from Text: A Preliminary Study Using GENIA Corpus . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Mohd Hafizul Afifi Abdullah, Norshakirah Aziz, Said Jadid Abdulkadir, Emelia Akashah Patah Akhir, and Noureen Talpur Human Evacuation Movement Simulation Model: Concepts and Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 Noor Akma Abu Bakar, Siew Mooi Lim, and Mazlina Abdul Majid A Data Mining Approach to Determine Prospective Debtor of Unsecured Credit (Case Study: Bank XYZ in Indonesia) . . . . . . . . . . . . 138 Devi Fitrianah, Anita Ratnasari, S. Ayu Eka Fuji, and Siew Mooi Lim Date Palm Leaves Discoloration Detection System Using Deep Transfer Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 Shamma Alshehhi, Shamma Almannaee, and Maad Shatnawi Solving Drinking-Water Challenges: Supply and Temperature in a Smart Poultry Monitoring System Using IoT System . . . . . . . . . . . . . . 162 Ahmed Y. Mohammed, Harith A. Hussein, and Moceheb Lazam Shuwandy Offline Marker-Less Augmented Reality Application for Exploring Threatened Historical Places . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 Hasan Badir Althewaynee, Maytham M. Hamood, and Harith A. Hussein A Robust Tuned K-Nearest Neighbours Classifier for Software Defect Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181 Abdullah B. Nasser, Waheed Ghanem, Antar Shaddad Hamed Abdul-Qawy, Mohammed A. H. Ali, Abdul-Malik Saad, Sanaa A. A. Ghaleb, and Nayef Alduais

Contents

xiii

Smart Virtual Robot Automation (SVRA)-Improving Supplier Transactional Processes in Enterprise Resource Planning (ERP) System: A Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 Tiong Yew Tang, Narishah Mohamed Salleh, and Mikkay Ei Leen Wong Recent Trends in Software Engineering Design and Implementation of Modified Vedic Multiplier Using Modified Decoder-Based Adder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207 Arti Kumari, Saurabh Kharwar, Sangeeta Singh, Mustafa K. A. Mohammed, and Salim M. Zaki Design and FPGA Implementation of Matrix Multiplier Using DEMUX-RCA-Based Vedic Multiplier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 Balivada Yashwant Kumar, Saurabh Kharwar, Sangeeta Singh, Mustafa K. A. Mohammed, and Mohammed Dauwed Analysis and Modeling of Brushless DC Motor PWM Control Technique Using PSIM Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225 Ola Hussein Abd Ali Alzuabidi, Mohammed Abdulla Abdulsada, and Mohammed Wajeeh Hussein Single-Bit Architecture for Low Power IoT Applications . . . . . . . . . . . . . . 235 Reeya Agrawal, Sangeeta Singh, Mustafa K. A. Mohammed, and Mohammed Dauwed Hybrid Fuzzy Logic Active Force Control for Trajectory Tracking of a Quadrotor System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 Sherif I. Abdelmaksoud, Musa Mailah, and Tang H. Hing Semantic Analysis of Moving Objects in Video Sequences . . . . . . . . . . . . . 257 Emad Mahmood Ibrahim, Mahmoud Mejdoub, and Nizar Zaghden Improved Automatic License Plate Recognition System in Iraq for Surveillance System Using OCR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 Yasir Dawood Salman, Hussam S. Alhadawi, Ahmed Salih Mahdi, and Fahad Taha AL-Dhief Do We Use the Right Elements for Assurance Case Development? . . . . . . 278 Abdul Rehman Gilal, Abdul Sattar Palli, Jafreezal Jaafar, Bandeh Ali Talpur, Ahmad Waqas, and Ruqaya Gilal Development of a Mobile Application for Scheduling Electric Vehicle Charging in Wind Energy Powered Facility . . . . . . . . . . . . . . . . . . . 287 Misbah Abdelrahim, Ammar Ahmed Alkahtani, Gamal Alkawsi, Yahia Baashar, and Sieh Kiong Tiong

xiv

Contents

Evaluating Websites Audit Tools: A Case Study of the Amazon Website . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298 Mohammed Fahad Alghenaim, Nur Azaliah Abu Bakar, and Fiza Abdul Rahim Blockchain and Internet of Things (IoT): A Disruptive Integration . . . . . 308 Nazanin Moosavi and Hamed Taherdoost Emerging Technologies in Education Implementing UX Model at Dijlah University College . . . . . . . . . . . . . . . . . 319 Omar Sabraldeen Aziz, Zeena Tariq, and Zena Hussain Integration of Face-to-Face Instruction and CALL in Computer-Assisted Cooperative Learning . . . . . . . . . . . . . . . . . . . . . . . . . 337 Amr Abdullatif Yassin, Norizan Abdul Razak, and Tg Nor Rizan Tg Mohamad Maasum Smart Techniques for Moroccan Students’ Orientation . . . . . . . . . . . . . . . 361 Morad Badrani, Adil Marouan, Nabil Kannouf, and Abdelaziz Chetouani A General and Theoretical Background of English Academic Writing with Reference to Saudi EFL Context . . . . . . . . . . . . . . . . . . . . . . . 367 Asma Abdullah Alasbali, Harmi Izzuan Baharum, and Zuhana Mohamed Zin Measuring Educator Satisfaction of Learning Analytics for Online Learning Systems in Malaysia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382 Nur Maisarah Shahril Khuzairi, Zaihisma Che Cob, and Thaharah Hilaluddin Digital Support, Teacher Support, and Blended Learning Performance: Investigating the Moderating Effect of Gender Using Multigroup PLS-SEM Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392 Mohammed Ali Al-Awlaqi, Ahmed Mohammed Taqi, Nor Hasliza Binti Md Saad, and Nezar Al-Samhi The Research on Design and Application of Dynamic Mathematics Integrable Ware Design Model in Junior High School . . . . . . . . . . . . . . . . . 402 Jianlan Tang, Jerito Pereira, Shiwei Tan, and Tommy Tanu Wijaya Exploring the Accuracy of Mathematics Students on the Final Semester Assessment Based on Racsh Model Analysis in Timor-Leste . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416 Jerito Pereira, Jianlan Tang, Bento Soares, Rafiantika Megahnia Prihandini, and Tommy Tanu Wijaya

Contents

xv

ZOOM-ing into a New Pedagogy: Permanent Adoption of Online Teaching and Learning in Private Higher Education Institution in Malaysia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426 Zufara Arneeda Zulfakar, Fitriya Abdul Rahim, Nor Haliza Che Hussain, Azrina Ahmad, Cham Tat-Huei, and Eugene Cheng-Xi Aw Analytical Review and Study on Student Performance Prediction: A Challenging Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436 Amlan Jyoti Baruah and Siddhartha Baruah Artificial Intelligence Applications in Education . . . . . . . . . . . . . . . . . . . . . . 451 Hamid S. Al-shakri and Khalid I. Al-Hamdani Setting Up a Dedicated Virtual Reality Application for Learning Critical Thinking and Problem-Solving Skills . . . . . . . . . . . . . . . . . . . . . . . . 459 Mohamed Benrahal, El Mostafa Bourhim, Ali Dahane, Oumayma Labti, and Aziz Akhiate Online Learning Readiness and Satisfaction Among Undergraduate Students . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469 Nurulhaida Hasim and Junaidah Yusof User Acceptance of Augmented Reality in Education: An Analysis Based on the TAM Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481 El Mostafa Bourhim and Oumayma Labti UTAUT2 Model to Explain the Adoption of Augmented Reality Technology in Education: An Empirical Study in Morocco . . . . . . . . . . . . 491 Mohamed Benrahal, El Mostafa Bourhim, Ali Dahane, Oumayma Labti, and Aziz Akhiate Question Guru: An Automated Multiple-Choice Question Generation System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501 Abdul Rehman Gilal, Ahmad Waqas, Bandeh Ali Talpur, Rizwan Ali Abro, Jafreezal Jaafar, and Zaira Hassan Amur Importance and Implications of Theory of Bloom’s Taxonomy in Different Fields of Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515 Abdul Momen, Mansoureh Ebrahimi, and Ahmad Muhyuddin Hassan Learning Chemistry with Interactive Simulations: Augmented Reality as Teaching Aid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526 Mohamed Benrahal, El Mostafa Bourhim, Ali Dahane, Oumayma Labti, and Aziz Akhiate Upskilling Educators for Pandemic Teaching: Using Video Technology in Higher Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536 Chee Heong Lee, Pek Hoon Er, Tiny Chiu Yuen Tey, Priscilla Moses, Phaik Kin Cheah, and Tat-Huei Cham

xvi

Contents

Analysing English for Science and Technology Reading Texts Using Flesch Reading Ease Online Formula: The Preparation for Academic Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546 Lubna Ali Mohammed, Musheer Abdulwahid Aljaberi, Antony Sheela Anmary, and Mohammed Abdulkhaleq Research on Continued Intention to Adopt E-Learning in Beijing University During Covid-19 Epidemic in China . . . . . . . . . . . . . . . . . . . . . . 562 Zhao Ming Sheng, Poh Hwa Eng, and Tat-Huei Cham Intelligent Health Informatics Real-Time Healthcare Surveillance System Based on Cloud Computing and IoT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575 Radwan Nazar Hamed and Azmi Shawkat Abdulbaqi A Review of Agent-Based Model Simulation for Covid 19 Spread . . . . . . 585 Samar Ibrahim Dengue in Bangladesh: Strategic Assessment Considering the Future Outbreak and Hospital Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . 603 Md. Zahidur Rahman and Nur Mohammed Covid-19 Vaccine Public Opinion Analysis on Twitter Using Naive Bayes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613 Samar Ibrahim and Sheriff Abdallah Medical Application of Deep Learning-Based Detection on Malignant Melanoma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627 Abdulkader Helwan and Mohamad Khaleel Sallam Ma’aitah Privacy-Preserving Data Aggregation Scheme for E-Health . . . . . . . . . . . . 638 Matthew Watkins, Colby Dorsey, Daniel Rennier, Timothy Polley, Ahmed Sherif, and Mohamed Elsersy Classification of Skeletal Muscle Fiber Types Using Image Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647 Mehdy Mwaffeq Mehdy, Sarah Raad Mohammed, Nasser N. Khamiss, and Anam R. Al-Salihi Modeling the Intention to Use AI Healthcare Chabot’s in the Indian Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 657 Aishwarya Nagarathinam, Aarthy Chellasamy, N. Elangovan, and Sangeetha Rengasamy Exploring the Technology Acceptance of Wearable Medical Devices Among the Younger Generation in Malaysia: The Role of Cognitive and Social Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 667 Way Zhe Yap, Bee Chuan Sia, Hong Lip Goh, and Tat Huei Cham

Contents

xvii

Can LSTM Model Predict the Moroccan GDP Growth Using Health Expenditure Features? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 680 Ismail Ouaadi and Aomar Ibourk Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691

Artificial Intelligence and Data Science

ECG Signal Classification Based on Neural Network Bashar Al-Saffar1 , Yaseen Hadi Ali1(B) , Ali M. Muslim2 , and Haider Abdullah Ali3 1 Department of Computer Techniques Engineering, Al Salam University College, Baghdad,

Iraq [email protected] 2 Department of Computer Science, Dijlah University College, Baghdad, Iraq [email protected] 3 Department of Computer Engineering Techniques, Madenat Alelem University College, Baghdad, Iraq [email protected]

Abstract. Cardiovascular diseases (CVDs) are the world’s leading cause of mortality. The current method of diagnosing the disease is the analysis of an electrocardiogram (ECG). The physicians find it difficult to accurately diagnose abnormal heart behavior. However, early and precise detection of cardiac abnormalities helps in providing appropriate treatment to patients. The development of automated ECG classification is an emerging tool in medical diagnosis for effective treatment. In this paper, an effective technique based on Artificial Neural Networks (ANN) is described to classify ECG data into two classes: normal and abnormal. In this context, ECG data are obtained from UCI Arrhythmia databases where the classification is conducted using MATLAB platform. The experimental findings demonstrate that the proposed technique achieves a high classification accuracy of 92.477%, allowing it to effectively detect ECG signal abnormalities and implement it to diagnose heart disease. Keywords: ECG classification · Cardiovascular disease · Artificial neural networks · MATLAB

1 Introduction Cardiovascular diseases (CVDs) are the main cause of death worldwide [1]. Heart diseases account for more than 30% of all deaths in humans, with over 17 million people dying each year [2]. According to the world health organization (WHO), cardiovascular diseases caused the death of more than 17.5 million people in 2012 [3]. Moreover, the number of death increased annually which is reached 17.9 million in 2019 [3], representing 32% of all worldwide deaths. Among the various cardiac diseases that fall under the category of cardiovascular diseases are hypertension, heart attacks, strokes and arrhythmia [1]. A classifier that can diagnose CVDs early, may assist to lower death rates by providing timely care. The electrocardiogram (ECG) is a useful tool for checking a © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 3–11, 2023. https://doi.org/10.1007/978-3-031-20429-6_1

4

B. Al-Saffar et al.

patient’s heart state [4]. The term ECG refers to an electrocardiogram, which is an electrical record of the heart’s contractile action that may be easily obtained using electrodes put on the patient’s chest. A heart generates small electrical impulses that travel through the heart muscle [5]. An ECG equipment can detect these impulses and shows the results on paper as a trace. A physician is then interpreting these results. ECG assists in determining the source of chest pain symptoms as well as detecting irregular heart rhythms. ECGs from normal healthy hearts have a unique shape. Any abnormality in the heart rhythm or destruction to the heart muscle can alter the heart’s electrical activity, changing the shape of the ECG. An ECG may be recommended by a doctor for people who are at risk of heart disease due to a history of heart issues, smoking, being overweight, having high cholesterol, diabetes, or having high blood pressure [6]. Since cardiac disorders have a high mortality rate, early diagnosis and exact differentiation of ECG signals are critical for patient therapy [7]. The classification of ECG data using neural network approaches can give valuable information to clinicians in order to validate the diagnosis. After recognizing the abnormality, heart disease can be diagnosed and the patient treated better. The captured ECG signals may have noises such as power line interference, baseline wandering, instability of electrode-skin contact, muscle noise, electrosurgical noise and instrumentation. These noisy data cause cardiac arrhythmias to be misclassified. As a result, before classification, ECG data must be preprocessed [8]. In this study, a modern technique is used to facilitate the automatic early detection of the arrhythmia in order to reduce the number of deaths by applying specific treatments to the detected diseases. This article is based on the Artificial Neural Network (ANN). The ANN is a classification or detection technique that can differentiate between normal and abnormal cardiac arrhythmias. The presented work’s performance and robustness are assessed using the UCI Arrhythmia dataset [9]. This dataset is already processed by denoising the ECG signal, extracting essential information from the ECG input signal, and minimizing the ANN classifier’s training time without sacrificing system accuracy. The article is organized as follows: Sect. 1 provides an introduction, while Sect. 2 gives background knowledge of ECG. The approach employed in this investigation is described in Sect. 3. Section 4 discusses the results in-depth, and the final section is allocated to the conclusion.

2 Background The electrocardiogram (ECG) is a diagnostic tool that captures the electrocardiography activity of the heart over a period of time. It gathers and captures the electrodes attached to the skin of certain biological organisms and saves the relevant contents in a specific format [10]. Several electrodes are often implanted on human limbs during the detecting procedure. These electrodes always appear in pairs, and are referred to as leads. The LL + RL electrode combination, sometimes known as one- or two-lead electrocardiograms, is employed in this research. This approach is frequently utilized in a single diagnostic [11]. And corresponds to the trend of rapid diagnostics which is employed in this paper. In the test result, each electrode will be assigned an ECG signal map. In this research, the signal is unified. Although the ECG signal has a very clear periodicity, it has become a challenging research area for the identification of ECG abnormalities due to the diversity

ECG Signal Classification Based on Neural Network

5

of noise and random elements generated by external sources [12]. As a result, based on earlier work by many researchers and specialists [13], the ECG dataset must be processed before classification. The electrical activity of the human heart is interpreted via electrocardiography. ECG signals are non-stationary waves that fluctuate based on the individual’s cardiac state. A typical ECG cycle, which includes P-wave, QRS complex, and T-wave. P Wave – defines the heart’s atrial depolarization (contraction). PR Segment– Indicates the AV node’s delay. PR Interval – present the whole electrical activity of the heart before the impulse reaches the ventricles. QRS Complex - Indicates ventricular depolarization of the heart, with Q Wave – representing the first negative deflection after the P wave, but before the R wave, R Wave – representing the first positive deflection after the P wave, and S Wave – representing the first negative deflection after the R wave. T Wave – Indicates ventricular re-polarization (heart relaxation). ST-Segment – Because the atrial cells are relaxed and the ventricles are contracted, no electrical activity can be seen. QT interval - represents the period from ventricular depolarization repolarization [8].

3 Methodology This section primarily presents data processing, concepts, and applications. Figure 1 depicts the complete procedure of the whole approach, which consists of two key steps: firstly, the dataset is preprocessed. Secondly, the processed ECG data is immediately fed into the ANN model to finalize the training and classification.

Fig. 1. ECG data flowchart for classification

6

B. Al-Saffar et al.

3.1 ECG Dataset The execution of this work necessitates the collection of a dataset including digitized ECG records for the computational analysis of many distinct patients with 16 diseases. As a result, we made use of the well-known ANN Repository (UCI) arrhythmia database. This study’s datasets will be drawn from the UCI database. A total of 452 ECG signal instances are used. This dataset comprises 279 properties, 206 of which are linearly valued while the remaining are nominally valued. In our research, we will divide the samples into two main categories: normal (245 instances) and abnormal (207 instances) [9]. 3.2 ECG Dataset Pre-processing Preprocessing of raw ECG signals is necessary to reduce noises such as power-line interference, baseline drift, and high-frequency noises caused by muscle contractions and electrode movements, which can interfere with fiducial point detection and heartbeat classification and make a substantial contribution to the overall classification outcome. In a clinical setting, The collected ECG signals are typically combined with various interferences [14]. To extract the useable signal, the original data must be de-noised in order for the classification to be more precise. In the field of ECG denoising, low-pass filters, bandpass filters, and wavelet transform are commonly utilized [15, 16]. In this article, we made use of the processed dataset from the UCI database. Since the ANN has the advantage of automatic feature extraction from within the signal [17], this study only conducts simple filtering on the signal, which can improve the network’s generalization and decrease signal distortion. In addition, there were some missing data which we processed it by utilizing MATLAB software interpolation function. Finally, the processed ECG data are employed directly as input to the ANN model. 3.3 ECG Classification Many authors have utilized various types of neural networks to classify ECG data. In this study, Artificial Neural Networks (ANNs) are employed for the classification. ANNs are self-adaptive, data-driven, non-linear, accurate, fast, scalable, and robust to noise [18]. The advantages of ANN are as follows: (1) it enables a non-linear mapping between inputs and outputs utilizing activation functions such as sigmoid to handle nonlinear problems such as ECG signal classification. (2) It can produce outcomes that are comparable to or better than statistical or deterministic techniques. (3) ANN can simulate the lower frequencies of the ECG, which are fundamentally non-linear, adaptively. (4) ANN reduces the ECG signal’s time-varying and nonlinear noise features [19]. A Multilayer Perceptron (MLP) network was used to create the classification model. An MLP algorithm is a sort of Feed-Forward Neural Network in which the type contains one or more hidden layers. In general, this model can model complicated non-linear data. An MLP neural network is made up of three layers: an input layer (source nodes), one or more hidden layers (computation nodes), and an output layer. In this study, the number of neurons in the input layer, 279 were considered based on the characteristics used for classification. The number of neurons in the output layer is constant since there

ECG Signal Classification Based on Neural Network

7

are only two types of classes: normal and abnormal. Four hidden layers are utilized, with a sigmoid activation function. The recognition process is divided into two stages: training and testing. In the training stage which is 70%, weights are calculated based on the ANN algorithm. The testing portion is 30% and it is used to assess network performance. The classification is done using ANN tool box, MATLAB software package version R2019a to receive processed data representing the ECG signal to be classified as either normal or abnormal (representing cardiac arrhythmia). Patients can be properly treated with medicine and care if the arrhythmia is discovered early.

4 Results and Discussion This section evaluates the performance of the proposed classifier. Researchers utilize a variety of metrics to assess the categorization accuracy of neural networks. Specificity, sensitivity, accuracy, receiver operating characteristic (ROC), and other metrics are utilized in ECG classification. Furthermore, researchers employ the confusion matrix as the main performance metric. Sensitivity, specificity, and accuracy are evaluation metrics derived from the confusion matrix and are discussed in the next subsection. 4.1 Classification Evaluation Metrics Accuracy Accuracy is a data measurement that accurately determines correctness [20]. It is the ratio of the sum of True Positives (TP) and True Negatives (TN) to a total number of data inputs provided. The confusion matrix for signal recognition is required for accuracy computation. The accuracy is calculated mathematically as follows: Accuracy = (TP + TN)/(TN + FN + TP + TN)

(1)

Sensitivity Sensitivity is the sum of all positives recognized as positive by the algorithm [20]. It is the ability of a network to recognize signals belonging to the same class. It is the proportion of True Positives (TP) to the total of True Positives (TP) and False Negatives (FN). It’s also referred to as detection probability, recall, and true positive rate. The confusion matrix for signal categorization is required for specificity calculation. Sensitivity is calculated mathematically as follows: Sensitivity = TP/(TP + FN)

(2)

Specificity Specificity is defined as the proportion of all negatives properly predicted by the algorithm [20]. It is the ratio of the total number of True Negatives (TN) to the total number of

8

B. Al-Saffar et al.

True Negatives (TN) and False Positives (FP). The confusion matrix for signal categorization is required for specificity calculation. The specificity is calculated mathematically as follows: Specificity = TN/(TN + FP)

(3)

where TP is an actual positive value that is predicted positively and corresponds to the actual value, indicating that the patient has the disease and the test is positive. TN represents the negative expected value which corresponds to the actual value, indicating that the patient has no disease and the test is negative. Similarly, FP denotes a false positive value that is incorrectly predicted as negative when the true value is positive, which means that the patient does not have the disease but the test is positive. FN indicates a false negative value that is incorrectly predicted to be positive when the real value is negative, indicating that the patient has the disease but the test is negative. 4.2 Confusion Matrix Figure 2 depicts the confusion matrix produced for signal classification into Normal and Abnormal. A confusion matrix is a technique for determining how effectively a classifier is able to identify groups of different classes. The sensitivity, accuracy and specificity are shown below, along with the features that were properly and wrongly classified based on the confusion matrix. Sensitivity: 0.9303% Specificity: 0.92031% Accuracy: 0.92477% Correctly Classified Instances 418 Incorrectly Classified Instances 34

Fig. 2. The proposed system’s confusion matrix

ECG Signal Classification Based on Neural Network

9

4.3 Performance Evaluation Performance evaluation is a critical task in neural networks. So, in the case of a classification challenge, we may rely on a ROC curve. A ROC curve is a graph that depicts a classification model’s performance overall classification levels. This curve depicts two parameters: True Positive Rate (TPR) and False Positive Rate (FPR). The ROC curve depicts the relationship between sensitivity (TPR) and specificity (1 – FPR). Classifiers that produce curves closer to the top-left corner yield better results, as seen in Fig. 3, representing our model below.

Fig. 3. The provided model’s ROC

To investigate the link between the recognition rate of test data and the number of iterations, we assess the accurate rate of the test set by varying the training durations while keeping the size and learning rate constant. Figure 4 depicts the experimental findings, which indicate that the error rate decreases as the number of iterations increases. According to the graph, the best validation performance is 0.33928 at epoch 25.

5 Conclusion In today’s society, cardiovascular disease is a huge public health issue. The ECG is extremely important in the early detection of cardiac arrhythmia. The classification of ECG signals is crucial in the process of combining medical and computer technology since it aids in the prevention and diagnosis of cardiovascular disease. The Artificial Neural Network (ANN) technique is used in this study to acquire higher-quality ECG classification. From this study, ECG data are classified into two classes i.e., Normal ECG and Abnormal ECG signals. According to the findings of this article, neural networks are suitable options for ECG classification in terms of accuracy on training and testing datasets. In this study, specificity, sensitivity, and accuracy are utilized to assess the classifier’s performance. A confusion matrix can be used to calculate these metrics. The

10

B. Al-Saffar et al.

Fig. 4. Cross entropy vs epochs

ANN model learns outstanding features and automatically completes classification. The developed ANN model shows excellent performance in terms of overall classification accuracy of 92.477%, sensitivity of 93.03%, and specificity of 92.031%. It is shown that the ANN model, which was originally developed to handle two classes, is applicable in the field of signal processing. The proposed approach will reduce the proportion of human deaths caused by heart disease. The outcomes of the testing have resulted in a robust and rapid decision assistance system. However, future research will look at what improvements may be made to the suggested system for identifying normal and abnormal ECG signals in order for it to grow into a system capable of classifying abnormal ECG signals into bradycardia, tachycardia, and so on.

References 1. Oresko, J.J., Jin, Z., Cheng, J., Huang, S., Sun, Y., Duschl, H.: A wearable smartphone-based platform for real-time cardiovascular disease detection via electrocardiogram processing. IEEE Trans. Inf. Technol. Biomed. 14(3), 734–740 (2010) 2. Mc Namara, K., Alzubaidi, H., Jackson, J.K.: Cardiovascular disease as a leading cause of death: how are pharmacists getting involved? Integr. Pharm. Res. Pract. 8, 1 (2019) 3. World Health Organization: Global status report on noncommunicable diseases. World Health Organization (2014) 4. Mukhometzianov, R., Carrillo, J.: CapsNet comparative performance evaluation for image classification (2018). arXiv Preprint arXiv:180511195 5. Kalid, N., et al.: Based on real time remote health monitoring systems: a new approach for prioritization “Large Scales Data” patients with chronic heart diseases using body sensors and communication technology. J. Med. Syst. 42(4), 1–37 (2018). https://doi.org/10.1007/ s10916-018-0916-7

ECG Signal Classification Based on Neural Network

11

6. Jambukia, S.H., Dabhi, V.K., Prajapati, H.B.: Classification of ECG signals using machine learning techniques: a survey. In: 2015 International Conference on Advances in Computer Engineering and Applications, pp. 714–721. IEEE, India (2015) 7. Hamid, R.A., Albahri, A.S., Albahri, O.S., Zaidan, A.A.: Dempster–Shafer theory for classification and hybridised models of multi-criteria decision analysis for prioritisation: a telemedicine framework for patients with heart diseases. J. Amb. Intell. Human. Comput. Springer, Berlin, Heidelberg (2021) 8. Sathya, R., Akilandeswari, K.: A novel neural network based classification for ECG signals. Int. J. Recent Innov. Trends Comput. Commun. 3(3), 1554–1557 (2015) 9. Dua, D., Graff, C.: UCI machine learning repository. University of California, School of Information and Computer Sciences, Irvine (2017) 10. Li, C., Zheng, C., Tai, C.: Detection of ECG characteristic points using wavelet transforms. IEEE Trans Biomed Eng 42(1), 21–28 (1995) 11. Krikler, D.M.: Heart disease: a textbook of cardiovascular medicine. Br. Heart J. 68(2), 250 (1992) 12. Winter, D.A., Rautaharju, P.M., Wolf, H.K.: Measurement and characteristics of over-all noise content in exercise electrocardiograms. Am. Heart J. 74(3), 324–331 (1967) 13. Tang, Z., Zhao, G., Ouyang, T.: Two-phase deep learning model for short-term wind direction forecasting. Renew. Energy 173, 1005–1016 (2021) 14. Zhang, D.: Wavelet approach for ECG baseline wander correction and noise reduction. In: 27th Annual Conference of Engineering in Medicine and Biology, pp. 1212–1215. IEEE, Shanghai (2006) 15. Bazi, Y., Alajlan, N., AlHichri, H., Malek, S.: Domain adaptation methods for ECG classification. In: International Conference on Computer Medical Applications (ICCMA), pp. 1–4. IEEE, Tunisia (2013) 16. Gao, J., Zhang, H., Lu, P., Wang, Z.: An effective LSTM recurrent network to detect arrhythmia on imbalanced ECG dataset. J. Healthcare Eng. 2019 (2019) 17. Muslim, A.M., Mashohor, S., Mahmud, R., Al Gawwam, G., binti Hanafi, M.: Automated feature extraction for predicting multiple sclerosis patient disability using brain MRI. Int. J. Adv. Comput. Sci. Appl. 13(3) (2022) 18. Al-Sharafi, M.A., Al-Emran, M., Iranmanesh, M., Al-Qaysi, N., Iahad, N.A., Arpaci, I.: Understanding the impact of knowledge management factors on the sustainable use of AIbased chatbots for educational purposes using a hybrid SEM-ANN approach. Interact. Learn. Environ. 1–20 (2022) 19. Xue, Q., Hu, Y.H., Tompkins, W.J.: Neural-network-based adaptive matched filtering for QRS detection. IEEE Trans. Biomed. Eng. 39(4), 317–329 (1992) 20. Zheng, Z., Chen, Z., Hu, F., Zhu, J., Tang, Q., Liang, Y.: An automatic diagnosis of arrhythmias using a combination of CNN and LSTM technology. Electronics 9(1), 121 (2020)

A Review of Long Short-Term Memory Approach for Time Series Analysis and Forecasting Nur Izzati Ab Kader1

, Umi Kalsom Yusof1(B) , Mohd Nor Akmal Khalid2 and Nik Rosmawati Nik Husain3

,

1 School of Computer Sciences, Universiti Sains, 11800 Penang, Malaysia

[email protected], [email protected] 2 School of Information Science, Japan Advance Institute of Science and Technology, 1-1

Asahidai, Nomi, Ishikawa 923-1292, Japan [email protected] 3 Department of Community Medicine, School of Medical Sciences, Universiti Sains Malaysia, Kota Bharu 16150, Kelantan, Malaysia [email protected]

Abstract. The long short-term memory (LSTM) approach has evolved into cutting-edge machine learning techniques. It belongs to the category of deep learning algorithms originating from Deep Recurrent Neural Network (DRNN) forms. In recent years, time series analysis and forecasting utilizing LSTM can be found in various domains, including finance, supply and demand forecasting, and health monitoring. This paper aims to analyze the previous recent studies from 2017 to 2021, emphasizing the LSTM approach to time series analysis and forecasting, highlighting the current enhancement methods in LSTM. It is found that the applications of LSTM in the current research related to time series involve forecasting or both. The finding also demonstrated the current application and advancement of LSTM using different enhancement techniques such as hyperparameter optimization, hybrid and ensemble. However, most researchers opt to hybridize LSTM with other algorithms. Further studying could be applied to improve LSTM performance, especially in the domain study, in which the LSTM enhancement technique has not been widely applied yet. Keywords: Long short-term memory · Time series analysis · Time series forecasting · Deep learning

1 Introduction Over the past years, machine learning (ML) has made numerous remarkable improvements in developing complex learning algorithms and successful methodologies. Amongst the advancements was the growth of artificial neural network (ANN) into progressively deep neural network topologies with advanced learning capabilities, dubbed deep learning (DL). DL models have outperformed other shallow machine learning © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 12–21, 2023. https://doi.org/10.1007/978-3-031-20429-6_2

A Review of Long Short-Term Memory Approach

13

algorithms and traditional techniques for data analysis in a variety of applications [1]. The LSTM algorithm is a deep learning method derived from the DRNN form. In 1997, Hochreiter and Schmidhuber introduced LSTM as a subclass of DRNN [2]. The learning capabilities of LSTM have had a significant impact on a variety of fields, both practically and theoretically, to the point that it has become a state-of-the-art model. LSTM is the sub of recurrent neural network that outperforms other deep learning models in solving time series problems. Due to its long-term memory capability, LSTM is frequently regarded as one of the fundamental deep-learning algorithms for time series [3]. A time series is a collection of numbers accompanied by date and time stamps. Typically, data is collected consistently, including every hour, day, month, or year. Time series are often classified into two types: analysis and forecasting. While these two domains are closely related, they perform differently: time series analysis is concerned with defining the intrinsic structure of time series data and interpreting the hidden characteristics to extract useful information; regression analysis is concerned with extracting useful information from time-series data (such as pattern or seasonal variance). Time series forecasting is a process that entails training machine learning/deep learning models on historical time series data and then consuming them to forecast future predictions [4]. Time series analysis and forecasting utilizing LSTM have been used for various applications in recent years, including finance, supply and demand forecasting, and health monitoring. Additionally, numerous scientific areas and economic sectors rely substantially on the application of time series [5]. While almost every design can be utilized for any application, certain architectures are better suited to particular data types. This review aims to analyze the previous recent studies from 2017 to 2021, emphasizing the LSTM approach to time series analysis and forecasting, highlighting the current enhancement methods in LSTM. It is significant to find the current state-of-the-art application of LSTM for time series analysis and forecasting to help other researchers fill in the current gap and provide a better solution. The remainder of this work is divided into the following sections: Sect. 2 discusses the history of LSTM and time series. Section 3 contains past references to relevant works. Section 4 presents an overview based on studies that have been reviewed. Finally, Sect. 5 concludes.

2 Background Study 2.1 Long Short-Term Memory Approach The LSTM architecture was developed to address a weakness in RNNs: the exploding/vanishing gradient challenge. The error flow is maintained by LSTM using special units called gates, which allow for weight changes and the truncation of the gradient when its output is not needed. The addition of gating units to the LSTM enables it to retain a long-term memory for the trends in primary data [6]. Even with noisy, incompressible input sequences, LSTMs can bridge time intervals of more than 1000-time steps while maintaining short time lag capabilities. The architecture imposes a continual error flow through the internal states of a specific unit referred to as a memory cell. The architecture of the LSTM is depicted in Fig. 1.

14

N. I. A. Kader et al.

Fig. 1. The LSTM consists of forget gate, input gate and output gate.

Basic Components in LSTM. The most well-known architecture is the vanilla LSTM unit, which consists of forget gate, input gate and output gate. The LSTM architecture comprises a collection of recurrently connected sub-networks dubbed memory blocks. The memory block’s purpose is to maintain its state and regulate the flow of information through nonlinear gating units throughout time. LSTMs frequently contain blocks that contain multiple LSTM units (gates) [7, 8]. LSTM block is composed of three or four gates. Logical functions are used to implement the gates, which compute a value between 0 and 1. The gates control the flow of data into and out of the memory of the LSTM blocks. an input gate regulates when new values are permitted to enter the memory; a forget gate regulates how long weights remain in memory, and an output gate governs how memory data are used to compute the block’s output activation [9]. A memory cell is a more advanced form of the LSTM unit. It is built around a central linear unit that is self-contained. The memory cell gets data from gate units referred to as gated recurrent units (GRU), and it can transport the data unmodified through the block. GRU is a more straightforward implementation of the LSTM method. LSTM has a greater number of parameters than GRU. for example, GRU lacks an output gate, which is one of the distinctions between LSTM and GRU. The ability of LSTMs to learn influenced a wide variety of domains, both practically and conceptually, to the point that they became a state-of-the-art model. Enhancement Methods in LSTM. Numerous hyperparameters such as weight, layers, and neurons must be changed to obtain satisfactory results with LSTM. The performance of the LSTM is configuration dependent. It is critical to avoid network under- and overfitting [10]. Hyperparameter optimization is one way for optimizing the performance of the algorithm. Users of any algorithm can set the parameters to any value they choose or apply a tuning approach to determine the optimal parameters for processing the data. Grid search is another way that researchers employ [11]. Grid search is an exhaustive search approach that will try every conceivable combination of hyperparameters inside a given hyperparameters subset. The use of metaheuristic algorithms is another strategy that has grabbed the curiosity of academics. In the context of hyperparameter optimization, this methodology is defined as a strategy for “learning to learn” about the subject [4].

A Review of Long Short-Term Memory Approach

15

These algorithms are stochastic in nature, and they are referred regarded as natureinspired algorithms because of their resemblance to the behavior of animals in the wild. Unlike genetic algorithms, which have their origins in biology, ant colonies and particle swarms have their origins in ethology. Simulated annealing, on the other hand, is a realworld physical process. Using metaheuristics to find optimal or near-optimal solutions to complex problems with several peaks and valleys can be more effective than using classic gradient-based techniques, which can become stuck in local optima [12]. Following that, hybrid approach is another option that is inspired by combining two different algorithms together. When dealing with complicated real-world systems that involve unknown mixed patterns, a single model is insufficient to handle the situation. Combining several models is one of the most frequently stated techniques in the literature. When there is a lack of a complete individual model that can capture numerous patterns in data at the same time [13], hybridization is usually used to compensate. Apart from the hybrid, another technique is called an ensemble. However, ensemble models and hybrid techniques employ the concept of information fusion in more or less distinct ways. Multiple homogeneous weak models are frequently merged at the output level in ensemble classifiers, adopting a variety of merging algorithms classified as fixed and learning combiners. On the other hand, both have the potential to dramatically improve the quality of reasoning and the flexibility of real-world solutions. 2.2 Time Series In time series, it is assumed that the value of a data point X at a given time is connected to past values. Due to the fact that the time series is calculated in discrete time units, the values can be stated using Eq. 1 [14]. X = x(1), x(2), . . . , x(t)

(1)

x t is the most recently computed value. The majority of time series issues attempt to anticipate x(t + 1) by utilizing prior function values that are closely related to the expected value. Time series are often classified into two types: analysis and forecasting. Time Series Analysis. Time-series analysis is used to determine the effect of time varying exogenous variables on a result across time [15]. After the model has been fitted and modified for time-varying extraneous variables, it is critical to examine residual dependency. Time series analysis is frequently used by data scientists for the following reasons: (i) Develop a thorough understanding of the structural underpinnings of historical time series data. (ii) Increase the consistency of time series function interpretation in order to more effectively educate the problem domain. (iii) To enable a more lucrative and comprehensive historical data collecting, preprocessing, and feature engineering of high quality. Time Series Forecasting. For time series forecasting, the trends will be coupled with the observation value. It is classified as either univariate or multivariate. A univariate time series is made up of a single variable measured against time, whereas a multivariate time series has numerous variables at each timestamp [16]. This massive collection of time series data is evaluated for forecasting applications, including weather forecasting,

16

N. I. A. Kader et al.

stock market forecasting, and sales forecasting. Forecasting challenges are classified into single-step and multi-step forecasting based on the number of possible prediction steps. Single-step and multi-step problems can be separately solved using single-step and multi-step forecasting.

3 Related Works Based on the study conducted, LSTM were found to be used in a variety of domains such as solar irradiation, financial stock market, engineering, environmental science, medical and transportation. Table 1 provides studies in time series analysis and forecasting using LSTM. Note that it does not cover all the available studies. Instead, it gives an overview of current techniques used in time series analysis and forecasting with their application. For the application of time series, it was found that the previous researchers focused on either analysis, forecasting or both. Studies that focus on time series analysis can be found in [37, 38]. Zeng et al. [37] apply the time-series analysis approach to enhance gesture categorization and to properly quantify the resemblance of the training and test data’s micro-Doppler (MD) signatures. The maximal instantaneous Doppler frequencies exhibited in the spectrograms describe the MD signatures. Mendonçaet al. [38] advocated using LSTM to search for a cyclic alternating pattern indicative of sleep instability. Additionally, various studies have been conducted on the application of time series to forecasting, such as [17, 18, 22, 27, 29, 33, 40–42]. Previously, [17] used LSTM to analyze data on solar irradiation. It is necessary to know the specific amount of electricity available from various sources and at various time intervals: minutes, hours, and days, while operating a solar-powered electricity-generating system. As a result, they evaluated the LSTM and Gated Recurrent Unit (GRU) algorithms to determine whether they are suitable for and competitive on solar irradiation data. Besides, [22] presented a hybrid approach that combined LSTM and GA. This paper analyses the temporal property of stock market data using daily Korea Stock Price Index (KOSPI) data and proposes a systematic technique for choosing the time window size and topology for the LSTM network using GA. Another study [31] used GA to optimize the LSTM model as well. Sorkun et al. [17] and Chniti et al. [18] used LSTM for forecasting financial markets with complicated characteristics such as non-stationarity, non-linearity, and sequence correlation. LSTM exceeds established benchmarks such as DNN, random forest, and logistic regression. In [25], the proposed LSTM ensemble forecasting algorithm successfully combines the outcomes of several forecasts (predictions) generated by a group of independent LSTM networks. Following that, [27] compared LSTM and SVM predictions for financial stock volatility and found that LSTM produces more accurate forecasts for even large time intervals. This conclusion that LSTM outperforms conventional techniques is further supported by research from [28] on petroleum prediction. They used numerous layers of LSTM to dramatically improve the model’s capacity to perform temporal tasks and grasp the structure of the data series. On the other hand, the [29] study establishes a framework for time series forecasting using recurrent neural networks that encompasses feature engineering, feature importance, point and interval predictions, and forecast assessment. An empirical study made use of both LSTM and GRU networks.

A Review of Long Short-Term Memory Approach

17

Table 1. Related studies in LSTM for time series analysis and forecasting. Study

Year

Application of time series

Enhancement technique

Domain study

[17]

2017

Forecasting

None

Solar irradiation

[18]

2017

Forecasting

None

E-commerce

[19]

2017

Both

Hybrid

Multidisciplinary

[20]

2017

Forecasting

None

Multidisciplinary

[21]

2017

Forecasting

None

Stock market

[22]

2018

Forecasting

Hybrid

Stock market

[23]

2018

Forecasting

None

Financial market

[24]

2018

Forecasting

None

Financial

[25]

2018

Forecasting

Ensemble

Engineering

[26]

2018

Forecasting

None

Stock market

[27]

2019

Forecasting

None

Multidisciplinary

[28]

2019

Forecasting

Hyperparameter optimization (GA)

Petroleum production

[29]

2019

Forecasting

None

Multidisciplinary

[30]

2019

Forecasting

Hybrid

Multidisciplinary

[31]

2019

Forecasting

Hyperparameter optimization (GA)

Multidisciplinary

[32]

2020

Both

None

Environmental science

[33]

2020

Forecasting

None

Water quality IoT systems

[34]

2020

Forecasting

None

Medical (Covid19)

[35]

2020

Forecasting

None

Medical (Covid19)

[36]

2020

Both

Hybrid

Atmosphere

[37]

2020

Analysis

None

Remote sensing

[38]

2021

Analysis

None

Medical (sleep instability)

[39]

2021

Forecasting

Hybrid

Train delays

[40]

2021

Forecasting

Hybrid

Informatic

[41]

2021

Forecasting

Hybrid

Electrical load

[42]

2021

Forecasting

Hybrid

Electrical load

Additional research on time series analysis and forecasting is available in [19, 32, 36]. For instance, [32] suggested a hybrid neural network called TreNet that blends convolutional neural networks (CNNs) and long short-term memory (LSTM). In this

18

N. I. A. Kader et al.

combination, LSTM was utilized to capture the long-range dependency in historical trend data. Mbatha and Bencherif [36] with their research on biophotovoltaic (BPV) devices, they looked for trends, seasonal components, and irregular components. LSTM was utilized to forecast the one-step-ahead current density using lagged values of the output and light status (on/off). As can be seen from the examples above, LSTM is an excellent model for time series analysis and forecasting. Numerous other uses exist for forecasting scenarios, including wind speed forecasting, building energy consumption forecasting, carbon emission levels forecasting, air quality monitoring, and aircraft delays forecasting. The power of the LSTM denotes the model’s extensive applicability in areas where other recurrent neural networks are likely to fail.

4 Discussion This section covers the analysis of related works in Sect. 3. This study aims to analyze recent research on LSTM application to time series analysis and forecasting. It is different from the current review on LSTM as this study specifies the scoping review to only LSTM related to time series analysis and forecasting, emphasizing the current enhancement methods in LSTM. From the studies reported, it is found that applications of LSTM that are related to time series are more prone to forecasting or both. A minimal study that focuses only on analysis from the observation, as the study focusing only on time series analysis usually aims to understand some trends from the data and find helpful information. It is because, based on the literature, a method such as Regression models applied more often for time series analysis, ANNs are then used, followed by Box and Jenkins’ ARIMA with an explanatory variable (ARIMAX), and the researchers frequently use support vector machines (SVMs) compared to LSTM. Time series forecasting, on the other hand, is restricted to the examination of time series data only. Time series forecasting predicts future values based on data that has already been observed in the present. Predicting future values would allow one to be more prepared for the challenges that lie ahead. For example, in the stock market, it is important to understand the current trend using time series analysis, but it is also beneficial to predict values in order to maximize profits. As an additional point of reference, a frequent practice for the enhancement technique is to hybridize LSTM with other algorithms, such as the hybridization of CNNs with LSTM undertaken by [19]. As a result of the findings, the majority of the studies observed an improvement in LSTM efficiency. In terms of application to domain studies, LSTM has demonstrated a wide range of applications, including studies in Covid19, which is a current global issue. The advantage of LSTM makes it preferable in many domains. For instance, LSTM is suitable for forecasting, notably when dealing with time series, because it is capable of capturing non-linearity. Besides, LSTM supports multiple inputs and is excellent at pattern extraction in input data which sequences are relatively long. It operates by “unfolding” the network over time and guaranteeing that the weights of the links remain constant. By including LSTMs into the DRNN, it is as though the network gains a memory device capable of remembering context from the very beginning of the input. It can help accurately model large and long historical data with high accuracy. These highlighted characteristics make LSTM differ from other time-series approaches. However,

A Review of Long Short-Term Memory Approach

19

based on the study conducted in different domain studies, many research studies still do not involve enhancement techniques for LSTM as the application of deep learning, such as LSTM, which is an extension of machine learning, is still new in some domain studies. Thus, there is still room for improvement for researchers to explore and improve the performance of LSTM.

5 Conclusion This study aims to analyze recent research on the application of LSTM to time series analysis and forecasting, emphasizing the current enhancement methods in LSTM that have been achieved. The finding has demonstrated the current application and advancement of LSTM using different enhancement techniques such as hyperparameter optimization, hybrid and ensemble. This review also illustrated the ability of LSTM for time series analysis and forecasting in different domain studies such as financial, engineering, medical and multidisciplinary studies. In future, further customization of LSTM architectures could be applied to improve LSTM performance, especially in the domain study, in which the LSTM enhancement technique has not been widely applied yet. Acknowledgments. The authors would like to convey their appreciation to the Ministry of Higher Education Malaysia for the support provided in completing the current investigation. The funding for this research comes from the Fundamental Research Grant Scheme under the Project Code: FRGS/1/2019/ICT02/USM/02/3.

References 1. Janiesch, C., Zschech, P., Heinrich, K.: Machine learning and deep learning. Electron. Mark. 31(3), 685–695 (2021). https://doi.org/10.1007/s12525-021-00475-2 2. Hochreiter, S., Schmidhuber, J.: Long short-term memory. Neural Comput. 9(8), 1735–1780 (1997) 3. Thi Kieu Tran, T., Lee, T., Shin, J.Y., Kim, J.S., Kamruzzaman, M.: Deep learning-based maximum temperature forecasting assisted with meta-learning for hyperparameter optimization. Atmosphere 11(5) (2020) 4. Lazzeri, F.: Machine Learning for Time Series Forecasting with Python. Wiley (2020) 5. Leijnen, S., Veen, F.V.: The neural network zoo. In: Multidisciplinary Digital Publishing Institute Proceedings, vol. 47, no. 1 (2020) 6. Yu, X., Li, D.: Important trading point prediction using a hybrid convolutional recurrent neural network. Appl. Sci. 11(9) (2021) 7. Zazo, R., Lozano-Diez, A., Gonzalez-Dominguez, J., T. Toledano, D., Gonzalez-Rodriguez, J.: Language identification in short utterances using long short-term memory (LSTM) recurrent neural networks. PLOS One 11(1) (2016) 8. Zhang, R., Liu, Y., Sun, H.: Physics-informed multi-lstm networks for metamodeling of nonlinear structures. Comput. Methods Appl. Mech. Eng. 369 (2020) 9. Nguyen, M., Sun, N., Alexander, D.C., Feng, J., Yeo, B.T.: Modeling Alzheimer’s disease progression using deep recurrent neural networks. In: 2018 International Workshop on Pattern Recognition in Neuroimaging (PRNI), pp. 1–4. IEEE (2018)

20

N. I. A. Kader et al.

10. Tran, T.T.K., Bateni, S.M., Ki, S.J., Vosoughifar, H.: A review of neural networks for air temperature forecasting. Water 13(9) (2021) 11. Rahman, M., Islam, D., Mukti, R.J., Saha, I.: A deep learning approach based on convolutional lstm for detecting diabetes. Comput. Biol. Chem. 88 (2020) 12. Bouktif, S., Fiaz, A., Ouni, A., Serhani, M.A.: Multi-sequence LSTM-RNN deep learning and metaheuristics for electric load forecasting. Energies 13(2) (2020) 13. Hajirahimi, Z., Khashei, M.: Hybrid structures in time series modelling and forecasting: a review. Eng. Appl. Artif. Intell. 86, 83–106 (2019) 14. Bhanja, S., Das, A.: Impact of data normalization on deep neural network for time series forecasting (2018). arXiv preprint arXiv:1812.05519 15. Zeger, S.L., Irizarry, R., Peng, R.D.: On-time series analysis of public health and biomedical data. Annu. Rev. Pub. Health 27, 57–79 (2006) 16. Nielsen, A.: Practical Time Series Analysis: Prediction with Statistics and Machine Learning. O’Reilly Media, Inc. (2019) 17. Sorkun, M.C., Paoli, C., Incel, Ö.D.: Time series forecasting on solar irradiation using deep learning. In: 2017 10th International Conference on Electrical and Electronics Engineering (ELECO), pp. 151–155. IEEE (2017) 18. Chniti, G., Bakir, H., Zaher, H.: E-commerce time series forecasting using lstm neural network and support vector regression. In: Proceedings of the International Conference on Big Data and Internet of Thing, pp. 80–84 (2017) 19. Lin, T., Guo, T., Aberer, K.: Hybrid neural networks for learning the trend in time series. Hybrid neural networks for learning the trend in time series. In: Proceedings of the TwentySixth International Joint Conference on Artificial Intelligence, pp. 2273–2279 (2017) 20. Yunpeng, L., Di, H., Junpeng, B., Yong, Q.: Multi-step ahead time series forecasting for different data patterns based on LSTM recurrent neural network. In: 2017 14th Web Information Systems and Applications Conference (WISA), pp. 305–310. IEEE (2017) 21. Zhuge, Q., Xu, L., Zhang, G.: LSTM neural network with emotional analysis for prediction of stock price. Eng. Lett. 25(2) (2017) 22. Chung, H., Shin, K.S.: Genetic algorithm-optimized long short-term memory network for stock market prediction. Sustainability 10(10) (2018) 23. Fischer, T., Krauss, C.: Deep learning with long short- term memory networks for financial market predictions. Eur. J. Oper. Res. 270(2), 654–669 (2018) 24. Yan, H., Ouyang, H.: Financial time series prediction based on deep learning. Wirel. Pers. Commun. 102(2), 683–700 (2018) 25. Choi, J.Y., Lee, B.: Combining LSTM network ensemble via adaptive weighting for improved time series forecasting. In: Mathematical Problems in Engineering, vol. 2018 (2018) 26. Wu, C.H., Lu, C.C., Ma, Y.F., Lu, R.S.: A new forecasting framework for bitcoin price with LSTM. In: 2018 IEEE International Conference on Data Mining Workshops (ICDMW), pp. 168–175. IEEE (2018) 27. Liu, Y.: Novel volatility forecasting using deep learning–long short-term memory recurrent neural networks. Expert Syst. Appl. 132, 99–109 (2019) 28. Sagheer, A., Kotb, M.: Time series forecasting of petroleum production using deep lstm recurrent networks. Neurocomputing 323, 203–213 (2019) 29. Petneházi, G.: Recurrent neural networks for time series forecasting (2019). arXiv preprint arXiv:1901.00069 30. Fathi, O.: Time series forecasting using a hybrid arima and lstm model. Velvet Consult. (2019) 31. Gorgolis, N., Hatzilygeroudis, I., Istenes, Z., Gyenne, L.G.: Hyperparameter optimization of lstm network models through genetic algorithm. In: 2019 10th International Conference on Information, Intelligence, Systems and Applications (IISA), pp. 1–4. IEEE (2019) 32. Okedi, T.I., Fisher, A.C.: Time series analysis and long short-term memory (LSTM) network prediction of BPV current density. Energy Environ. Sci. 14(4), 2408–2418 (2021)

A Review of Long Short-Term Memory Approach

21

33. Thai-Nghe, N., Thanh-Hai, N., Chi Ngon, N.: Deep learning approach for forecasting water quality in iot systems. Int. J. Adv. Comput. Sci. Appl. 11(8), 686–693 (2020) 34. Chimmula, V.K.R., Zhang, L.: Time series forecasting of covid-19 transmission in Canada using LSTM networks. Chaos Solitons Fractals 135 (2020) 35. Zeroual, A., Harrou, F., Dairi, A., Sun, Y.: Deep learning methods for forecasting covid-19 time-series data: a comparative study. Chaos Solitons Fractals 140 (2020) 36. Mbatha, N., Bencherif, H.: Time series analysis and forecasting using a novel hybrid lstm data-driven model based on empirical wavelet transform applied to total column of ozone at buenos aires, argentina (1966–2017). Atmosphere 11(5) (2020) 37. Zeng, Z., Amin, M. G., Shan, T.: Arm motion classification using time-series analysis of the spectrogram frequency envelopes. Remote Sens. 12(3) (2020) 38. Mendonça, F., Mostafa, S.S., Morgado-Dias, F., Ravelo-García, A.G.: On the use of patterns obtained from LSTM and feature-based methods for time series analysis: application in automatic classification of the cap a phase subtypes. J. Neural Eng. 18(3) (2021) 39. Teo, T.W., Choy, B.H.: in. In: Tan, O.S., Low, E.L., Tay, E.G., Yan, Y.K. (eds.) Singapore Math and Science Education Innovation. ETLPPSIP, vol. 1, pp. 43–59. Springer, Singapore (2021). https://doi.org/10.1007/978-981-16-1357-9_3 40. Fang, Z., Dowe, D. L., Peiris, S., Rosadi, D.: Minimum message length in hybrid arma and LSTM model forecasting. Entropy 23(12) (2021) 41. Dudek, G., Pełka, P., Smyl, S.: A hybrid residual dilated LSTM and exponential smoothing model for midterm electric load forecasting. IEEE Trans. Neural Netw. Learn. Syst. (2021) 42. Rafi, S.H., Deeba, S.R., Hossain, E.: A short-term load forecasting method using integrated CNN and LSTM network. IEEE Access 9, 32436–32448 (2021)

Applying Modified LBP for 2D ECG Images Classification Anfal Hamid Hammad1 and Azmi Shawkat Abdulbaqi2(B) 1 College of Computing and Information Technology, University of Anbar, Ramadi, Iraq

[email protected] 2 College of Computer Science and Information Technology, University of Anbar, Ramadi, Iraq

[email protected]

Abstract. A novel texture classification feature extraction (FeExt) approach that is rotation and histogram equalization (HistEq) resistant was suggested in this study. For this reason, the concept of Modified Local Binary Patterns (MLBP) was used to reflect the key local structural features of distinct textures. We also integrate MLBP with the Aura Matrix (AruMtrx) scale as a second layer for texture analysis of 2D ECG images, taking advantage of MLBP patterns’ global spatial dispersion. Three significant contributions are made by the proposed technique. (a) The suggested MLBP method captures the most important local structure properties of the texture of 2D ECG images (corners and edges); (b) the suggested technique extracts global information using AruMtrx scale based on prominent patterns’ spatial distribution (DomPats) that MLBP creates; and (c) the proposed method rotates, and the HistEq is strong. MIT-BIH and PTB Diagnostic ECG texture datasets were used to evaluate the proposed approach, which included the classification test on randomly rotated images and the HistEq. The proposed method’s classification accuracy is superior to that of other image attributes, according to the results of experiments. Keywords: Electrocardiogram (ECG) · Cardiovascular Disease (CVD) · Modify Local Binary Pattern (MLBP) · Histogram Equalization (HistEq)

1 Introduction Cardiovascular Disease (CVD) is one of the most common causes of sudden mortality across the globe. A bad lifestyle, which includes an unhealthy food, strain and stress, cigarette smoking, and insufficient exercise, causes CVD. Concurrent atrial and ventricular arrhythmias are a negative side effect of cardiovascular disease. Arrhythmia is a condition in which the heart rate fluctuates excessively due to a defective heartbeat, causing blood pumping failure. One of the most effective feature extractors is the two-dimensional local binary pattern, which extracts texture properties from a 2-D image by comparing each signal sample (image pixel) with its neighbor samples in a small vicinity. Because no training is required, feature extraction may be done rapidly and simply with fresh data sets. Furthermore, as shown in Fig. 2, the image-size dimension of the feature space may be © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 22–31, 2023. https://doi.org/10.1007/978-3-031-20429-6_3

Applying Modified LBP for 2D ECG Images Classification

23

reduced to the number of histogram bins is the radius of the neighborhood, and is the number of neighbor samples relative to the center pixel. The neighbor sample is assigned a 1 if its value is more than or equal to that of the center sample, and a 0 if its value is less than that of the center pixel. Texture analysis plays an important part in a variety of computer vision applications. Apps, image retrieval, facial image analysis, and motion analysis are among examples. In texture classification, extracting constant features from the HistEq [1, 2] is a difficult operation. This problem is particularly important since the application of rotationsensitive FeExt and HistEq techniques is severely limited [3]. The electrocardiogram (ECG) is a useful tool for monitoring the electrical activity of the heart. The status of the heart is frequently monitored and examined using an ECG signal. ECG signals may also be used to define arrhythmia, which is important for early detection of heart failure. These symptoms last for a long time, necessitating the services of a skilled, patient, and knowledgeable physician.

2 Literature of Related Works Several scientists experimented with various ways to extract the characteristics of the rotation invariant texture a few decades ago. The authors “Soodeh Nikan, Femida Gwadry-Sridhar,” and “Michael Bauer” [4] suggested a number of segmentation algorithms for adaptive beats based on the median value of the intervals “R-R,” which reduces misclassification since each section comprises surrounding beats. “V. Mondéjar-Guerra, J. Novo, J. Rouco, M.G. Penedo, and M. Ortega” [5] proposed an automatic ECG classification method based on the use of various SVNs (Support Vector Network). The method to ECG descripting is determined by the amount of time between consecutive beats and the shape of the beats. Other writers, including “Meryem Regouid, Mohamed Touahria, Mohamed Benouis,” and “Nicholas Costen” [6], proposed a multi-modal biometric system based on ECG-ear-iris biometrics at the feature level. “Mohammed Benouse” and “Meryem Regouid” [8] presented and implemented a novel ear-based biometric identification system based on shifted 1D-LBP and analyzed to recognize the various weaknesses/benefits of 1D-LBP, Shifted-1DLBP, and 1D-MR-“Mohammed Benouse” and “Meryem Regouid” [8], presented and implemented a novel ear-based bio. 1D-LBP was used to decrease the size of each ear image vector after it was preprocessed and transformed to 1D space. The authors “Daban Abdulsalam Abdullah, Muhammed H. Akpnar, and Abdulkadir Engür” [7] proposed an efficient approach for ECG beat classification. The proposed approach differs from signal processing methods in which ECG beats were transformed to ECG pulse images and subsequently extracted characteristics using image processing techniques. Well-known SVN techniques were used using linear, quadratic, cubic, and Gaussian kernel functions. The results of experiments including 5,000 and 10,000 ECG beat samples, separated into five groups, were presented. “G. Suseendran, Noor Zaman, M. Thyagaraj,” and “M. Thyagaraj” [9] each offered an approach for predicting CVD in which an initial heart local binary pattern was produced and PCA was used in combination with an ANN. Where LBP and PCA are used to extract the feature and reduce the size of the feature set, while ANN is used to classify and verify the system.

24

A. H. Hammad and A. S. Abdulbaqi

3 Methodology In this suggested technique, a 2D ECG image is transformed to a grayscale image, the local binary pattern (LBP) is generated, and then the shape parameters are assessed for health classification, as shown in Fig. 1. Electrocardiogram description; preprocessing; feature extraction; classification; and comparison were the modules of the suggested system.

Fig. 1. The system methodology

3.1 Preprocessing Data preprocessing refers to any sort of processing performed on raw data in order to prepare it for further processing. Preprocessing data is a data mining method that converts data into a format that can be processed more readily and efficiently for a user’s aim, such as in a neural network [10, 11]. During preprocessing, the heart rate dataset will be submitted to the program. The software analyzes all of the heartbeat sequences in the collection. The first step is to download the heartbeat dataset and convert it to text for use as program input [13]. The ECG (usually the top signal) was digitally filtered for band-pass to enhance QRS complexes after phase shift correction in the filter, and each rhythm label was transferred to a substantial local tip. A few number of loud chimes are manufactured by hand [12, 14].

Applying Modified LBP for 2D ECG Images Classification

25

3.2 Extracting the Features Diagnosis of cardiovascular disease. PQRST waves may be seen in a single cardiac cycle of an ECG signal. This feature extraction approach determines the amplitude and time intervals within ECG data for further analysis. Each human’s heart functioning is defined by the amplitudes and intervals of the P-QRS-T segment. Feature extraction helps to reduce the amount of data in a dataset that isn’t needed. Finally, minimizing the data makes it simpler for the machine to create the model with less effort, as well as speeding up the machine learning learning and generalization processes. 3.3 Modified LBP (MLBP) The pixels of the 2D ECG image are initially discriminated in the standard LBP approach [9] by finding the various among the pixel of the center and its surrounds using the function of step u (x), i.e. u(x) = One when x > Zero else u(x) = Zero

(1)

As a unique description, the labels’ sequence of adjacent to each pattern is used. Patterns are uniform if the transitions between “Zero” and “One” are smaller than or equal to “Two” (UniPat). UniPats, for example, are “01100000” and “11011111.” The feature vector in the complete 2D ECG image is UniPat of HistEq [9]. To do multi-resolution analysis, several values for m and R may be used, where m refers to the pixel number in relation to the pixel center and R refers to the distance between the pixel and every pixel of neighboring pixels. Traditional LBP, on the other hand, cannot capture the majority of the pattern information in the texture of 2D ECG images by simply searching in “Unified LBPs” for certain textures whose DomPats aren’t “uniform LBPs,” notably for textures with uneven shapes and edges. In the PTB Diagnostic ECG datasets, Table 1 displays the ratios of “uniform LBPs” with varied “radius R” in the texture of 2D ECG image. There are many stages to implement MLBP: 1. The 2D ECG image is converted to grayscale image space. 2. Choose the regions (P) that surround the center pixel for each pixel (gp) in the image the coordinates of gp are defined by:      2π p 2π p , gcy + Rcos , (2) gcx − Rsin P P 3. Set the P next to the center pixel (gc) as a threshold. 4. If the value of the neighboring pixel is more than or equal to the value of the center pixel, One is set, and Zero is set otherwise. 5. Calculate the value of LBP by making a binary number using numbers next to the center pixel in a counterclockwise sequence. The LBP-central pixel code is a binary integer (or its decimal counterpart) used as a specialized local texture feature.

26

A. H. Hammad and A. S. Abdulbaqi

LBP(gpx, gpy)

p−1 

S(gp − gc)x2p

(3)

p=0

Table 1. Shows the percentage of samples in the PTB diagnostic ECG Dataset. Texture

P = 7, R = 2

P = 15, R = 1

P = 25, R = 2

Rec.Bk #0001

52.73

41.34

31.73

Rec.Bk #0002

57.11

45.73

38.10

Rec.Bk #0003

53.18

39.95

24.82

Rec.Bk #0004

47.69

32.42

20.76

Rec.Bk #0005

54.83

43.04

34.18

Rec.Bk #0006

25.91

33.04

24.17

Rec.Bk #0007

50.05

46.18

33.06

Rec.Bk #0008

57.85

48.82

36.11

For the goal of fully explaining the key patterns seen in the texture of 2D ECG images, the standard LBP to MLBP was enlarged.

4 MIT-BIH and PTB Diagnostic ECG Datasets The MIT-BIH Arrhythmia Data Collection has been used at over 500 locations throughout the globe since 1980 to evaluate arrhythmia detectors and conduct fundamental heart dynamics research. It may be credited with much of the current acknowledgment of the value of shared datasets, both for basic research and to create and evaluate medical devices, thanks to its cooperation with the AHA (American Heart Association) database. Clinical evaluations from 500 patients make up the full PTB Diagnostic ECG. There were no prerequisites or data selections made in order to impose clinical realism. Consequently. PTB Diagnostic ECG is now accessible online, and it will be updated and maintained as long as the data is helpful for clinical study [11].

5 Information on Spatial Distribution (SDI) of Dominate Patterns Despite the fact that modified local binary patterns may encode dominant pattern information more consistently, efficiently, and robustly for random rotation, SDI for DomPats is currently absent. More precisely, with the MLBP alone, we can only discern what the DomPats are in a texture of a 2D ECG image. We have no clue where these DomPats are, however. The following diagram depicts a Normal two-dimensional ECG signal, which is used to estimate the dimensions of the other signals, as well as whether they are Normal or Abnormal, depending on the wave peaks’ amplitude.

Applying Modified LBP for 2D ECG Images Classification

27

Fig. 2. a, b Various types of 2D ECG texture

Such information is important for characterizing the texture of a 2D ECG image. To clarify the concepts of SDI, the distribution of a prominent pattern in several textures of 2D ECG images gathered from the PTB Diagnostic ECG [14] database is presented in Fig. 1. According to our results, the four textures of 2D ECG images in Fig. 2 were incorrectly identified using only the standard LBP or MLBP classification methods. This is because, although MLBP may be able to characterize the sorts of DomPats in a texture more accurately, the SDI of these patterns is lost. The identical set of DomPats appears in all four textures, with highly similar proportions, one of which is the pattern type indicated by 24. As can be seen, the number of such DomPats in each texture is the same. However, the pattern’s distribution properties in each of the four textures of 2D ECG images are vastly different. As a consequence, as the Experimental results section will demonstrate, the SDI is a critical attribute for detecting textures. For extracting features from distribution dominating pattern maps, the gray level AruMtrx (GL AruMtrx) approach [11] was used. Aura [11]: For region system N, labeled ϑ B(A, N), the aura of A in relation to B is provided by: For the region system N, given two subsets A, B, C, S, the aura of A in regard to B, denoted B(A, N), is given by: ϑB (A, B) = UsA (B ∩ Ns )

(4)

Aura metrics: [11]: Aura metric of A in relation to B, represented by m(A, B), is calculated utilizing the same notations as in Eq. (8):  m(A, B) = |N8 ∩ B| (5) s∈A

where JAI is the overall number of items in A for a particular subset A C S. Alcin et al. [11] GLAruMtrx: Let N denote the image’s gray level sets over S, and Si, 0 I G-1 denote the region system over S. The ECG image’s gray level AruMtrx over N, abbreviated A, is then computed as follows:   A = A(N) = m(Si , Sj ) (6) where G is the total number of the gray levels in the image, Si = {s ∈ S|xs = i} is the set of gray level equivalent to the level of ith, and m(Si, Sj) is the metric of aura among Si and Sj definite in Eq. (5). The finding in the GLAruMtrx 4-elements are significant and strongly features, which can define the SDI of the DomPats and are utilized as features

28

A. H. Hammad and A. S. Abdulbaqi

to complements with the MLBP. The following Fig. 3 refers to the unfiltered QRS wave (Raw QRS) based on the peaks measures. Due to it is the most significant element of the cardiac signal, obtaining the measurements of that QRS Wave is critical because it allows the expert to identify whether the ECG signal is normal or abnormal.

Fig. 3. Non filtered QRS wave

6 Experiential Findings On two different datasets with big texture image sets, we tested our recommended technique: (1) 24-textures from the MIT-BIH dataset [13]; (2) 47 images from the PTB Diagnostic ECG dataset [14]. Random rotation was shown to be effective in the experiments (Every ECG image in the testing and training sets was rotated by a random angle ranging from 0 to 360°). To assess the method’s resiliency, HistEq was used in both the training and testing sets for each sample. The remainder of each class’s images were used as testing sets, while half of each class’s images were used as training sets. Eight frequently used texture classification algorithms were compared to the provided method. In our experiment, we used SVN as the classifier. The SVN’s kernel was the Gaussian Radial Basis Function (RBF). Experiments using MIT-BIH Datasets include the following: Table 2 compares the classification accuracy of several strategies in a variety of scenarios. According to the testing findings, the suggested MLBP technique can already outperform the other eight methods in a variety of circumstances. Furthermore, using the DomPats SDI with the MLBP enhances classification performance when compared to using the MLBP alone. The PTB Diagnostic ECG database has been used in the following experiments: In the PTB Diagnostic ECG database, it contains the most texture classes [14] (47-classes). One of the most noticeable elements of this database is the number of texture classes it contains. As a consequence of the small distances between classes in the feature space, classifying a large number of textures might be problematic. This dataset may be used to assess how accurate the characteristics of each strategy are at defining the texture of 2D ECG images. The experiment’s findings are shown in Table 3. Table 2 shows the performance of several characteristics in the MIT-BIH Dataset. In

Applying Modified LBP for 2D ECG Images Classification

29

the final two rows, the results of the recommended ways are documented. The highest classification accuracy is marked in bold for each test (column). Table 2. Performance of various features in the dataset of MIT-BIH Classification (%) Feature

Original texture

HistEq texture

Randomly rotated texture

HistEq and rotated texture

Patient Rec.Bk. #0001

98.06

87.73

81.32

62.71

Patient Rec.Bk. #0002

91.67

75.00

91.20

76.83

Patient Rec.Bk. #0003

98.61

91.67

83.51

61.36

Patient Rec.Bk. #0004

90.07

60.28

88.64

58.42

Patient Rec.Bk. #0005

96.70

84.33

50.63

40.75

Patient Rec.Bk. #0006

95.83

86.52

93.75

81.65

Patient Rec.Bk. #0007

93.57

70.00

87.35

58.31

Patient Rec.Bk. #0008

97.22

96.30

92.75

91.50

MLBP

98.61

98.61

96.78

96.76

The results of several 6464 ECG image resolution characteristics in the PTB Diagnostic ECG Database are shown in Table 3. The outcomes of our approaches are shown in the final two rows. Each test’s greatest classification accuracy is shown in bold (column).

7 Conclusion This article introduces a novel MLBP approach based on classic LBP. We also observed that since the DomPats SDI integrates the DomPats position information in the texture, it is a highly robust feature for defining the features of the texture of 2D ECG images. It was tested against eight commonly used approaches using two databases: MIT-BIH and PTB Diagnostic ECG. Experiments have shown that our technique is effective at texture classification and is robust to HistEq and random rotation. Our proposed technique also has the benefit of being computationally simple, since the characteristics may be obtained with only a few computations and comparisons, and no image filtering is necessary.

30

A. H. Hammad and A. S. Abdulbaqi

Table 3. Shows the PTB diagnostic ECG database findings for various 64 × 64 ECG image resolution parameters. Classification (%) Feature

Original texture

HistEq texture

Randomly rotated texture

HistEq and rotated texture

Patient Rec.Bk. #0001

90.43

61.17

80.13

54.46

Patient Rec.Bk. #0002

85.11

39.89

83.75

45.81

Patient Rec.Bk. #0003

76.06

44.15

63.25

33.56

Patient Rec.Bk. #0004

60.64

27.66

61.53

36.80

Patient Rec.Bk. #0005

46.81

45.27

27.04

35.63

Patient Rec.Bk. #0006

77.13

71.26

76.59

68.25

Patient Rec.Bk. #0007

69.15

38.83

63.68

38.21

Patient Rec.Bk. #0008

72.87

69.15

70.25

62.58

MLBP

82.98

79.79

84.18

76.02

References 1. Mohammed, Y.R., Basil, N., Bayat, O., Mohammed, A.H.: A new novel optimization techniques implemented on the AVR control system using MATLAB-SIMULINK (2020) 2. Karthikeyan, P., Mithra, K. P., Sriaartee, P., Swetha, K.: Effective content based image retrieval using modified image retrieval algorithm. Extraction 35, 46 3. Abdulwahid, M.M., Wasel, N.B.M.: Design and implementation of water level tank model by using scada system. Inform.: J. Appl. Mach. Electr. Electron. Comput. Sci. Commun. Syst. 1(1), 63–69 (2020) 4. Mondéjar-Guerra, V., Novo, J., Rouco, J., Penedo, M.G., Ortega, M.: Heartbeat classification fusing temporal and morphological information of ECGs via ensemble of classifiers. Biomed. Signal Process. Control 47, 41–48 (2019) 5. Peimankar, A., Jajroodi, M.J., Puthusserypady, S.: Automatic detection of cardiac arrhythmias using ensemble learning. In: TENCON 2019–2019 IEEE Region 10 Conference (TENCON), pp. 383–388. IEEE (2019, October) 6. Abdullah, D.A., Akpınar, M.H., Sengür, ¸ A.: Local feature descriptors based ECG beat classification. Health Inform. Sci. Syst. 8(1), 1 (2020). https://doi.org/10.1007/s13755-020-001 10-y 7. Regouid, M., Benouis, M.: Shifted 1D-LBP based ECG recognition system. In: International Symposium on Modelling and Implementation of Complex Systems, pp. 168–179. Springer, Cham (2018)

Applying Modified LBP for 2D ECG Images Classification

31

8. Dornan, L., Pinyopornpanish, K., Jiraporncharoen, W., Hashmi, A., Dejkriengkraikul, N., Angkurawaranon, C.: Utilisation of electronic health records for public health in Asia: a review of success factors and potential challenges. BioMed Res. Int. 2019 (2019). 9. Yazid, M., Rahman, M.A.: Variable step dynamic threshold local binary pattern for classification of atrial fibrillation. Artif. Intell. Med. 108, 101932 (2020) 10. Muhammad, G., Rahman, S.M.M., Alelaiwi, A., Alamri, A.: Smart health solution integrating IoT and cloud: a case study of voice pathology monitoring. IEEE Commun. Mag. 55(1), 69–73 (2017) 11. Alcin, O.F., Siuly, S., Bajaj, V., Guo, Y., Sengur, A., Zhang, Y.: Multi-category EEG signal classification developing time-frequency texture features based Fisher vector encoding method. Neurocomputing 218, 251–258 (2016) 12. Li, Y., Cui, W., Luo, M., Li, K., Wang, L.: Epileptic seizure detection based on time-frequency images of EEG signals using Gaussian mixture model and gray level co-occurrence matrix features. Int. J. Neur. Syst. 28(07), 1850003 (2018) 13. Yuan, Q., Zhou, W., Xu, F., Leng, Y., Wei, D.: Epileptic EEG identification via LBP opera-tors on wavelet coefficients. Int. J. Neur. Syst. 28(8), 1850010 (2018)

Integrating Ontology with Imaging and Artificial Vision for a High-Level Semantic: A Review Malak Belkebir1(B)

, Toufik Messaoud Maarouk2

, and Brahim Nini1

1 Research Laboratory on Computer Science’s Complex Systems, Larbi Ben M’hidi University,

Oum El Bouaghi, Algeria [email protected] 2 ICOSI Laboratory, Khenchela University, Khenchela, Algeria

Abstract. Nowadays, imaging and artificial vision are becoming important fields of research. Nevertheless, despite the high accuracy that has been achieved by their techniques, most of them remain “quantitative approaches”; their efficiency depends on the computational capacity and requires a large amount of data for training and testing. In addition, there are still many deficiencies such as the semantic gap between the low-level visual information and high-level semantic knowledge, the recognition of complex scenes and so on. With the aim of providing less costly and more efficient solutions, the new trend is to make it possible for the aforementioned fields to support knowledge representation techniques i.e. “qualitative-approaches”, which is at first sight ontological. This article reviews several research-works that have demonstrated the effectiveness of combining those approaches. This combination has certain advantages, among which: closing its semantic gap, reducing the data rate required by making logical inferences, enhancing the querying capability. Keywords: Ontology · Deep Learning · Data-driven approach · Knowledge-driven approach · Semantic gap

1 Introduction Imaging and artificial vision are important fields of research nowadays. Their importance relates to their underneath applications. These last have reached a high level of usefulness thanks to the used techniques. As an example, we can cite the very famous one, called deep learning; which has proven its capabilities in terms of results, either at the theoretical and research level, or at the application level. Despite the high accuracy that has been achieved in the obtained results, most of these techniques are “data-driven approaches”, i.e. quantitative approaches, which use inductive inference; meaning that the results are obtained from existing information in image data such as pixels or pixel morphologies. In this sense, their efficiency depends on the computational capacity of computers and also require a large amount of data for training and testing. The larger it is, the higher the efficiency is. In addition, there are still many deficiencies such as providing fine-grained objects, being detected or separated, © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 32–41, 2023. https://doi.org/10.1007/978-3-031-20429-6_4

Integrating Ontology with Imaging and Artificial Vision

33

the recognition of complex scenes and the identification of complex high-level semantics between low-level visual information and semantic knowledge. With the aim of providing less costly and more efficient solutions, researchers oriented themselves to seeking semantic-oriented approaches using deductive inference. This new trend should make it possible for the field of imagery and that of artificial vision to support knowledge representation techniques -in this case: formal representation-, which is at first sight “ontological” that is based on qualitative approaches. An ontology is a formal, explicit specification of a shared conceptualization [13]. Xu Chuan_yun et al. present the birth of the integration of Ontology in the domain of image semantics recognition [1]. This paper uses OWL (Web Ontology Language) with (DL) Description Logic just to describe image semantics levels. In the field of imaging and artificial vision, any information semantic is divided as follows (see Fig. 1) [2].

Fig. 1. Semantic levels in imaging and artificial vision.

More clearly, these fields need to rely on a formal basis. This mixture will be a contribution in addition to a support for existing imaging techniques. The contribution is in the sense of searching a new vision of existing techniques so that they can be formalized. The support will be in the sense of improving the sophistication of systems/works by making logical inferences. Many researches show that the combination of “quantitative” and “qualitative” approaches has some advantages, among which: extracting the high-level semantic of image/frame, also the scalability of the computer vision systems is improved, the large data rate required for a high accuracy is reduced, the querying capability is enhanced and the finest granularity of objects components is detected. This review paper presents the state-of-the-art of some works that succeeded in this combination, with the aim to help researchers to propose new ideas to contribute to this field. For this, the paper is organized as follows: Sect. 2 presents the review papers [3, 4]. Section 3 provides a summary of several works’ papers in different fields such as human behavior analysis [5–7, 18, 20], image recognition/classification [9, 10], image retrieval [11, 12, 23] and Risk Prediction [14–16]. Finally, we finish with a conclusion in Sect. 4.

2 Review Papers The state-of-the-art approaches combining imaging techniques with ontology are reviewed in several papers [3, 4]. These papers present the recent works that integrate the ontology with Deep Learning techniques in the domain of recognition, which has proved

34

M. Belkebir et al.

their efficiency and importance in terms of recognition accuracy and semantic reasoning. This importance appeared in the following contributions that make systems able to: converging to the human ability to recognize a large number of objects, inferring additional semantics from the existent information in the scene, and enhancing the segmentation results with a semantic-based one [3]. Moreover, this combination has demonstrated its efficiency in treating Deep Learning issues providing a high-level semantic information, and improving the recognition accuracy [4].

3 Works Paper 3.1 Integration of Ontology with Human Behavior Analysis The limitations of data-driven approaches to analyze human behaviors have inspired Mojarad et al. [18] to propose a hybrid framework for human behavior analysis, which combines these approaches with Knowledge-Driven approaches. The authors aimed at characterizing the human context even in the absence of information by inferring new knowledge about his/her behaviors, with the use of the logical reasoning and the exploitation of contextual information. In the first aforementioned approach, the techniques used are CNN-LSTM model for the recognition of the temporal information of Human Activities together with a statistical algorithm to extract their contexts. Whereas in the second stage, the results are mapped to an Ontology called Human AcTivity (HAT) that conceptualizes the human activities and their context. Finally, a high-level expressive logic-based formalism called (‘Answer Set Programming (ASP)’) [19] is then exploited to represent human behaviors and used as an inference-engine for inferring new behaviors.This hybrid approach has proven to be effective in enhancing Ambient Assisted Living (AAL) systems and personal assistance robots to improve people’s quality of life in terms of autonomy, well-being, and safety. With the same previous aims, Mojarad et al. [20] proposed in another paper a hybrid context-aware framework combining a machine-learning model and probabilistic reasoning to detect abnormal human behavior. Steps are shown in Fig. 2 and it is organized as follow: Human activity recognition. An LSTM model is used to classify input data into a set of labels describing human activities (activity, location, and object). Human behavior analysis. The obtained labels are analyzed in terms of locations, used objects, times of the day, duration, frequencies, and activity sequences. Mapping to an ontology. The proposed Human AcTivity (HAT) ontology, inspired by the ConceptNet semantic network [21] is used to conceptualize human activities, human behaviors and their contexts. Abnormal human behavior detection. Human activity predictions is generally uncertain. Mapping uncertainty over ontology will not yield good performance in the context of activity or behavior prediction. So that a Markov Logic Network (MLN) [22], combining logic and probability to handle data uncertainty by assigning weights to FOL

Integrating Ontology with Imaging and Artificial Vision

35

rules (First-Order Logic) defined by an expert, is used to detect abnormal human daily living behaviors. Equation 1 is an example of a FOL rule in MLN. ω Act (Eating, time) ∧ Loc(Bedromm, time) ⇒ AbnormalActLoc (Eating, Bedroom, time)

(1)

Fig. 2. Architecture of the proposed framework [20].

The important novelty of the proposed approach is that it considers the contexts of human behavior in handling the uncertainty of human daily living behaviors for detecting abnormal human behaviors with high accuracy. In Ly N. Q. et al. paper, a method for recognizing the specific behavior “left luggage and loitering at the corridor behaviors” (see Fig. 3) based on “Behavior Ontology” is proposed [5].

Fig. 3. Example of “left luggage and loitering at the corridor behaviors” [5] where the Result is: “Person A left his luggage”.

The authors’ idea is that, by integrating the ontology with prior knowledge and scene processing that use the strategy “Divide and Conquer”, new behaviors are inferred compared to just using deep learning and this is done without the necessity of the data of the entire behavioral process. In addition, the result can be reused in other cases. In the work of J. Vizcarra et al., an Ontology-based methodology for human behavior indexing with multimodal video data is proposed (see Fig. 4) [6]. This approach interprets multimodal human behavior annotations and Deep Neural network outputs in a formal computable format, where each notation (expression with operands and operators) is

36

M. Belkebir et al.

transformed into a SPARQL query, and then converts them into a High-Level semantic knowledge graph (KG). This KG is exploited to analyze complex behavior patterns and to index videos. The novelty of this work in terms of query capability is that the resulting computable human behavior makes it possible to retrieve such a specific scene without reviewing the entire video file, in contrast to other works, which provide the entire video that contains the requested action/scenario. Figure 5 is the result of the query “Grab caregiver’s upper limb AND Grab care receiver’s trunk” in the elderly care domain.

Fig. 4. The proposed methodology steps [6].

Fig. 5. Querying result of “Grab caregiver’s upper limb AND Grab care receiver’s trunk”.

W. F. Youssef et al. proposed an automatic generic approach based on interactions between two objects for video surveillance description for the security and safety in the modern fight against crimes. This approach is considered a high-level generalist layer for its analysis systems [7]. The authors’ idea is that the variety of objects, scenes and behaviors natures/types requires an abstract level of information to reduce this large size of description scope. This article is based on this idea and hence, presents an ontology named "Video-Surveillance-Description Ontology”. It combines “low-level primitives” of objects basic features (size, color, locations, speed and others), which are extracted using Deep Learning, with “ontological information”. The Table 1 is an excerpt of the example of video surveillance generated description: taken from the scene “fight_runaway2”, of the database “caviar” [8], where a person enters the scene (frame 193), then starts fighting with “person 2” (frame 321). 3.2 Integration of Ontology with Image Recognition/Classification C. Mazo et al. aimed at improving the classification results in the domain of medical imaging. They propose a method, which is an ontology-based automatic reclassification of cardiovascular tissues and organs in histological images [9]. This method combines the automatic classification (prior knowledge of experts and result of imaging classification) with the histological ontology, in order to refine the classification of organs and epithelial

Integrating Ontology with Imaging and Artificial Vision

37

Table 1. Example of video surveillance generated description [7]. Frame number 193

321

States Descriptions Object 1: "Deformable" object "1" enters the scene, from "C" spot, on the "Left Middle" of the "Outside" area of the camera field of view, heading "Up Left", having respectively "regular" shape, "small" surface, and "slow" speed. Object 1: "Deformable" object "1" moves, in "F" spot, on the "Left Middle" of the "Inside" area of the camera field of view, heading "Left Middle", "No big changes occurring respectively on" its shape, and "No big changes occurring respectively on" its Surface, and having respectively considerable "decreasing" of its Speed. Object 2: "Deformable" object "2" moves, in "F" spot, on the "Left Middle" of the "Inside" area of the camera field of view, heading "Up Left" , "Toward" the object "1", "No big changes occurring respectively on" its shape, and "No big changes occurring respectively on" its Surface , and having respectively "stable" Speed. Object 1 and Object 2: The two objects are respectively "Merged", A "Physical" "Aggressive" Interaction occurs between them.

tissues. By using SPARQL’s query and RDF triples, a new “organ-label” will be decided according to the behavior of false positives in the classification process, or is confirmed that it is already correctly classified. The proposed method proved its effectiveness in reducing classification errors as shown in (Fig. 6).

Fig. 6. (a) Block diagram of the TP and the FP of the initial and the refined classifications, by organ. (b) TP and FP blocks with automatic classification and improvement process [9].

A. Reiza et al. propose a fashion recognition approach by combining deep-learningbased detection with semantic reasoning (see Fig. 7) [10]. The aim of this article is to augment the semantic of the recognized objects in fashion images and separate the fine granular categories in the latter, by linking semantic reasoning techniques to image recognition techniques. In fact, in contrast to Deep Learning, it is not necessary that all target objects appear in the training phase. It is therefore possible to deduce the fine-grained categories of fashion clothing from those in the image, so that the result is more akin to their recognition by the human eye. The proposed approach differs from other state-of-the-art in that the deduction is based on the identified scene. 3.3 Integration of Ontology with Image Retrieval The famous Content Based Image Retrieval (CBIR) still have difficulties and challenges in term of semantic indexing, such as semantic gap, long-tail problem, large intra/interclass divergence. A. Zga et al. proposed an ontological semantic model to solve the last

38

M. Belkebir et al.

Fig. 7. Sequential steps of the work.

mentioned issues with the aim of extracting semantic visual relationship in sport images [23]. The authors used the ontology to represent the visual information obtained with VGG16 -a Deep Learning model- with lexical formal description. Moreover, a statistical ranking module that aims to filter false positives/negatives of class proposals is proposed. The result is an input of semantic ontological module and visual relationship module models, which aim to achieve a high prediction of objects and their relationships (see Fig. 8).

Fig. 8. The labels “human is playing” and “a ball”, become “football-players are playing with a soccer-ball” with the proposed approach [23].

U. Manzoor et al. propose an Ontology-based Image Retrieval (OIR) system in the mammalian domain [11, 12]. This system retrieves semantically the relevant images upon user’s request. 3.4 Integration of Ontology with Risk Prediction One of the limitations of Deep Learning or Computer Vision is knowledge reasoning; i.e. it cannot link the low-level visual information with high-level semantic interpretation. Wu et al. [14] aimed at closing this semantic gap to facilitate the management of safety of on-site construction by semantically reasoning hazards from images. The authors integrated a formal ontology with computer vision algorithms so that it can converge to the thinking model of safety managers. More precisely, the authors formally represented the safety regulatory knowledge with ontology and SWRL, and used the Deep Learning model ‘Mask R–CNN’ developed

Integrating Ontology with Imaging and Artificial Vision

39

in their previous work [15] to detect visual objects with their spatial relationships. Results of the latter would be transferred into instances in the developed ontology, where hazards can be inferred by comparing extracted visual information from construction images with pre-defined SWRL rules (see Fig. 9).

Fig. 9. Framework of the proposed method [14].

In the same domain, Fang et al. [16] proposed a Knowledge Graph, which is a logical semantic expression for identifying hazards on construction sites. The paper presents a novel semantic computer vision-based approach, combining its algorithms (ResNet) with ontological reasoning, in order to develop this KG.

Fig. 10. The workflow of the proposed hybrid semantic computer vision approach [16].

Fig. 11. Semantic computer vision detection results [16].

This KG can automatically detect and identify construction hazards, based on the distance and spatial information. The KG’ data is modeled using the Neo4j database “objects-attributes-relationship” [17]. The ontology is created using the Graph Database Language.

40

M. Belkebir et al.

The authors address the semantic gap of computer vision when it is unable to recognize rules or objects that are newly defined, or when it is unable to extract semantic relationships between detected objects. Figure 10/Figure 11 present the steps/results.

4 Conclusion Despite the high accuracy that has been achieved by Deep Learning techniques, most of them are “data-driven approaches”, which is costly either on the large amount of data required, or on the necessary computational capacity of computers. Moreover, most of their results are considered as “medium-level” semantic. It cannot link the visual information extracted from images/frames -which is considered “low-level”- with high-level knowledge. Thus, to reduce this semantic gap, the new trend is to use a powerful mechanism for structuring, organizing the knowledge and reasoning. That is ontology. In this paper, we synthesized some works on imaging and artificial vision axes integrating ontology, such as human behavior analysis, image recognition/classification, image retrieval, augmented reality and risk prediction. In addition, we highlighted the importance of combining “data-driven approaches” and “knowledge-driven approaches”, providing a brief analysis of the state-of-the-art. This combination will be a turning point in this field, due to its contributions both in the spatial and temporal domains. It allows providing additional inferred information with high classification accuracy that Deep Learning or computer vision techniques cannot reach alone with few rates of data. Moreover, it is still possible to get better results, especially in the “temporal” field. Acknowledgments. The authors acknowledge the financial support of the Directorate General for Scientific Research and Technological Development (DGRSDT, MESRS) and the Research Laboratory on Computer Science’s Complex Systems (ReLaCS2 ) for this work.

References 1. Xu, C.Y., Yang, D., Zhang, Y.: Image semantics recognition based on ontology and description logics. In: 2009 International Conference on Multimedia Information Networking and Security, vol. 2, pp. 33–36, November 2009. IEEE (2009) 2. Lam, T., Singh, R.: Semantically relevant image retrieval by combining image and linguistic analysis. In: International Symposium on Visual Computing, vol. 2145, pp. 770–779, November 2006. Springer, Berlin, Heidelberg (2006) 3. Bhandari, S., Kulikajevas, A.: Ontology based image recognition: a review. In: CEUR Workshop Proceedings, vol. 2145 (2018) 4. Ding, Z., Yao, L., Liu, B., Wu, J.: Review of the application of ontology in the field of image object recognition. In: Proceedings of the 11th International Conference on Computer Modeling and Simulation, pp. 142–146, January 2019 (2019) 5. Ly, N.Q., Truong, A.M., Nguyen, H.V.: Specific behavior recognition based on behavior ontology. In: Król, D., Madeyski, L., Nguyen, N.T. (eds.) Recent Developments in Intelligent Information and Database Systems. SCI, vol. 642, pp. 99–109. Springer, Cham (2016). https:// doi.org/10.1007/978-3-319-31277-4_9

Integrating Ontology with Imaging and Artificial Vision

41

6. Vizcarra, J., Nishimura, S., Fukuda, K.: Ontology-based human behavior indexing with multimodal video data. In: 2021 IEEE 15th International Conference on Semantic Computing (ICSC), pp. 262–267, January 2021. IEEE (2021) 7. Youssef, W. F., Haidar, S., Joly, P.: Generic Video Surveillance Description Ontology. In: BDCSIntell, pp. 83–88 (2018) 8. CAVIAR: Context Aware Vision using Image-based Active Recognition. http://homepages. inf.ed.ac.uk/rbf/CAVIAR/. Accessed 30 Dec 2021 9. Mazo, C., Trujillo, M., Alegre, E., Salazar, L.: Ontology-based automatic reclassification of tissues and organs in histological images. In: Proceedings of the 12th Alberto Mendelzon International Workshop on Foundations of, vol. 390, pp. 1–4 (2018) 10. Reiz, A., Albadawi, M., Sandkuhl, K., Vahl, M., Sidin, D.: Towards more robust fashion recognition by combining of deep-learning-based detection with semantic reasoning. In: AAAI Spring Symposium: Combining Machine Learning with Knowledge Engineering (2021) 11. Manzoor, U., Ejaz, N., Akhtar, N., Umar, M., Khan, M. S., Umar, H.: Ontology based image retrieval. In: 2012 International Conference for Internet Technology and Secured Transactions, pp. 288–293, December 2012. IEEE (2012) 12. Manzoor, U., Balubaid, M.A., Zafar, B., Umar, H., Khan, M.S.: Semantic image retrieval: an ontology based approach. Int. J. Adv. Res. Artif. Intell. 4(4), 1–8 (2015) 13. Studer, R., Benjamins, V.R., Fensel, D.: Knowledge engineering: principles and methods. Data Knowl. Eng. 25(1–2), 161–197 (1998) 14. Wu, H., Zhong, B., Li, H., Love, P., Pan, X., Zhao, N.: Combining computer vision with semantic reasoning for on-site safety management in construction. J. Build. Eng. 42, 103036 (2021) 15. Fang, W., et al.: A deep learning-based approach for mitigating falls from height with computer vision: convolutional neural network. Adv. Eng. Inform. 39, 170–177 (2019) 16. Fang, W., Ma, L., Love, P.E., Luo, H., Ding, L., Zhou, A.O.: Knowledge graph for identifying hazards on construction sites: integrating computer vision with ontology. Autom. Constr. 119, 103310 (2020) 17. Miller, J.J.: Graph database applications and concepts with Neo4j. In: Proceedings of the Southern Association for Information Systems Conference, vol. 2324, no. 36. Atlanta, GA, USA, March 2013 (2013) 18. Mojarad, R., Attal, F., Chibani, A., Amirat, Y.: A context-aware hybrid framework for human behavior analysis. In: 2020 IEEE 32nd International Conference on Tools with Artificial Intelligence (ICTAI), pp. 460–465, November 2020. IEEE (2020) 19. Bonatti, P., Calimeri, F., Leone, N., Ricca, F.: Answer set programming. In: A 25-Year Perspective on Logic Programming, pp. 159–182 (2010) 20. Mojarad, R., Attal, F., Chibani, A., Amirat, Y.: A hybrid context-aware framework to detect abnormal human daily living behavior. In: 2020 International Joint Conference on Neural Networks (IJCNN), pp. 1–8, July 2020. IEEE (2020) 21. Speer, R., Chin, J., Havasi, C.: Conceptnet 5.5: an open multilingual graph of general knowledge. In: Thirty-first AAAI Conference on Artificial Intelligence, February 2017 (2017) 22. Richardson, M., Domingos, P.: Markov logic networks. Mach. Learn. 62(1), 107–136 (2006) 23. Zga, A., Nini, B.: Visual relationship extraction in images and a semantic interpretation with ontologies. Int. J. Intell. Inf. Database Syst. 15(2), 223–247 (2022)

Agent-Based Simulations for Aircraft Boarding: A Critical Review Thaeer Kobbaey

and Ghazala Bilquise(B)

Higher Colleges of Technology, Computer Information Science Department, Dubai, UAE {tkobbaey,gbilquise}@hct.ac.ae

Abstract. Aircraft boarding has a direct influence on the operational cost of an airline. Therefore it has become imperative for the airline industry to find better boarding methods that minimize the boarding time thereby reducing the turnaround time of flights. Agent-Based Simulations (ABS) offer a way to investigate optimal boarding strategies. Complex interactions between multiple passengers during the process of boarding can be modelled using ABS to discover the factors causing most delays and to compare the performance of suggested new models. This study performs a critical review of 12 studies that investigate the aircraft boarding problem using ABS to compare their results. The study classifies the reviewed papers into three groups based on the objective of the investigation: (1) studies that evaluate the efficiency of various boarding strategies, (2) studies that evaluate new methods for aircraft boarding, and lastly, (3) studies that evaluate the impact of COVID-19 restrictions on aircraft boarding. Keywords: Agent-based modeling · Aircraft boarding · Boarding strategies · Simulation · COVID-19

1 Introduction The airline industry incurs millions of dollars in losses every year due to delays in turnaround time. Turn-around time may be defined as the total time the aircraft spends on land in between flights. Reducing this time even by one minute could save the airline up to $30 for every turn-around [1], leading to a yearly saving of over 50 million dollars [2]. A reduction in turnaround time also benefits airports, since it would allow the airports to accommodate more flights [3]. Therefore reducing turn-around time is an essential goal for successful airline operations. Minimizing the passenger boarding time is one of the strategies undertaken by airlines to shorten turn-around time. Boarding is defined as the time taken to get all the passengers to get seated in a flight before take-off. Reducing boarding time not only influences the operational cost of an airline but may also have a direct impact on passenger comfort by reducing waiting time leading to a relaxed journey [2]. As a result, numerous boarding strategies, such as boarding passengers randomly, in various groups, and more, have been investigated over the years some of which have been put into practice by airlines [4, 5]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 42–52, 2023. https://doi.org/10.1007/978-3-031-20429-6_5

Agent-Based Simulations for Aircraft Boarding

43

Over the last decade, the boarding problem has been studied using agent-based simulation (ABS). ABS is suitable for modelling complex behaviour and interactions among entities known as agents, which are otherwise hard to model using traditional mathematical modelling techniques [4]. Several studies have used ABS to investigate the boarding problem to understand the factors that lead to boarding delays, and to suggest new boarding strategies, especially taking into consideration social distancing requirements for COVID-19. In these studies, passengers are represented as agents in an ABS, with a defined set of attributes and the interaction between the agents allows a simulation to be created that captures complex behaviour and interferences during boarding. Although the boarding problem is critical for the airline industry, there are no research papers that provide a comprehensive critical review of the relevant literature investigating boarding challenges using ABS. To that effect, this paper aims to review 12 recent studies that have investigated the airline boarding process using ABS and present the outcomes of the studies. The studies that use agent-based modelling to analyze or recommend effective airplane boarding strategies were selected. Section 2 presents the background information on the boarding problem, highlighting the terminologies and some of the common boarding strategies recommended in the literature for alleviating this problem. Section 3 presents a review of research using ABS for investigating the boarding problem. Finally, Sect. 4 ends with a conclusion and future.

2 Background Information Boarding time is the total time taken for all the passengers to be seated in a flight before take-off. Passengers enter the aircraft forming a queue until they get to their assigned seat. The aisle of an aircraft is usually very narrow and does not allow passengers to pass other passengers in front of them. Two types of interferences normally occur during boarding, which delays passengers from getting to their seat – aisle interference and seat interference. An aisle interference occurs when a passenger blocks all the passengers behind them in the aisle while storing luggage in the overhead compartment. Seat interference occurs when the middle seat or aisle seat is blocked by a passengers who are already seated. These interferences substantially impact boarding time. 2.1 Boarding Strategies In a comprehensive study of aircraft boarding, Jaehn and Neumann [2] described three methods of boarding an airplane with assigned seats – random seating, by group and by individual. In the random strategy, passengers board the plane in any random order. In group boarding there are several strategies for boarding passengers based on certain predefined groups such as by seat type or seat location and more. Finally, boarding by seat is where every passenger is called out individually to board by his/her seat allocation. The most common among these are described below. Random. This strategy allows passengers to board in any order, usually on a first come first serve basis [6].

44

T. Kobbaey and G. Bilquise

Back to Front. In this strategy, passengers are usually divided into three groups as shown in Fig. 1. The first group consists of passengers seated in the last third of the plane, the second group consists of passengers seated in the middle third and finally the last group are the passengers seated in the front [2].

Fig. 1. Back to front

WMA (Window Middle Aisle). Also known as outside in, this strategy aims to board the passengers in groups of three to reduce seat interference. Priority is given to the window seat passengers followed by the middle seat and finally the aisle seat passengers. Figure 2 shows the WMA method in which passengers of group 1 are boarded first followed by group 2 then 3 [6].

Fig. 2. WMA

Reverse Pyramid. This strategy is a combination of both back to front and the WMA strategy. Boarding starts from the rear end of the aircraft. Passengers are boarded in groups of 4 or 5, whereby the window seat passengers at the back are given priority to board first. Figure 3 shows the reverse pyramid strategy using five groups.

Fig. 3. Reverse Pyramid

Steffen method. This is an individual seating method proposed by Steffen [7]. In this strategy, the passengers are boarded individually in a numerical order shown to reduce

Agent-Based Simulations for Aircraft Boarding

45

both seat and isle interference. Boarding starts from the rear end for every individual passenger seated in the alternate row. The strategies most commonly employed by airlines are random and back to front due to their simplicity and ease of implementation. Besides the strategies described above, there are several other strategies investigated in the literature but are not used in practice [2, 8]. All the strategies are based on the assumption of complete passenger compliance and that passengers arrive on time for boarding. Boarding strategies often impact passenger satisfaction, as they may prevent families from boarding together, or may not be easily understood by the passengers [2]. Therefore factors such as passenger friendliness and ease of implementation are also crucial to a successful strategy. Several studies have also recommended structural changes to minimize boarding time. Wallace [9] proposed a strategy called the flying carpet, in which a separate area is used to sort up to 40 passengers for boarding in groups. Zeineddine [10] proposed a novel algorithm in which passengers are organized in queues after the last passenger checks in. The algorithm sorts the passengers in an optimal queue based on seat location, group boarding (such as families), and same row boarding. Some studies have also recommended dynamic seat configurations that reduce aisle interference by allowing passengers to bypass the passenger in front [11]. ABS offers an effective method to investigate these strategies with various factors involved in boarding.

3 Literature Review This section presents a critical review of 12 studies that apply ABS to evaluate boarding strategies. The studies are classified into three groups based on the objective of the study. First is the group of studies that evaluate the efficiency of existing boarding strategies. In the second group are the studies that propose and evaluate new methods of aircraft boarding and the last group consists of studies that evaluate the impact of COVID-19 on aircraft boarding. All the reviewed studies build agent-based models under the assumption that the aircraft has a single aisle and all seats are economy class since business and first-class passengers are always boarded first and do not contribute to the boarding time [2]. Table 1 provides a summary of all the factors considered by the reviewed studies. 3.1 Studies Evaluating Various Boarding Strategies This section reviews studies that have investigated existing boarding strategies by using ABS to find the optimal strategy and identify the factors that impact boarding. The common parameters considered in all studies is the seat interference time and luggage storage time and passenger walking time. The method of computing the luggage storage time differs in each study. Moreover, each study also focuses on other specific parameters such as aircraft size, flight occupancy level, and overhead bin occupancy. Four studies have been reviewed in this category. Delcea et al. [4] compare the efficiency of the unassigned random seating strategy against other popular assigned seating strategies using an ABS. Since the random seating

46

T. Kobbaey and G. Bilquise

method largely depends on passenger’s choice of seat, the authors simulated human behaviour using passenger preference data. The study considers all the basic parameters for modeling such as walking speed, luggage storage time, and seat and aisle interference. Aircraft occupancy levels (60%, 80%, and 100%) and luggage situations (types and number of luggage) were used as independent variables to determine the boarding time. Results showed that the reverse pyramid strategy had the least boarding time regardless of the flight occupancy and passenger luggage. The worst performing strategies were the random methods regardless of seat allocation. The study fails to consider other human factors such as family seating in groups. In another study Delcea et al. [6] analyzed 24 boarding strategies using ABS. The authors considered three categories of seating strategies – random (n = 1), group seating (n = 19), and individual seating (n = 4). The study aimed to determine the boarding strategy that takes the least amount of time to board. Five independent configurable variables were considered in the simulations – the aircraft type, number of passengers, percentage of passengers with luggage, luggage storing time, and the boarding method. Passenger agents were assigned the property of speed, luggage, luggage store time, seated, and an assigned seat. 267,000 simulations were executed using NetLogo, with each boarding strategy executed 100 times with the same variable setting and the average time was recorded. Results showed that the back-to-front seating methods were best regardless of the aircraft size of the aircraft, the luggage storing requirements, and the number of passengers. Furthermore, the overall boarding time was influenced by the luggage carrying passengers and storing time required for the luggage. The study does not consider several factors such as the seat interference time, the number of luggage items to be stored, and occupancy of the overhead bin. Moreover the luggage storage time is randomly assigned to each passenger, whereas in reality luggage storage time is based on the number and size of luggage items, and occupancy of the overhead bin. The study does not incorporate human factors into the agent-based model such as delays or unwillingness to cooperate. Wittmann [19] analyzed the popular boarding strategies to identify the best strategy that minimizes boarding time and increases customer satisfaction. The passengers walking speed, the number of luggage items, and luggage storage time with respect to bin occupancy are the factors considered in the model. The back to front boarding strategy was found to be inferior compared to the WMA and reverse pyramid. Results showed that when fast walking passengers are boarded first, the time taken to board is reduced, thus concluding that slower passengers such as family, the determined ones and people carrying luggage should be boarded last. Boarding time was also impacted by the number of luggage items and bin occupancy rate. The study further went on to perform optimizations to produce a passenger-friendly boarding pattern that reduces boarding time. The result of the optimization revealed that as compared to conventional boarding strategies the optimized boarding in groups results in lower boarding time and also allowed passengers to board in groups. The study did not consider passenger compliance as a factor of boarding time. Delcea et al. [12] investigate the impact of isle and seat interference on boarding delays. The model is based on parameters such as walking speed, luggage storing time, assigned row, and seat. Each seat or isle interference it is counted and recorded in the

[6]

[12]

[4]

[13]

[8]

[14]

[15]

[16]

[11]

[17]

[18]

[19]

1

2

3

4

5

6

7

8

9

10

11

12

N/A

CA

CA

Netlogo

Netlogo

Netlogo

Netlogopython

Netlogo

Netlogo

Netlogo

Netlogo

Netlogo

ABS

CA = Cellular Automation

Study

#

✓ ✓

✓

✓

✓ ✓

✓

✓

✓ ✓

✓

✓ ✓ ✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

Number of luggage items

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

Walking speed

✓

✓

✓

Passenger compliance

✓

✓

✓

Two door boarding

✓

Aircraft size

Number of passengers

✓

✓

✓

✓

✓

Luggage size

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

Luggage storage time

Table 1. Summary of factors considered in reviewed studies

✓

Overhead bin availability

✓

✓

✓

Arrival rate

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

Seat interference

Individual walking speed

Passenger interaction

Passenger Interaction

Social distance

Social distance

Seat preference

Seat preference

Additional parameters

Agent-Based Simulations for Aircraft Boarding 47

48

T. Kobbaey and G. Bilquise

agent property to note the impacted agents. During the simulation the different types of interferences, duration and affected passengers were monitored and recorded. Results showed that several strategies such as WMA, reverse pyramid and back to font cause zero seat interference as expected. The authors’ also studied the strategies that worked best in case of no isle interference (when there is no hand luggage). Overall results showed that the best boarding strategy to reduce isle and seat interference are the ones where rear passengers are boarded first. The study did not consider several parameters that impact boarding such as number and size of luggage items, overhead bin occupancy and passenger compliance. 3.2 Studies Evaluating Infrastructural Changes or Proposing New Methods This section provides a review of studies that use ABS to evaluate new strategies such as new boarding methods or changes to the cabin layout. A total of four studies have been reviewed in this category. Schultz [11] used ABS to investigate the impact of six boarding strategies on a one-door and two-door aircraft and also studies the impact of structural changes for faster boarding. In addition to common parameters, the model considers the arrival rate of passengers and passenger compliance factors for optimal boarding. The nonconforming passengers are randomly repositioned within the group to similar noncompliance. Stochastic functions are used to determine the time of passenger arrival time, seat interference time, and luggage storing time. Results showed that the reverse pyramid and individual boarding strategies lead to a significant reduction of boarding time when two aircraft doors are used for boarding. The study does not take into account the time taken to sort the passengers before boarding. Milne et al. [13] used ABS to investigate the boarding optimization in a two-door aircraft where passengers are transported to the plane via two apron buses. The parameters considered in the simulation are the aircraft size, occupancy rate, luggage, the capacity of the aircraft, and the capacity of the apron buses. Agents are assigned a walking speed that varies based on luggage that is being carried. Luggage storing delay is computed based on the number of handbags and overhead bin occupancy. The greedy algorithm proposed by the authors populates the first apron bus with passengers in such a way that it reduces seat interference such as loading window seat passengers first. Results showed that the greedy algorithm reduces the boarding time by 8.33% against the popular algorithms used in literature, and up to 43.72% when compared to the random boarding algorithm. The study does not non-conforming passengers who may board the incorrect apron bus. Furthermore, dividing the passengers to reduce seat interference will most likely cause families to be separated while boarding, which in turn leads to passenger dissatisfaction. The study by Milne et al. [8] presents a Mixed Integer Programing (MIP) to divide passengers into groups before they are assigned to apron buses. The model suggests assigning seats on the bus based on the passenger’s vertical seat location in the aircraft. Passengers with assigned seats that are further from the airplane door are allocated to the first bus and the remaining to the second bus. The model also suggests that passengers are assigned to the bus randomly but taking into consideration their seat’s vertical location (window, middle, or aisle) and giving priority to the window seat passenger. The simulation and experiment results showed that this model works best when passengers have

Agent-Based Simulations for Aircraft Boarding

49

no luggage and only a few numbers of passengers are traveling alone. The suggested boarding approach relies on many assumptions that often do not exist in the real world. In another study, Milne et al. [16] use an ABS to investigate thirteen new boarding methods for assigning passengers to an apron bus for boarding in a two-door aircraft, while assuming a fully boarded flight. Some of the suggested strategies minimize the total boarding time by up to 36%, in comparison with other popular strategies such as the backto-front method. The simulation, based on NetLogo, considers several parameters as part of the boarding process and integrates them into the passenger agent including passenger speed, number, size of luggage items, and time to store the luggage. The suggested methods are then evaluated against the back-to-front methods as it is considered the best current method used so far. The results show that all the suggested methods improved the total boarding time by up to 36%. The proposed method operates under the assumption that the buses and the airplane are fully loaded and that passengers will comply with the boarding strategy. 3.3 Studies Evaluating Boarding with Restrictions – COVID-19 COVID-19 has significantly changed the boarding process giving rise to new parameters such social distancing, interaction between passengers, and probability of infection. The reviewed studies in this section examine the impact of COVID-19 restrictions on boarding time and analyze the risk of spreading the virus using an ABS. Schultz and Fuchte [17] investigate the transmission of the COVID-19 virus during boarding and deboarding using an ABS. The study also recommends strategies to reduce the risk of spreading the virus. The ABS model parameters specific to transmission, such as distance between passenger agents, duration, and intensity of interaction to compute the probability of transmission. Six existing boarding strategies were investigated to show that the traditional boarding strategies, without social distancing, significantly increase the transmission of the virus when an infected patient is present. The risk of infection is reduced by 75% with social distancing using random boarding strategy, however, the boarding time is nearly doubled. Furthermore, restricting the number of hand luggage items further reduced the transmission risk in a random strategy by up to 50% as aisle interference is significantly reduced. However, the group and individual boarding strategies with social distancing lead to the lowest transmission rate with regular hand luggage allowance. The model normal aircraft capacity and does not consider the transmission rates based on reduced aircraft capacity while keeping an empty airplane seat between passengers. Milne et al. [15] evaluate the performance of several boarding methods using busses while taking into consideration the safe distance between passengers in the bus and the airplane by leaving all the middle seats unoccupied. The ABS is used to evaluate three popular boarding strategies. The authors also evaluate several luggage scenarios in their simulation by relating the percentage of passengers to their luggage size, weight, and number. The simulation results showed the reverse pyramid boarding method achieved the shortest boarding time and have the best health metrics while the back to front approach has the highest boarding time. The study is limited to evaluation of methods that use apron busses and assume a two door airplane boarding.

50

T. Kobbaey and G. Bilquise

The study by Schultz and Soolaki [18] focuses on the aircraft boarding problem by investigating the COVID-19 transmission risk. The suggested method aims to reduce the time required to complete the boarding process in addition to lowering the risk of virus transmission. They recommend forming groups that guarantee minimum interaction between groups and minimum individual interaction by using a genetic algorithm. The main assumption of the suggested approach is that members of the same group may interact with each other while interaction between different groups is kept to a minimum. This assumption is suitable real-life scenarios where family members, colleagues, and friends tend to travel together. The authors apply a transmission model based on cellular automata to evaluate the risk of virus transmission by calculating the probability of a certain passenger getting infected by taking into consideration the passenger movement and interaction. Moreover, the authors introduce a new mathematical MA full capacity seating and does not investigate the risk based on keeping the middle seat empty. Milne et al. [14] present six new boarding method and evaluate their performances against another two methods. The work is focused on improving the boarding process by reducing the total boarding time while reducing the probability of virus spreading by maintain social distancing. The presented methods assume that only one door is used for boarding and that the middle seats are empty to maintain social distancing rules. The parameters of the simulation include, walking time, seat interference, aisle seat risk and window seat risk. The aisle seat risk and the window seat risk are calculated based on the probability of an infected person advancing toward the seat. The simulation results show that using back to front improves boarding time and reduce the health risk for passengers. It will also reduce the spread of the virus if an infected passenger is on board.

4 Conclusion The purpose of this study is to provide a thorough critical review of aircraft boarding using agent-based simulation. This paper presents a review of 12 studies that investigate the boarding problem using ABS. The most common boarding strategies simulated in the studies were random, WMA, back to front, reverse pyramid, and the Steffen method. Some studies also proposed new methods and evaluated the efficiency of those methods using ABS. Furthermore, in the last two years, there have been several studies that have researched the impact of COVID-19 on aircraft boarding and determine effective strategies to combat the transmission of the virus. Passengers were represented as agents in the ABS and the environment was a singleaisle aircraft (like A320). The most commonly studied parameters for analyzing the boarding time were the agent walking speed, luggage storage time, and seat interference time. While some studies assigned a random number for storing luggage, others considered the size of the luggage, number of luggage items and bin occupancy to compute the storage time. Some studies investigated only one door boarding, others also investigated boarding from front and rear door. Use of apron buses for boarding, change of cabin layout was also investigated. While these methods provided promising results, they are challenging to implement. In addition, studies that investigated the impact of COVID-19 considered additional parameters such as social distancing and passenger

Agent-Based Simulations for Aircraft Boarding

51

interaction to measure the boarding time and transmission risk. However, none of these studies considered the use of face masks in reducing the spread of the virus. The findings of the studies confirm that the group boarding strategies, such as reverse pyramid and WMA, perform the best under the operational constraints. Although the individual seating strategy, like the Steffen method, outperforms several other strategies, however this strategy is ideal when there is a large percentage of individual travelers as opposed to families. Furthermore, it was revealed that the random seating (with or without assigned seats) perform the worst in comparison with other boarding strategies. Overall it is evident that ABS provides an efficient method for evaluating various boarding strategies to examine the factors that impact the boarding process. However, the parameters used for evaluation are not consistent among the studies.

References 1. Nyquist, D.C., McFadden, K.L.: A study of the airline boarding problem. J. Air Transp. Manag. 14, 197–204 (2008) 2. Jaehn, F., Neumann, S.: Airplane boarding. Eur. J. Oper. Res. 244, 339–359 (2015) 3. Schmidt, M.: A review of aircraft turnaround operations and simulations. Prog. Aerosp. Sci. 92, 25–38 (2017) 4. Delcea, C., Cotfas, L.A., Salari, M., Milne, R.J.: Investigating the random seat boarding method without seat assignments with common boarding practices using an agent-based modeling. Sustainability 10 (2018) 5. Bazargan, M.: A linear programming approach for aircraft boarding strategy. Eur. J. Oper. Res. 183, 394–411 (2007) 6. Delcea, C., Cotfas, L.A., Paun, R.: Agent-based evaluation of the airplane boarding strategies’ efficiency and sustainability. Sustainability 10 (2018) 7. Steffen, J.H.: A statistical mechanics model for free-for-all airplane passenger boarding. Am. J. Phys. 76, 1114–1119 (2008) 8. Milne, R.J., et al.: Airplane boarding method for passenger groups when using apron buses. IEEE Access 8, 18019–18035 (2020) 9. Wallace, R.: The flying carpet (2013). http://the-flying-carpet.com/ 10. Zeineddine, H.: A dynamically optimized aircraft boarding strategy. J. Air Transp. Manag. 58, 144–151 (2017) 11. Schultz, M.: Implementation and application of a stochastic aircraft boarding model. Transp. Res. Part C Emerg. Technol. 90, 334–349 (2018) 12. Delcea, C., Cotfas, L., Cr, L., Molanescu, A.G.: Are seat and aisle interferences affecting the overall airplane boarding time ? An agent-based approach. Sustainability 1–23 (2018). https://doi.org/10.3390/su10114217 13. Milne, R.J., et al.: Greedy method for boarding a partially occupied airplane using apron buses. Symmetry (Basel) 11, 1–19 (2019) 14. Milne, R.J., Delcea, C., Cotfas, L.A.: Airplane boarding methods that reduce risk from COVID-19. Saf. Sci. 134, 105061 (2021) 15. Milne, R.J., Delcea, C., Cotfas, L.A., Ioanas, C.: Evaluation of boarding methods adapted for social distancing when using Apron buses. IEEE Access 8, 151650–151667 (2020) 16. Milne, R.J., Delcea, C., Cotfas, L.A., Salari, M.: New methods for two-door airplane boarding using apron buses. J. Air Transp. Manag. 80, 101705 (2019) 17. Schultz, M., Fuchte, J.: Evaluation of aircraft boarding scenarios considering reduced transmissions risks. Sustainability 12, 5329 (2020)

52

T. Kobbaey and G. Bilquise

18. Schultz, M., Soolaki, M.: Analytical approach to solve the problem of aircraft passenger boarding during the coronavirus pandemic. Transp. Res. Part C Emerg. Technol. 124, 102931 (2021) 19. Wittmann, J.: Customer-oriented optimization of the airplane boarding process. J. Air Transp. Manag. 76, 31–39 (2019)

Adaptive Systems and Materials Technologies in Civil Engineering: A Review Mohammed A. Ahmed(B) Civil Engineering Department, Dijlah University Collage, Baghdad, Iraq [email protected]

Abstract. Smart structure systems and smart materials technologies have a vital role in many types of manufacturing in this century. Various construction materials can considerably improve functioning. Smart Materials have a vital role in the evolution of building technology. These materials are part of a smart structural system that can detect its surroundings and perform like living systems. The study state and recognize the relation between adaptive structure system and construction materials “smart concrete” in the construction scope through analysis, new techniques and innovative methods. The study’s goal is to explore how these two professions are inextricably linked, developing ever more symbiotic as time goes on. The paper presented the various categories of smart structure systems and materials that have unique properties. Furthermore, the paper revealed the effective behavior of smart materials that take into consideration at every level of design progression of civil engineering. Keywords: Adaptive system · Smart materials · Smart concrete

1 Introduction A smart structure is a typical building or bridge that is equipped with smart materials, adaptive systems, or both as shown in Fig. 1. Basic civil engineering structures, such as buildings and bridges, or their sections, are used to carry and transmit two types of loads, static and dynamic, such as gravity and earthquakes, respectively. The term “smart material” refers to materials with unique features. Adaptive systems are intelligent systems that can able regulate themselves automatically, in response to the altering of the environment [1]. Structures become intelligent as a result of accompanying adaptive systems and smart materials (artificial or natural), allowing them to self-monitor and adjust to environmental changes. A smart structure is capable of detecting any alterations in the atmosphere or the system, diagnosing any complications at damaged places, storing and analyzing measured data. Furthermore, dictating appropriate actions promptly to advance system performance and maintain structural integrity and safety. The employment of adaptive system approaches to minimize the wind effect and earthquake response of massive building structures is still an innovative research and development technology [2]. The notion of smart material (active-material) is offered as a high-tech material that responds intelligently to climate changes through material characteristics when installed © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 53–65, 2023. https://doi.org/10.1007/978-3-031-20429-6_6

54

M. A. Ahmed

in a construction. They have a structural response (for example, creating an induced deformation field), in response to non-mechanical motivation (for example, electrical or magnetic stimulation) or a non-mechanical reaction (for example, creating an electric charge), and in response to mechanical stress (for example, an applied force) [3]. 1.1 What Are the Benefits of Smart Structures? Researchers were encouraged to look for alternatives because of limitations in the standard method to structure construction. Because of limitations in the traditional technique, researchers were driven to investigate an alternative approach to structural design for the following reasons [2]: • In order to discharge dynamic energy, the material damping is kept to a minimum. • Structures rely on stiffness only to withstand loads (earthquake and wind loads). • A limited load resistance and limited capability of dissipating energy. so, they are unable to adjust to constantly changing environmental excitations like winds or earthquakes.

Fig. 1. Materials and system role in design operation

2 Research Objectives This exploratory study intends to learn about the types of smart structures in civil engineering, as well as the elements that make a traditional structure smart. The major goal and purpose is to explore the importance and occupation of smart adaptive systems and smart materials as acquiescent ways in building or bridges, as well as their effects on structural behavior in civil engineering. This was accomplished through the classification, analysis, and behavior of intelligence (adaptive materials and systems), as well as an overview of the intelligence (materials and systems) receiving transmission structure. Building and bridging systems become smarter. It also looks at how novel technologies are being used in the building industry.

Adaptive Systems and Materials Technologies

55

3 Smart Structure System (Adaptive System) The development of smart structures is focused on several sides, including the integration and downsizing of sensors and smart materials in the structure’s components. Smart systems’ strength is manifested in their ability to bring forward contact with the outside world. Miniaturization and component integration [4]. The use of sensors or smart materials in the connections of components to the structure serves the health of the structure over its lifetime. The smart structure capabilities like monitoring, control of self-repair, dynamics of a given environment, self-monitoring, self-repair, and low-maintenance structures [5]. 3.1 Smart Structure Technology for Civil Engineering The traditional technique of designing structures (buildings, bridges, and towers) has limitations in terms of load capacity and energy consumption. To resist excitation forces and disperse kinetic energy, they are fully reliant on their stiffness and low damping. These buildings are passive because they are incapable of adapting to the ever-changing and unpredictable effects of wind and earthquakes. To enhance the structure’s strength and ductility to tolerate higher stimuli, two ways are required: first, expanding the crosssection of the essential parts of the unknown structure, and second, employing resistant building materials [2]. This might result in a design that is both ineffective and pricey. The inefficiency of customary wind and earthquake-resistant designs prompted the incorporation of smart technology in civil engineering projects [6]. It is feasible to perform the following with smart structure technology: • Add devices and provide systems to the traditional structure to make it more earthquake resistant. • Structures can disperse dynamic energy and withstand seismic force reduction. • It may greatly improve the seismic performance of structural systems, making it an appealing solution for increasing structural safety and ease of structural maintenance. • Use of the system construction resources and labor may be saved with a smart structure, lowering structural weight and expenses. 3.2 Type of Smart Structural System of Civil Engineering The type of structural system used in smart structures may be classified in the current study see Fig. 2, and these parts mostly focus on the introduction of the following: • A system that is Passive (e.g. Base-isolation system and Passive energy-Dissipation systems). • A system that is Semi-active (e.g. Semi-active damper system). • A system that is Active (e.g. Active control system). • A system that is Hybrid (e.g. Hybrid control system).

56

M. A. Ahmed

smart structure Passive system Base-isolation system

semi-active system

Active control system

Hybrid control system

Semiactive damper system

Passive energy-Dissapation systems Fig. 2. Smart structure system.

3.3 Passive System A passive system is a structural system with a simple, easily fabricated mechanism to withstand certain stimuli. But it cannot have the capacity to adapt to any change in external stimuli because this system lacks the strength and external senses to stimulate and respond [2]. • Base-isolation system: Basic isolation is a proper example of how the passive control strategy may be used. An adaptable basis is obtained by mounting a structure on a particular low lateral stiffness material, like flexible rubber see Fig. 3 [7]. • Passive energy-Dissipation systems: Mechanical devices are used in passive energy dissipation systems to distribute most of the energy which affects the structure as shown in Fig. 4, minimizing structural reactivity and the risk of structural damage [8].

Fig. 3. Elastomeric-type bearings [7]

Adaptive Systems and Materials Technologies

57

Fig. 4. Friction dampers: X-braced [8]

3.4 Semi-active System Semi-active systems are only useful for suppressing structural reactions to the dominant mode as shown in Fig. 5, such as wind-induced structural vibrations. • Semi-active dampers: which combine adaptive mechanisms to increase efficiency and intelligence are the logical progression of passive energy dissipation technology. Their adaptive technology gathers sensory and structural data to improve the damper’s performance by modifying its behavior based on these inputs [9].

Fig.5. Semi-active damper system [9]

3.5 System of Active Control The System of active control specialized equipment, such as electro-hydraulic actuators, to deliver precise regulated forces in response to seismic loads by returning recorded structural reactions as shown in Fig. 8. This control force can reduce structure vibrations produced by wind, traffic, and earthquakes by acting as extra damping. Most active control devices, which are active tendon, active bracing, and pulse generating systems, have been invented by researchers [10]. Active control seismic response has gotten a lot of attention in recent years for the following reasons:

58

M. A. Ahmed

• Increased control efficiency: current systems may now function to their full potential. • Ground movement adaptability: The system can detect ground movement and alter its control determinations accordingly. • Control selectivity: Active control systems can be developed for a range of reasons, like, safety of structure safety or social relaxation. • Applicability to various excitation devices: The structural reactivity to wind or seismic motion can be effectively reduced by a lively system (Fig. 6).

Fig. 6. Active control system [10]

3.5.1 Active Control Systems at a Basic Level The basic configuration of such a system is shown in Fig. 7, which is made up of three types of elements: sensors, actuators, and a regulator with a well-defined control process [11]. • Sensor(s): might be positioned at the structure’s base to detect external stimuli. • Controller: It collects sensor data, analyzes it, and creates the appropriate signals to control the actuator using a specified control algorithm. • Actuator: the actuator creates the appropriate control forces.

Measurements

Controller

Measurements

Control signal

Sensors

Power supply

Actuators

Sensors

Control forces

Earthquake excitation

Structure

Structural response

Fig. 7. Basic schematic of an active system for smart structure [11]

Adaptive Systems and Materials Technologies

59

3.6 Hybrid Control System A mixed control system combines active and passive control technologies. A hybrid system combines the benefits of both strategies while also improving the limitations of each methodology independently. However, the systems cross, the limitations of all systems (passive, active, and semi-active), while achieving the capabilities of the active system and the reliability of the passive system [12].

4 Smart Materials The term “smart materials” refers to a category of material systems that have distinct features [5]. Smart materials are divided into two categories: non-structural functional materials that can tolerate a wide range of stresses, and general operational functions such as shape change and stress sensing. These unusual materials have a functional capacity in addition to being activated, subjected to external loads, and capable of deforming relative to the material in which they are utilized. Furthermore, additional functionalities of these materials can be ordered, resized, and blended with other technical materials, according to the new study [3]. The response of the “smart” material to shape change under external load, which is reversed in the elongation of the model, enables modest [13]. 4.1 Characteristics of Smart Materials Smart materials can reversibly change their properties in response to various stimulus reactions. The basic characteristics that distinguish smart materials from more traditional materials used in construction are classified as shown in Fig. 8. Furthermore, they are molecules of materials and mixture assemblies. These materials have special properties, so given a possibility to utilize them in wide applications in construction and maintenance of a building [14]. Characteristics of smart materials are classified as follows:

Property change Energy exchange

Material exchanges Smart Materials

Fig. 8. Classification of smart materials

• Change properties of materials: materials have the ability to change one or more properties like thermal, mechanical, optical, electrical and magnetic, in response to altering the environmental condition [15].

60

M. A. Ahmed

• Materials, energy exchange: These materials are also known as “first law” materials since they convert input energy into a different form in order to amount produced of energy agree with the law of thermo-dynamics [16]. • Material exchange: These features refer to the material’s distinct size and direct impact. Size of a material element has ability to be reduced by removing or reducing secondary transfer arrays and additional components. Also, in certain cases even the power connection and packaging. Materials having bidirectional changing properties or energy-exchanging behavior can often be reversible or directed [17]. 4.2 Application of Smart Materials in Civil Engineering Smart concrete is a constantly developing set of smart materials that have an impact on the development of building structures and their safety. In this group, smart concrete plays a significant role [18]. By utilizing reinforcing material, concrete becomes stronger than ordinary concrete. Adding special materials have unique properties to traditional concrete, so can be obtained as “smart concrete”, as shown in Fig. 9. The additional cost of the material rises by around 30%, but it is still substantially less than the cost of installing the sensor in the structure. It can withstand higher bending pressures and absorb more energy before it breaks. Concrete is an excellent choice for highway ballast since it has the ability to produce odors [19]. The “smart” materials can be altering form in response, through subjected to external force from outside. They may be used, for example, as a small direct motor since this shape change to elongate [5]. A collection of material systems having distinctive features is referred to as “smart materials”. Smart materials are those that do not change form but instead have important qualities, such as and magneto- and electro properties [13]. 4.2.1 Self-healing Concrete Materials of self-repair are materials, that have considerable ability to minimize or eliminate the requirement for routine inspection or monitoring because they can selfrepair once damaged. The following three traits are required for a material to be selfhealing. First, provide means for storage of material inside concrete. Second, the material must be capable of releasing the repair material when it is required, and third, the material must be capable of self-healing (i.e., the repair agent must be active) [20]. Two types of self-healing materials were studied: • Auto-genic healing material: material has the ability to recover itself immediately without needing an external action Such as cementitious materials. Water interacting with cellular voids of un-hydrated cement in the concrete is the most prevalent cause, as seen in Fig. 10 [21]. • Auto-nomic healing: is defined as healing that occurs as a result of the release of an internal healing agent. Encapsulation and bacterial inclusion are two regularly employed approaches as seen in Fig. 11. Both of these procedures are capable of repairing fractures, but each has its own set of restrictions [22].

Adaptive Systems and Materials Technologies

61

Concrete

Add carbon fibers

Add phase change material (Shape Memory Alloys)

Add bacteria or Cementations materials

Reducing concrete cracking Concrete detect internal stress

Self healing concrete

Smart concrete Fig. 9. Application of smart materials

Self-healing concrete’s primary premise is that when minor fractures emerge, the selfhealing agent inside the concrete will start to restore the structure by repairing concrete itself and retrain original functional or near-original level l. This process will continue throughout time, extending the structure’s lifespan [23].

Fig. 10. Autogenic self-healing [26]

4.2.2 Shape Memory Alloys (SMA) Shape memory alloys (SMA) are alloys that, when heated over a particular temperature, may revert to their original shape after being inelastically deformed at ambient temperature. Shape Memory Effect (SME) and superelasticity are two unique features of SMA.

62

M. A. Ahmed

Fig. 11. Crack healing by bacteria [21]

The form memory effect is a phenomenon that occurs when shape memory alloys are heated and revert to their original shape. Nitinol (characteristics of nickel-titanium) is the name of the substance that has the phenomenon of shape memory. Nitinol remains the most often used shape memory alloy due to its superior thermo-mechanical and thermo-electric characteristics. Shape memory alloys’ superelastic feature refers to their capacity to withstand huge quantities of inelastic deformation and regain their shape after loading [24]. In many constructions, reinforced concrete beams with shape memory alloys are used. • Beams that are subjected to flexural and, as a result, reinforced concrete beams are prone to tensile cracking. As a result, deflection control was one of the early uses of SMA wire [25]. With this knowledge, the control of beam deflection with SMA was investigated, as illustrated in Fig. 12. The control system is proposed to be mounted to obtain the deflection measurement of concrete beams by using a sensor. The system will be activated SMA wires and deliver current when the deflection exceeds the allowable limit. This procedure is repeated until the beam deflection is slightly below the permitted deflection [26]. The temperature change must produce a force to counterbalance the deflection. • The residual stress can regulate the deflection and breadth of the crack in the loaded state and close the fracture in the no-load state by pre-stressing the wire or SMA strand in the reinforced concrete beam. The performance of a reinforced concrete piece may be considerably improved by designing it as indicated in Fig. 13. Cross wires can regulate flexural fractures, whereas transverse SMA fibers have ability to govern cracks [27].

4.2.3 Carbon Fiber Reinforced Concrete (CFRC) Adding carbon fiber to concrete boosts electrical conductivity substantially. Because of their low density and excellent strength-to-density ratio, carbon fibers are considered a preferable material other than steel fibers. Furthermore, conductive property of carbon fiber gave the concrete ability to sense cracks. This conductive concrete has the potential to be utilized as a smart material that can identify faults without causing damage [28]. The concrete reinforced by carbon fibers, not only provides detection flow but also improves mechanics owing to the fibers’ inclusion. The bearing capacity assumption of

Adaptive Systems and Materials Technologies

63

Fig. 12. Control deflection system for beam using bar of shape memory alloy inside the beam [26]

Fig. 13. Reinforced concrete beam using shape memory alloy wire [23]

carbon fiber reinforced concrete is dependent on the volumetric resistance of concrete increases during fault formation and decreasing after crack closure. This enables the identification of deterioration in concrete buildings in real-time utilizing simple and low-cost electrical instruments [29]. Mechanical strength and selfmonitoring, are making carbon fiber reinforced concrete (CFRC), a more common and inexpensive cost. This is usually less expensive than using sensors or other structural health monitoring methods [18]. When occur cracks in CFRC, it’s possibly observed by detecting the alteration in resistance caused by a change in the path connection when system is loaded externally [30]. The optical fiber’s push and pull cause this change in resistance as shown in Fig. 14.

Fig. 14. Fiber that spans a micro-crack, overstated opening for clarity [25].

64

M. A. Ahmed

5 Conclusion • The systems and materials that are imperiled and employed throughout entire building, not only as a texture or surface. As a result, the designer should think of constructing systems and materials as functional elements with altered, adaptable, and effective behavior at every level of design progression. • In the time of modern technology, the progress of contemporary building structure needs tremendously for adaptive systems and smart materials have the ability to predict the altering of the environment. The better structural solutions contribute to the handling force behind the expansion of technology, and not just a projection of what is available. • Deficiency of knowledge is all barriers to Smart Materials Innovation implementation. For certain goods, dependability has been established. Furthermore, there is a shortage of measuring of success, particularly in the area of new technology testing and certification. • There is a relationship between smart material and adaptive systems in building construction that is compound, interactive, and complicated systems and structures This relationship has become a significant driver of inventive design, and new smart materials that are beginning to develop in the area of design procedure, and it provides us with the new way of possibilities that influence how we think. • A clear understanding of materials is considerably important to design structure by an engineer. The involving material affects the performance of the structure completely. Since most smart materials are in initial stages so designers think twice to use these applications which are not been tried.

References 1. Srinivasan, A.V., McFarland, D.M.: Smart structures, analysis and design. American Association of Physics Teachers (2001) 2. Cheng, F.Y.: Smart Structures: Innovative Systems for Seismic Response Control. CRC Press (2008) 3. Gaudenzi, P.: Smart Structures: Physical Behaviour, Mathematical Modelling and Applications. Wiley (2009) 4. Udayakumar, R., Khanaa, V., Saravanan, T.: Analysis of polarization mode dispersion in fibers and its mitigation using an optical compensation technique. Indian J. Sci. Technol. 6(6), 4767–4771 (2013) 5. Krishna, J.G., Thirumal, J.: Application of smart materials in smart structures. Int. J. Innov. Res. Sci. Eng. Technol. 4(7) (2015) 6. Soong, T., Spencer, B., Jr.: Supplemental energy dissipation: state-of-the-art and state-of-thepractice. Eng. Struct. 24(3), 243–259 (2002) 7. Kelly, T.E.: Base isolation of structures: design guidelines. Holmes Consulting Group Ltd. (2001) 8. Whittaker, A.S., et al.: Seismic testing of steel plate energy dissipation devices. Earthq. Spectra 7(4), 563–604 (1991) 9. Lin, Y.-K.: Probabilistic Structural Dynamics. Advanced Theory and Applications (1995)

Adaptive Systems and Materials Technologies

65

10. Soong, T.T., Costantinou, M.C.: Passive and Active Structural Vibration Control in Civil Engineering, vol. 345. Springer (2014) 11. Miller, R., et al.: Active vibration control of large civil structures. J. Eng. Mech. 114(9), 1542–1570 (1988) 12. Ramallo, J., Johnson, E., Spencer, B., Jr.: “Smart” base isolation systems. J. Eng. Mech. 128(10), 1088–1099 (2002) 13. Udayakumar, R., Khanaa, V., Kaliyamurthie, K.: High data rate for coherent optical wired communication using DSP. Indian J. Sci. Technol. 6(6), 4772–4776 (2013) 14. Michelle, A., Daniel, S.: Smart Materials and New Technologies for the Architecture and Design Professions. Architectural Press (2005) 15. Behnoosh, M., Hamid, R., Saeeid, P.: Architecture building sustainability regarding smart materials. J. Civil Eng. Urban. 4(4) (2014) 16. Ritter, A.: Smart Materials in Architecture, Interior Architecture and Design. Walter de Gruyter (2006) 17. Gharabaghi, M., Naghdi, A.: Identifying smart materials and applying them in residential spaces in cold climate case study: City of Hamadan. Int. Res. J. Appl. Basic Sci. Sci. Explor. Publ., 9(1), 51–62 (2014). www.irjabs.com 18. Shelvay, A.M.: Reinforced concrete: applicability of smart materials. Massachusetts Institute of Technology (2012) 19. Mir, B.A.: Smart materials and their applications in civil engineering: an overview. Int. J. Civil Eng. Constr. Sci. 4(2), 11–20 (2017) 20. Rogers, C.A., Giurgiutiu, V., Leung, C.K.: Smart materials for civil engineering application. Emerging materials for civil infrastructure: state of the art. Am. Soc. Civil Eng. (ASCE) (2000) 21. Schlangen, E., Joseph, C.: Self-healing processes in concrete. In: Self-healing Materials: Fundamentals, Design Strategies, and Applications (2009) 22. Wu, M., Johannesson, B., Geiker, M.: A review: self-healing in cementitious materials and engineered cementitious composite as a self-healing material. Constr. Build. Mater. 28(1), 571–583 (2012) 23. Zhong, W., Yao, W.: Influence of damage degree on self-healing of concrete. Constr. Build. Mater. 22(6), 1137–1142 (2008) 24. Song, G., Ma, N., Li, H.-N.: Applications of shape memory alloys in civil structures. Eng. Struct. 28(9), 1266–1274 (2006) 25. Maji, A.K., Negret, I.: Smart prestressing with shape-memory alloy. J. Eng. Mech. 124(10), 1121–1128 (1998) 26. Deng, Z., Li, Q., Sun, H.: Behavior of concrete beam with embedded shape memory alloy wires. Eng. Struct. 28(12), 1691–1697 (2006) 27. Choi, E., et al.: Recovery and residual stress of SMA wires and applications for concrete structures. Smart Mater. Struct. 19(9), 094013 (2010) 28. Chung, D.D.: Self-monitoring structural materials. Mater. Sci. Eng. R. Rep. 22(2), 57–78 (1998) 29. Chen, P.-W., Chung, D.D.: Carbon fiber reinforced concrete for smart structures capable of non-destructive flaw detection. Smart Mater. Struct. 2(1), 22 (1993) 30. Chen, B., Liu, J.: Damage in carbon fiber-reinforced concrete, monitored by both electrical resistance measurement and acoustic emission analysis. Constr. Build. Mater. 22(11), 2196– 2201 (2008)

A New Method for EEG Signals Classification Based on RBF NN Shokhan M. Al-Barzinji , Mohanad A. Al-Askari , and Azmi Shawkat Abdulbaqi(B) College of Computer Science and Information Technology, University of Anbar-Iraq, Anbar, Ramadi, Iraq {shokhan.albrzinji,mohanad.abdul,azmi_msc}@uoanbar.edu.iq

Abstract. Automation is necessary since traditional EEG assessments are tedious and time-consuming, particularly the outpatient kind. For this manuscript, the researchers focused on constructing a three-class EEG classifier using FeExt and RBFNN, which stands for Radial Basis Functional Neural Network. If FeExt is finished, RBFnn may be trained to equally recognize the trends. Seizure signals are one of the various anomalies that may be identified using the EEG signal. Stable, interactive, and seizure signals are the three different types of EEG signals. This manuscript’s goal is to classify EEG signals using RBFnn. EEG signal data were relied on the CHB-MIT Scalp EEG dataset. There are 55 various FeExt schemes investigated, and a classifier is constructed that is relatively quick and accurate. The 10 morphological features of the literature were not explored or compared with the extraction techniques. According to research, the multilayer perceptron with momentum learning rule is the best classifier topology, and the FeExt algorithms PCA, Bi-gonal 2.2, coif1, DCT, db9, Re-Bi-gonal 1.1, and sym2 perform better than others. The recorded results may be effectively classified for EEG rhythm for quick examination by a neurology professional. Therefore, quick, accurate diagnosis that saves time. Using a similar method, the EEG rhythm categorization for other brain illnesses may be used. Keywords: Electroencephalogram (EEG) · Radial Basis Functional Neural Network (RBFnn) · EEG rhythm classifier

1 Introduction EEG analysis forms the basis of brain disease (Neurons disease) diagnosis and is preferred since it is non-invasive, reliable, and less costly than other methods. It utilized to analyze the activity of the brain by cerebral waves recording that setting the head electrodes along the scalp. Diagnoses consistent with EEG are typically performed manually by practitioners of medicine. The key challenge in diagnosing brain failure is to evaluate each EEG rhythm and to correlate the distortions found in different brain diseases [1, 20]. Due to irregular brain signal can occur spontaneously, monitoring a (24-h/7 days) EEG

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 66–75, 2023. https://doi.org/10.1007/978-3-031-20429-6_7

A New Method for EEG Signals Classification Based on RBF NN

67

signal becomes very repetitive and time-consuming since it can involve massive of EEG signals. It is therefore required to automate the entire EEG signals classification process and preferably diagnose these signals accurately [2, 21]. Each scalp area produces waves that allow reflecting the cerebral-health status. The EEG analysis shows many anomalies in reported signal waves in the case of diseases. In order to facilitate clinical diagnosis, the detection of these anomalies helps the physician to estimate the disorder and its level. This process can be also used in biomedical researches to investigate cerebral disease characteristics. The major problem associated with EEG analysis is the amount of EEG signals available, especially for ambulatory EEG and the inter-patient variation in the morphology of the EEG signals. It becomes a time-consuming task for classifying/separating the EEG signal and is also prone to errors induced by manual intervention. This manuscript describes the design of an automated EEG classifier [3, 22]. Three types of EEG signals are considered here, healthy, interictal, and seizure signals. The EEG data were collected from the database of CHB-MIT Scalp EEG database. Generally, the FeExt schemes of Transform and morphological are mostly preferred. The three transformation mechanisms are discussed with three other morphological FeExt in this manuscript: DFT, PCA, and DWT [4, 23]. Average classification accuracy exceeding 98.2% was reached by the classifier. The methodology of the proposed system begins from The EEG data is collecting from the CHB-MIT Scalp EEG dataset. Forty-one different feature extraction schemes are examined, along with a compact set of statistical morphological features and a reasonably accurate and fast classifier is designed. Ten morphological features and these feature extraction methods have not yet been thoroughly examined and compared in the literature. The bipolar EEG channels were selected for analysis. The EEG data used in our study were from different patients (24-h EEG recorded) from both epileptic patients and normal subjects. Digitized data were stored on an optical disc for further processing. The manuscript was organized as follows: Section 2, The Description of Dataset was introduced. Section 3, Literature of the Related Works. Changing The Features of the Domain was presented in Sect. 4. Findings and Discussion, are described in Sect. 5. And finally, in Sect. 6, The Conclusion was presented.

2 The Description of Dataset This manuscript utilized the CHB-MIT Scalp EEG database. The portable wireless headset Emotive is utilized by the Brain-Computer Interfaces (BCI) applied in this work. In total, “16 Sensors”, “2 Reference Signals and 14 Channels” are available: “AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, AF4” [5]. Figure 1 depicts the Emotive headset showing the proper placement of its electrodes. The sensors are pre-dampened with a saline solution, then applied to the scalp according to the 10–20 international standard as shown in Fig. 2 [6]. The headset’s 14 channels are sampled at a rate of 128 Hz, then digitized with a 14-bit-per-sample resolution before being applied to a built-in fifth-order Butterworth digital filter that cuts-off frequencies above 64 Hz. Furthermore, two notch filters suppress the interference of 50/60 Hz caused by the power lines [7]. All the resulted sensor signals are included in the 0.2–64 Hz frequency band and are transmitted wirelessly via a proprietary encoding/modulation on a 2.4 GHz

68

S. M. Al-Barzinji et al.

Fig. 1. The proposed system structure

carrier to a USB module on the PC [8]. In Fig. 2, the first two sensor readings are the left front sensor (AF3 and F7) and the right front sensor (AF4 and F4) are the last two sensor readings. Blinks impact the frontal sensors mainly. Two involuntary blinks occur, seen in the signals as sharp rises and falls. Voluntary continuous blinking can easily be detected at the end of the signal [9].

Fig. 2. Separation of EEG Epoch Sample Showing 2- Involuntary Blinks of the Eye Followed by a Voluntary One’s Series. (A) An 8-s EEG Signals, (B) 14-Channel of EEG Signals, (C) Calculated Thresholds for the 14 Channels (First and Eighths were Classified as Art factual Components), (D) Extracted Involuntary Eye Blink Features of 14-Channel Signals.

EEG signals utilized in the manuscript was collected from this data set consisting of 5-sets, such as “Group A”, “Group B”, “Group C”, “Group D”, and “Group E”, obtained accord the following: “Group (A)” to “Group (B)”: This consists of a normal EEG signal with eyes open/close, respectively. “Group (C)” to “Group (D)”: During the seizure-free hemisphere formation interval of the brain’s hemisphere, EEG is recorded. (i.e. Inter

A New Method for EEG Signals Classification Based on RBF NN

69

Ictal (in Latin,) (Seizure in English). “Group (E)”: here, EEG during seizure disorder shall be recorded (i.e. Ictal) [10].

3 Literature Related Works This part includes a thorough discussion with many other machine learning classifiers previous similar studies on functional extraction using linear and nonlinear approaches. Many epileptic seizure identification methods according to linearity and non-linearity of EEG signals [11] have currently been published. Function extraction strategies are crucial to differentiating between non-seizure, seized and normal EEG behavior with machine learning algorithms in the methods that these studies suggest. This includes extraction of the subband frequency, analyzing the entropy, utilize of wavelet degradation, biggest exponent from Lyapunov, fractal estimation, exponent Hurst and cumulative-high-order. Lee et al. [12] has recently suggested fuzzy approximate entropy (fApEn) and extraction method of EEG dependent WT. Studies have suggested method of overcoming a classic wavelet transformation computational load, as stated by Chen [13] for epileptic seizure behavior classification, RBFnn and logistic regression. In addition, the epileptic seizer events were categorized by the utilize of RBFnn and logistic regression. Kumar et al. [14] and the SVN for functional classification recently reported fApEn method. EEG was breaking up into subbands of discreet wavelet transforms then measured for the disorderly behavior of EEG signals by fApEN of each subband. The authors along with the RBF have recorded highest rating precision with the SVN classifier. The literature review showed that most experiments of signal processing techniques and machine learning strategies for seizure activities were not able to achieve optimal outcomes seizure-free signals of the EEG or from seizure-free EEG activity from stable EEG data. FeExt of EEG is further helpful in classifying, recognizing patterns and detecting events. Handdesigned EEG extraction techniques cause poor analysis. Recurring auto encoders FeExt for EEG are then utilized [15]. Also, the echo-state network FxExt offers better grading and clustering. The classification of motor imaging is based on b and l spatial rhythm distribution. Gradient descent and recursive techniques of classification offer less precision and pace. Consequently, the EEG classification is performed by the Multilayer Perceptron Neural Network (MLP-NN) [16]. The speed and the accuracy of the convergence are measured and matched the metaheuristic algorithm efficiency. Neurocognitive ability is the human’s mental/cognitive potential and is utilized for research on neurology. The Neuro-cognitive effect is sleep scoring. The recurring neural network [17] with the utilize of long-term memory-blocks increase the accuracy of the classification involves sleep labeling, non-fast movement of the eyes and phase N1 in sleep, this mean a transition between drowsiness/wakefulness. Deep learning is utilized for the detection of temporal dependency in EEG [18]. For capturing high-level trends in EEG, LSTM [19] will be utilized. LSTM utilizes a fully linked layer for extracting stable, epileptic features and Softmax-layer for the extraction of expected output labels, and it also maintains a high detection efficiency in the detection of devices, such as eyes and muscles movements, and background noises in captured EEG, etc. The transformation between sleep periods in the EEG and extracting its features of the time-invariant are some of the main difficulties. In [20], the authors utilized bidirectional methods CNN and LSTM

70

S. M. Al-Barzinji et al.

to pass features accurate and F1 values for different data sets, contrasting their findings and various neural network techniques with statistics. Sample Entropy

Fuzzy Entropy =

N −m+1 m 1 ∅ (r, i) i=1 N −m+1 = −log 1 N −m m+1 (r, i) i=1 ∅ N −m  N −m+1 Nj=1−m+1 ∅m (r,i,j) N − m =1 N −m+1  N −m Nj=1−m ∅m+1 (r,i,j) N − m + 1 i=1 N −m

(1)

.

(2)

4 Changing the Features of the Domain In changing domains, signals may be properly represented. There are a number of benefits to representing a signal in morphing domains, including as Frequency domain representation decreases the size of the input vector by removing irrelevant or superfluous information, which is particularly advantageous when using signal neural network architectures, compression for effective data storage, and noise reduction [21]. In the literature, there is a lot of use of transforms such the DFT, DCT, PCA, and DWT in combination with the EEG. Only major transform domain components, such as signal shape, may be maintained without considerable information loss. The parameters used to pick features were 99% signal energy retention and percent root mean difference. These components are then used to create the training input vector for the RBFnn [22] (Table 1). Table 1. Examining the performance of several RBFnn models RBFnn Best Fusibility Standard Stroke Average model configuration accuracy accuracy accuracy accuracy (%) (%) (%) (%)

Time/standard/1000 Epochs seconds)

MLP

Single hidden 90.33 layer, 10 hidden layer neurons, momentum learning

90.88

93.55

91.173336

25.981

SOFM

8 × 8 size of the mapping

89

94.98

88.87

89.3445545 305.119

RBF

50 clusters, kernel adatron

89.33

91.18

88.66

95.522251

371.223

SVN

Null

91

93.27

91

98.246346

610

A New Method for EEG Signals Classification Based on RBF NN

71

5 Findings and Discussion Table 2 lists the results of all the schemes’ performance. The performances of each transform category are assessed in terms of percent average accuracy, percent stroke accuracy, and optimum data pre-processing time in order to narrow the search for the best strategy. Bi gonal 2.2, coif1, db9, Re_ Bi gonal 1.1, sym2, DCT, and PCA are the schemes that perform better. In the case of FFT, however, the data pre-processing time for various transformations is shown in Fig. 3. Transforms like FFT, DCT, PCA, and DWT may be used to FeExt for RBFnn-based pattern categorization of EEG signals. EEG signal amplitude, Mean Power Spectral Density (MPSD), Peaks distance, Energy of the Signal, Peaks area, Singular Decomposition Value (SVD), Area under the auto-correlation curve, and signal interval are just a few of the statistical morphological features that can be used with RBFnn-based EEG classification. This collection of statistical characteristics is small, and the results are represented in a feature vector with a decreased dimension (Table 3).

Fig. 3. Pre-processing time consumed by best-performing schemes.

6 Conclusion RBFnn model consisting of single-layer MLP with momentum learning was found to perform best concerning average accuracy, stroke accuracy, and training time. It was confirmed that a minimum of 200 rhythm/class is sufficient to train the classifier. This experimentation is important since it highlights the power of the RBFnn model to learn from a comparatively small amount of data. This is a welcome result that entails the possibility of a patient, adaptable (customizable) diagnostic system. Upon experimenting with 55 different combinations of feature vector formation for the three-class problem, it was found that the best-performing schemes in terms of percentage average accuracy, percentage stroke accuracy, and low data pre-processing time are: DCT, Bi_gonal 2.2, coif1, PCA, Re_ Bi_gonal 1.1, sym2, and db9 with the combinations often statistical morphological features each. PCA is a good candidate for FeExt since it offers good accuracy as well as a compact feature set.

72

S. M. Al-Barzinji et al.

Table 2. The identifications of the Patients (P_ID) in the Health Status (H_St) may be 0 = Healthy, or 1 Non-Healthy. P_ID

H_St

P_ID

H_St

P_ID

H_St

P_ID

H_St

P_ID

H_St

P_ID

H_St

1

1

11

1

21

1

31

0

41

0

51

0

2

1

12

1

22

1

32

0

42

0

52

0

3

1

13

1

23

1

33

0

43

0

53

0

4

1

14

1

24

1

34

0

44

0

54

0

5

1

15

1

25

1

35

0

45

0

55

0

6

1

16

1

26

1

36

0

46

0

–

–

7

1

17

1

27

1

37

0

47

0

–

–

8

1

18

1

28

1

38

0

48

0

–

–

9

1

19

1

29

1

39

0

49

0

–

–

10

1

20

1

30

1

40

0

50

0

–

–

Table 3. The effectiveness of different FeExt systems Sr. no.

Structure of FeExt

1

Number of nuron in hidden layer

Accuracy of the average (%)

Number of nuron in hidden layer

Stroke accuracy (%)

Preprocessing-data time

Bi_gonal 1.1 31

94.25667

27.85

94.75

27.68604

2

Bi_gonal 1.3 22

94.02333

23.85

94.45

28.41851

3

Bi_gonal 1.5 38

94.12333

38.85

94.35

29.08835

4

Bi_gonal 2.2

46

94.42333

13.85

94.55

29.11511

5

Bi_gonal 2.4 22

94.05667

30.85

94.55

29.49316

6

Bi_gonal 2.6 23

93.99

14.85

94.25

30.15026

7

Bi_gonal 2.8 30

94.22333

15.85

94.45

30.84684

8

Bi_gonal 3.1 15

94.32333

18.85

94.35

30.6519

9

Bi_gonal 3.3 43

94.02333

18.85

94.25

31.26219

10

Bi_gonal 3.5 29

94.29

29.85

94.45

31.87712

11

Bi_gonal 3.7 29

94.35667

20.85

94.65

32.5647

12

Bi_gonal 3.9 25

93.99

34.85

94.35

33.17013

13

Bi_gonal 4.4 34

94.02333

34.85

94.55

33.47596 (continued)

A New Method for EEG Signals Classification Based on RBF NN

73

Table 3. (continued) Sr. no.

Structure of FeExt

14

Number of nuron in hidden layer

Accuracy of the average (%)

Number of nuron in hidden layer

Stroke accuracy (%)

Preprocessing-data time

Bi_gonal 5.5 13

93.95667

13.85

94.25

34.04615

15

Bi_gonal 6.8 23

94.05667

1.85

94.25

35.06008

16

coif1

29

94.15667

7.85

94.55

22.83888

17

coif2

5

93.99

18.85

94.15

23.55712

18

coif3

23

94.05667

23.85

94.35

24.37332

19

coif4

38

93.95667

38.85

94.25

25.27603

20

coif5

19

93.95667

18.85

94.25

26.30116

21

db1

24

94.25667

11.85

94.75

22.03059

22

db2

37

94.15667

37.85

94.65

22.57694

23

db3

20

94.22333

28.85

94.75

23.09459

24

db4

23

94.22333

17.85

94.75

23.52142

25

db5

14

94.15667

15.85

94.65

24.12922

26

db6

16

94.22333

2.85

94.55

24.59219

27

db7

7

94.12333

11.85

94.55

25.08093

28

db8

41

94.05667

35.85

94.45

25.55431

29

db9

19

94.35667

19.85

94.65

27.41787

30

db10

46

93.50111

46.85

93.08

26.94038

31

Re_ Bi_gonal 1.1

6

94.29

17.85

94.75

27.68604

32

Re_ 32 Bi_gonal 1.3

94.09

24.85

94.45

29.72013

42

Re_ 2 Bi_gonal 3.9

93.82333

8.85

94.05

34.3755

43

Re_ 18 Bi_gonal 4.4

93.99

22.85

94.55

34.22256

44

Re_ 4 Bi_gonal 5.5

93.95667

4.85

94.45

34.818

45

Re_ 3 Bi_gonal 6.8

93.99

34.85

94.35

35.81682 (continued)

74

S. M. Al-Barzinji et al. Table 3. (continued)

Sr. no.

Structure of FeExt

Number of nuron in hidden layer

Accuracy of the average (%)

Number of nuron in hidden layer

Stroke accuracy (%)

Preprocessing-data time

46

sym2

27

94.22333

27.85

94.55

23.02003

7

sym3

26

94.15667

20.85

94.75

23.45988

48

sym4

21

94.09

22.85

94.65

54

FFT

17

94.59

17.85

94.85

55

PCA

20

94.12333

17.85

94.15

23.80114 129.2217 6.036395

References 1. Abdulbaqi, A.S., Nejrs, S.M., Mahmood, S.D., Panessai, I.Y.: A tele encephalopathy diagnosis based on EEG signal compression and encryption In: Anbar, M., Abdullah, N., Manickam, S. (eds.) Advances in Cyber Security ACeS 2020 Second International Conference, ACeS 2020, Penang, Malaysia, December 8–9, 2020, Revised Selected Papers Penang Malaysia 2020, 12 August 2020. Communications in Computer and Information Science, CCIS, vol. 1347, pp. 148–166. Springer, Singapore (2020) https://doi.org/10.1007/978-981-33-6835-4_10 2. MM, A.M., Mohammed, A.N.:. Review on chaotic theory using DNA encoding with image encryption. Inform.: J. Appl. Mach. Electr. Electron. Comput. Sci. Commun. Syst. 2(1), 14–19 (2021) 3. Alhamadani, B.N.: Distortion detection for multi class barcode using fine localization with image recognize. Inform.: J. Appl. Mach. Electr. Electron. Comput. Sci. Commun. Syst. 2(1), 61–65 (2021) 4. Alhamadani, B.N.: Implement image encryption on chaotic and discrete transform domain encryption. Inform.: J. Appl. Mach. Electr. Electron. Comput. Sci. Commun. Syst. 2(1), 36–41 (2021) 5. Zareapoor, M., Shamsolmoali, P., Yang, J.: Kernelized support vector machine with deep learning: an efficient approach for extreme multiclass dataset. Pattern Recogn. Lett. (2017) 6. Pereira, E.H., Canedo, V.B., et al.: A comparison of performance of K-complex classification methods using feature selection. Inform. Sci. 328, 1–14 (2016) 7. Saadeh, W., Khan, F.H., Altaf, M.A.B.: Design and implementation of a machine learning based EEG processor for accurate estimation of depth of anesthesia. IEEE Trans. Biomed. Circuits Syst. 13(4), 658–669 (2019) 8. Shoeb, A.H., Guttag, J.V.: Application of machine learning to epileptic seizure detection. In: Proceedings of the 27th International Conference on Machine Learning, pp. 975–982 (2010) 9. Zeng, W., Li, M., Yuan, C., Wang, Q., Liu, F., Wang, Y.: Classification of focal and non focal EEG signals using empirical mode decomposition (EMD), phase space reconstruction (PSR) and neural networks. Artif. Intell. Rev. 52(1), 625–647 (2019). https://doi.org/10.1007/s10 462-019-09698-4 10. Abdulbaqi, A.S., Najim, S.A.D.M., Mahdi, R.H.: Robust multichannel EEG signals compression model based on hybridization technique. Int. J. Eng. Technol. 7(4), 3402–3405 (2018) 11. Fu, K., Qu, J., Chai, Y., Dong, Y.: Classification of seizure based on thetime-frequency image of EEG signals using HHT and SVM. Biomed. Signal Process Control 13, 15–22 (2014)

A New Method for EEG Signals Classification Based on RBF NN

75

12. Lee, S.-H., Lim, J.S., Kim, J.-K., Yang, J., Lee, Y.: Classification of normal andepileptic seizure EEG signals using wavelet transform, phase-spacereconstruction, and Euclidean distance. Comput. Methods Programs Biomed. 116, 10–25 (2014) 13. Chen, G.: Automatic EEG seizure detection using dual-tree complexwavelet-Fourier features. Expert Syst. Appl. 41, 2391–2394 (2014) 14. Kumar, Y., Dewal, M., Anand, R.: Epileptic seizure detection using DWT basedfuzzy approximate entropy and support vector machine. Neurocomputing 133, 271–279 (2014) 15. Sun, L., Jin, B., Yang, H., Tong, J., Liu, C., Xiong, H.: Unsupervised EEG feature extraction based on echo state network. Inf. Sci. 475, 1–17 (2018) 16. Afrakhteh, S., Mosavi, M.-R., Khishe, M., Ayatollahi, A.: Accurate classification of EEG signals using neural networks trained by hybrid population-physic-based algorithm. Int. J. Autom. Comput. 17(1), 108–122 (2018). https://doi.org/10.1007/s11633-018-1158-3 17. Michielli, N., Acharya, U.R., Molinari, F.: Cascaded LSTM recurrent neural network for automated sleep stage classification using single-channel EEG signals. Comput. Biol. Med. 106, 71–81 (2019) 18. Hussein, R., Palangi, H., Ward, R.K., Wang, Z.J.: Optimized deep neural network architecture for robust detection of epileptic seizures using EEG signals. Clin. Neurophys. 130, 25–37 (2018) 19. Doborjeh, M.G., Wang, G.Y., Kasabov, N.K.: A spiking neural network methodology and system for learning and comparative analysis of EEG data from healthy versus addiction treated versus addiction not treated subjects. IEEE Trans. Biomed. Eng. pp 0018–9294 (2015) 20. Ng, W.Y., Tan, T.E., Movva, P.V., Fang, A.H.S., Yeo, K.K., Ho, D., ... & Ting, D.S.W.: Blockchain applications in health care for COVID-19 and beyond: a systematic review. Lancet Digital Health 3(12), e819–e829 (2021) 21. Amrani, G., Adadi, A., Berrada, M., Souirti, Z., & Boujraf, S.: EEG signal analysis using deep learning: a systematic literature review. In: 2021 Fifth International Conference On Intelligent Computing in Data Sciences (ICDS), pp. 1–8. IEEE (2021, October) 22. Kashani, M.H., Madanipour, M., Nikravan, M., Asghari, P., Mahdipour, E.: A systematic review of IoT in healthcare: applications, techniques, and trends. J. Netw. Comput. Appl. 192, 103164 (2021) 23. MM, A.M., Mohammed, A.N.: Review on Chaotic Theory using DNA encoding with image encryption. Inform.: J. Appl. Mach. Electr. Electron. Comput. Sci. Commun. Syst. 2(1), 14–19 (2021) 24. MM, A.M., Mohammed, A.N.: Enhancement of Similarity for Image Segmentation. Inform.: J. Appl. Mach. Electr. Electron. Comput. Sci. Commun. Syst. 2(1), 56–60 (2021)

Intelligent Fault Detection of Rotating Machinery Using Long-Short-Term Memory (LSTM) Network A. Anwarsha and T. Narendiranath Babu(B) School of Mechanical Engineering, Vellore Institute of Technology, Vellore, Tamil Nadu 632014, India [email protected], [email protected]

Abstract. Artificial intelligence is one of the fastest-growing fields across the board. In every corner of the globe, researchers are attempting to harness its promise. Artificial intelligence’s capabilities began to be harnessed across all industries with the onset of the fourth industrial revolution. All smart industries follow the deployment of predictive maintenance with the assistance of AI. With the use of deep learning, a subset of artificial intelligence, this article describes a method for diagnosing defects in rotating machinery. The long-short-term memory framework, a class of recurrent neural network, is used to classify the faults of a rotating machine element. The experiment uses vibration data collected from rolling element bearings under various fault circumstances. The findings indicate that the LSTM network is a promising method for spotting faults in rotating machine parts such gears, rolling element bearings, shafts, rotors, and so on. Keywords: Rotating machinery · Rolling element bearings · Fault diagnosis · Long-short-term memory · Deep learning · Artificial intelligence · Recurrent neural network

1 Introduction In line with the fourth industrial revolution goals and its impact on environmental sustainability, all industries have a strong focus on predictive maintenance. Predictive maintenance is gathering and analyzing data from machines in order to improve efficiency and streamline maintenance procedures. It not only aids in gauging the condition of the equipment but also in predicting when maintenance is required. There are a variety of methods available today for predicting the status of a machine and fixing the faults even in the early stages [1–6]. Figure 1 shows various artificial intelligence methods used in the defect classification of rotating machines. The long-short-term memory framework, a type of deep learning, is used in this article to discover defects in rotating machinery. Deep learning architectures for instance recurrent neural network (RNN), deep belief network (DBN), convolution neural network (CNN), autoencoders (AE), and others have been found as being capable of defect diagnostics of rotating machinery in recent years [7–10]. With the use of DBN and one-dimensional CNN, Li et al. [11] recommended a © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 76–83, 2023. https://doi.org/10.1007/978-3-031-20429-6_8

Intelligent Fault Detection of Rotating Machineryds

77

unique approach for identifying problems in rotating machinery in 2019. Using recurrent autoencoders, Kong et al. [12] suggested a procedure for measuring the fault severity and identifying early weak defects in rotating equipment. Li et al. [13] suggested a three-step intelligent failure diagnostic technique for rotating equipment on the basis of convolution neural network and Bayesian Gaussian mixture. For complex environments, a unique fusion deep learning approach on the basis of expanded deep CNN with wide first-layer kernels and LSTM framework is suggested in [14] to improve the independent learning ability and intelligent diagnostic accuracy of deep learning for rotating machinery. Han et al. [15] proposed a technique for the diagnosis of rolling element bearing faults using a bidirectional LSTM application and a CNN-capsule network. In a study, some researchers proposed a unique approach based on the stacked autoencoder and LSTM framework for the fault identification of REB [16]. Although many individuals have tried fault diagnosis using deep learning, RNN, notably LSTM technique has been rarely tried. This is what distinguishes this paper.

Fig. 1. Various types of artificial intelligence methods used in the defect classification of rotating machine.

2 An Overview of the LSTM Network The long-short-term memory network (LSTM) is a developed form of RNN that can alleviate the trouble of fading gradients in traditional RNN design. LSTM can store both present and historical learning information for a task in long-term memory, this property distinguishes the LSTM network from other networks [17]. The forget gate, input gate, update gate, and output gate are the four gates of an LSTM network as given in Fig. 2. First, let’s talk about the forget gate. This gate decides which information should be remembered and which should be ignored. The present input x t and the prior hidden stage ht−1 are sent to the sigmoid operator, which determines whether or not the

78

A. Anwarsha and T. Narendiranath Babu

old output information is required. The value of f t will be utilized later. The following equation can be used to express the forget gate at t.   (1) ft = σ (wf . ht−1 , xt + bf ) where f t refers to the forget gate, σ refers to the sigmoid operator, wf refers to the weight matrix between forget gate and input gate, ht−1 refers to the hidden state at the former time-stamp, x t represents the input at the present time-stamp, bf refers to the connection bias corresponds to the forget gate.

Fig. 2. Structure of an LSTM network.

The input gate is the next component, and it is responsible for updating the cell status. The second sigmoid function receives the present state x t and the formerly hidden state ht−1 . The values are changed from 0 to 1. One is for important information, and the other is for non-important information. The tanh function will then be used to pass the identical information from the hidden state and present state. The tanh function will construct a vector ct with all the possible values in the range of −1 and 1 to regulate the framework. The activation functions generate output values that are prepared for point-by-point multiplication. The following equation can be used to express the input gate.   (2) it = σ (wi ht−1 , xt + bi )   ct = tanh(wc ht−1 , xt + bc )

(3)

where it stands for input gate at time-stamp t, wi stands for weight matrix between input gate and output gate, bi stands for bias vector corresponding to input gate, ct stands for the value produced by tanh, wc stands for the weight matrix between cell state information and output, bc stands for bias vector corresponds to wc . The update gate, also known as the cell state, comes next. The forget gate and input gate have provided enough information to the network. The data of current status must

Intelligent Fault Detection of Rotating Machineryds

79

then be decided and set aside in the cell-state. The prior cell-state C t−1 is magnified by the value of f t . The framework then performs summation on the input vector’s outcome value, updating the cell state and assigning a new cell state C t to the network. Ct = ft ∗ Ct−1 + it ∗ ct

(4)

where C t stands for cell-state information at time-stamp t, C t−1 stands for cell state information at the previous time-stamp. The output gate, last but not least, controls the cost of the next hidden stage. This stage saves information from previous inputs. The 3rd sigmoid-operator receives the existing stage and former hidden stage values first. The tanh operator is then used to construct an updated cell-state from the old cell state. The framework selects which data the hidden stage should convey based on the final value. Prediction is based on this hidden state. The latest cell-state and hidden stage are then move forward to the next step.   (5) Ot = σ (wo ht−1 , xt + bo ) ht = Ot ∗ tanh(Ct )

(6)

where, Ot stands for output gate at time-stamp t, wo stands for weight matrix of output gate, bo stands for bias vector corresponds to wo , ht stands for the LSTM output.

3 Experimental Setup

Fig. 3. Experimental setup for the defect detection of a rolling element bearing.

This section explains how to set up an experimental testbed for rotating equipment defect diagnosis. The process is demonstrated using healthy and faulty rolling element bearings. A typical experimental setup for the defect detection of a rolling element bearing is given in Fig. 3. An induction motor, a dynamometer, and a torque transducer make up the system. Two test bearings can be provided to the induction motor, one is attached at the fan-end and the other is attached at the driving end. The vibration data for the experiment was taken from Case Western Reserve University’s bearing data center [18]. The accelerometers mounted to the induction motor’s casing are used to capture vibration data in both well and damaged circumstances. The vibration data is supplied into the MATLAB software, which uses it to implement the LSTM design (Table 1).

80

A. Anwarsha and T. Narendiranath Babu Table 1. Experimental dataset.

Fault types

Datasets

Motor load

Motor speed

Fault size

Normal state

12 datasets

0 HP

1797 rpm

No fault

Inner race fault

12 datasets

0 HP

1797 rpm

0.007 inch

Outer race fault

12 datasets

0 HP

1797 rpm

0.007 inch

Ball fault

12 datasets

0 HP

1797 rpm

0.007 inch

4 The Proposed Method In this part, it is described that an LSTM network may be utilised to diagnose faults using the vibration data from REB obtained with the aid of accelerometers. In our earlier research, machine learning methods were used to diagnose faults [19]. Before applying the ML model, feature engineering or feature extraction was done there. However, when using deep learning methods, feature extraction is not necessary. One benefit of deep learning architecture over machine learning techniques can be seen in this.

Fig. 4. Different layers of the proposed architecture.

Fig. 5. Training plot- accuracy vs. iteration.

A total of 48 vibration datasets, each containing 9000 entries were chosen to demonstrate the experiment, with 12 datasets corresponding to healthy bearings and labeled as ‘Healthy’, 12 datasets corresponding to inner race defect and labeled as ‘Inner Race Fault’, 12 datasets corresponding to outer race defect and labeled as ‘Outer Race Fault’, and the remaining 12 datasets corresponding to ball fault and labeled as ‘Ball Fault’. Then, to perform the LSTM network, divide the total datasets into training and testing. Here, 80% of the datasets are allocated to training and 20% to testing.

Intelligent Fault Detection of Rotating Machineryds

81

Fig. 6. Training plot-loss vs. iteration.

Fig. 7. Confusion matrix for LSTM network.

The LSTM architecture must then be defined. Long-term dependencies between time steps of sequence data can be learned by means of LSTM networks. The bidirectional LSTM layer is used in this experiment because it examines the series in both forward and backward paths. The architecture’s process is depicted in Fig. 4. It is made up of five-layer arrays. A sequence-input layer, a bi-directional-LSTM layer with 100 hidden levels, four fully-connected layers, one softmax layer, and a classification layer are all included. After that, choose from a variety of LSTM options, such as training choices, maximum epochs, mini-batch size, starting learning rate, sequence length, and so on. Then, by providing all of the aforementioned alternatives, train the network.

82

A. Anwarsha and T. Narendiranath Babu

5 Results and Discussion Figures 5 and 6 depict the training progress reports. The trained network’s accuracy is shown in Fig. 5, while its loss is shown in Fig. 6. The network’s accuracy has achieved 100% after the seventh iteration. Similarly, network’s loss has reduced to zero after sixteenth iteration. The next step is to estimate the training and testing accurateness, which denotes the classifier’s accurateness on the indications it was trained and tested on. Confusion matrix is employed to show the efficiency of a model on a collection of information for which the right values are recognized in classification issues. The goal class is the signal’s ground-truth condition, and the output class is the signal’s label assigned by the network. The class labels on the axes are healthy, inner race defect, outer race defect, and ball defect, respectively. From the confusion matrix given in Fig. 7, it can be found that the classification accuracy of the suggested approach is 100%. That means the long-short-term memory framework successfully classified all the conditions of the bearings.

6 Conclusions This research proposes a system for malfunction diagnostics of rotating machinery applying a long-short-term memory network. In this experiment, healthy and faulty rollingelement bearings were utilized to show the implementation of LSTM architecture in the defect detection of rotating machinery. The results show that the suggested approach’s classification accuracy is 100%, that means the LSTM network is a promising technique for detecting flaws in rotating machine components such as gears, rolling element bearings, shafts, rotors, and so on. The real-time data collected for this project is just for demonstration purposes. The percentage of accuracy may be affected if the dataset is too huge. In that case, feature engineering should be considered as well. This will be reflected in our future efforts.

References 1. Hamadache, M., Jung, J.H., Park, J., Youn, B.D.: A comprehensive review of artificial intelligence-based approaches for rolling element bearing PHM: shallow and deep learning. JMST Adv. 1(1–2), 125–151 (2019). https://doi.org/10.1007/s42791-019-0016-y 2. Nath, A.G., Udmale, S.S., Singh, S.K.: Role of artificial intelligence in rotor fault diagnosis: a comprehensive review. Artif. Intell. Rev. 54(4), 2609–2668 (2020). https://doi.org/10.1007/ s10462-020-09910-w 3. Liu, R., Yang, B., Zio, E., Chen, X.: Artificial intelligence for fault diagnosis of rotating machinery: a review (2018). https://doi.org/10.1016/j.ymssp.2018.02.016 4. Alshorman, O., Irfan, M., Saad, N., Zhen, D., Haider, N., Glowacz, A., Alshorman, A.: A review of artificial intelligence methods for condition monitoring and fault diagnosis of rolling element bearings for induction motor (2020). https://doi.org/10.1155/2020/8843759 5. Anwarsha, A., Narendiranath Babu, T.: A review on the role of tunable q-factor wavelet transform in fault diagnosis of rolling element bearings. J. Vibr. Eng. Technol. (2022). https:// doi.org/10.1007/s42417-022-00484-1

Intelligent Fault Detection of Rotating Machineryds

83

6. Anwarsha, A., Narendiranath Babu, T.: Recent advancements of signal processing and artificial intelligence in the fault detection of rolling element bearings: a review. J. Vibroeng. 24 (2022). https://doi.org/10.21595/JVE.2022.22366 7. Shao, H., Jiang, H., Wang, F., Wang, Y.: Rolling bearing fault diagnosis using adaptive deep belief network with dual-tree complex wavelet packet. ISA Trans. 69, 187–201 (2017). https:// doi.org/10.1016/j.isatra.2017.03.017 8. Liu, H., Zhou, J., Zheng, Y., Jiang, W., Zhang, Y.: Fault diagnosis of rolling bearings with recurrent neural network-based autoencoders. ISA Trans. 77, 167–178 (2018). https://doi.org/ 10.1016/j.isatra.2018.04.005 9. Jiang, H., Li, X., Shao, H., Zhao, K.: Intelligent fault diagnosis of rolling bearings using an improved deep recurrent neural network. Measur. Sci. Technol. 29 (2018). https://doi.org/10. 1088/1361-6501/aab945 10. Janssens, O., et al.: Convolutional neural network based fault detection for rotating machinery. J. Sound Vib. 377, 331–345 (2016). https://doi.org/10.1016/j.jsv.2016.05.027 11. Li, Y., Zou, L., Jiang, L., Zhou, X.: Fault diagnosis of rotating machinery based on combination of deep belief network and one-dimensional convolutional neural network. IEEE Access. 7, 165710–165723 (2019). https://doi.org/10.1109/ACCESS.2019.2953490 12. Kong, X., Li, X., Zhou, Q., Hu, Z., Shi, C.: Attention recurrent autoencoder hybrid model for early fault diagnosis of rotating machinery. IEEE Trans. Instr. Measur. 70 (2021). https://doi. org/10.1109/TIM.2021.3051948 13. Li, G., Wu, J., Deng, C., Chen, Z., Shao, X.: Convolutional neural network-based bayesian gaussian mixture for intelligent fault diagnosis of rotating machinery. IEEE Trans. Instr. Measur. 70 (2021). https://doi.org/10.1109/TIM.2021.3080402 14. Gao, Y., Kim, C.H., Kim, J.M.: A novel hybrid deep learning method for fault diagnosis of rotating machinery based on extended WDCNN and long short-term memory. Sensors 21 (2021). https://doi.org/10.3390/s21196614 15. Han, T., Ma, R., Zheng, J.: Combination bidirectional long short-term memory and capsule network for rotating machinery fault diagnosis. Measur.: J. Int. Measur. Confed. 176 (2021). https://doi.org/10.1016/j.measurement.2021.109208 16. Shi, H., Guo, L., Tan, S., Bai, X.: Rolling bearing initial fault detection using long short-term memory recurrent network. IEEE Access. 7, 171559–171569 (2019). https://doi.org/10.1109/ ACCESS.2019.2954091 17. Ning, S., Wang, Y., Cai, W., Zhang, Z., Wu, Y., Ren, Y., Du, K.: Research on intelligent fault diagnosis of rolling bearing based on improved ShufflenetV2-LSTM. J. Sensors 2022 (2022). https://doi.org/10.1155/2022/8522206 18. Case Western Reserve University: Bearing Data Center-Seeded Fault Test Data. https://eng ineering.case.edu/bearingdatacenter. Accessed 1 June 2022 19. Anwarsha, A., Narendiranath Babu, T.: Artificial intelligence-based fault diagnosis procedure for a sustainable manufacturing industry. In: IOP Conference Series: Earth and Environmental Science, vol. 1055, p. 012012 (2022). https://doi.org/10.1088/1755-1315/1055/1/012012

What People Post During the Movement Control Order (MCO): A Content Analysis of Intagram’s Top Posts Hong Lip Goh1

, Wen Hui Foo1 , Tat Huei Cham2 and Way Zhe Yap1(B)

, Bee Chuan Sia1

,

1 Faculty of Accountancy and Management, Universiti Tunku Abdul Rahman, Kuala Lumpur,

Malaysia {gohhl,siabc}@utar.edu.my, [email protected], [email protected] 2 UCSI Graduate Business School, UCSI University, Kuala Lumpur, Malaysia

Abstract. The outbreak of the COVID-19 pandemic in the year 2020 led the government of Malaysia to impose the Movement Control Order (MCO) to limit the spread of the virus, which resulted in people becoming heavily reliant on social media as a means of communication and expression of thought. Analysing the trends and sentiments of popular content posted on social media can provide insight into the market for marketers to plan and execute their social media strategies and marketing communication campaigns. Using qualitative content analysis (QCA), this research aims to identify the overall image content and caption sentiments reflected in the top Instagram posts posted during the first MCO, from the middle of March to June of the year 2020. A total of 200 posts were collected on Instagram and were filtered with the hashtags ‘#mco’ and ‘#cmco’. The content of the posts was categorised into food, gadgets, captioned photos, pets, activities, selfies, fashion, landscapes, portraits, and videos; while the sentiments of the posts were categorised into narratives of fear, fun, sadness, responsibility, and encouragement. The results show that portraits and narratives of fun were dominant during the first MCO, suggesting that positive and portrait-centric posts on social media are preferred over other types during a crisis, which might provide valuable insight for businesses to design their social media strategies. Keywords: Content analysis · Movement control order · Covid-19 · Instagram · Social media analysis

1 Introduction Ubiquitous internet access and the use of social media have become the norm of our current society. Social media refers to any platform where people can create, modify, share and discuss content posted on the internet. With the increasing usage of social media, an increasing number of people have begun to use these platforms as online, public diaries or, in the case of businesses or online “influencers”, as promotional tools © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 84–94, 2023. https://doi.org/10.1007/978-3-031-20429-6_9

What People Post During the Movement Control Order (MCO)

85

[1–4]. Following the outbreak of the COVID-19 pandemic, these activities have only increased, as social media has grown to be seen as a medium for crisis communication [5] and as a channel to voice one’s emotions. The Malaysian government imposed the Movement Control Order (MCO) to combat the spread of the virus [6, 7]. During the MCO, people were only allowed to leave the premises of their homes for essential activities, such as purchasing food or for medical purposes [1, 8]. The sudden outbreak of the pandemic and the need to be quarantined greatly impacted the livelihoods of most people [9], and social media became a window for crisis communication, leisure, communication with loved ones and a means to conduct business [10]. Social media provided notable support for the community during lockdown, as it helped create a sense of normalcy and helped maintain the social networks of people [11]. This prompted people to share their thoughts on the MCO and the COVID-19 pandemic on social media, as well as share how social media was assisting them in weathering out the difficult conditions they were under. The same could be seen in the activities of marketers and brand accounts, which bolstered their online activities during the MCO, since social media was their only channel for reaching the masses. The sentiments shown during this trying time, through messages posted on social media, are worth researching, as they might provide insight into trends and online phenomena that could be used by product and service providers to promote their products and services [12]. Instagram has been identified as a suitable platform to study, due to its relatively simple content presentation. It is currently one of the most popular social media platforms in the world [13] and is widely used in Malaysia [14]. Therefore, this study aims to examine the content posted on the Instagram accounts of Malaysians and analyse the sentiments reflected in them via the method of content analysis. The findings of this study will enable business owners to gain a better understanding of consumer needs, online trends and worthwhile sentiments that should be communicated to the public when using Instagram as a strategic channel of communication for promoting products and services.

2 Literature Review In studying the characteristics of messages, qualitative research via the method of mixed method content analysis is often used to ensure that the study is conducted systematically and objectively. These studied messages can be derived from written text, transcribed speech, graphical elements, pictorial images, moving images, music, sounds and even nonverbal behaviour [15, 16]. According to [17, 18], sentiment analysis can provide insight into an individual’s personal narrative of an event. This narrative may reveal the individual’s feelings of positivity towards the event, such as optimism, joy and happiness, or their feelings of negativity towards it, such as pessimism, fear and despair. Through content analysis, embedded messages from the media may be further analysed for inferential purposes, as well. This present study’s method of categorising the content of the images obtained from Instagram was adapted from the works of [19, 20]. Certain amendments were made to the foundational framework of those studies, so that the categories would be more

86

H. L. Goh et al.

relevant to the current study. For instance, both past studies did not include the video format as a content category, but, in this study, video was included as a category. As for the sentiment categories of the captions on the Instagram posts, a new framework was developed. While most of the categories were adapted from the framework of [21], the narrative of annoyance and the narrative of resistance were replaced by the narrative of sadness, which is an emotion category adapted from [22]. The narrative of encouragement was also added to match the context of this research. Table 1 are the categories used by this study to classify and identify the content on Instagram: Table 1. Categories for images and caption sentiments Images

Description

Food

Food, recipes, cakes, drinks, etc.

Gadgets

Electronic goods, tools, motorbikes, cars, etc.

Captioned photos

Pictures with embedded text, memes and so on

Pets

Animals, like cats and dogs, which are the focus of the picture

Activities

Both outdoor and indoor activities, or places where activities are done

Adapted from [19, 20]

Selfies

Self-portraits: only one subject (human) is present in the photo

Fashion

Shoes, costumes, makeup, personal belongings, etc

Landscapes

The features of an area of land are visible and focused on in the picture

Portraits

Representations of a person

Videos

Recordings of moving visuals

Captions

Description

Narrative of fear

The narrative of fear is the category for captions that express worry Adapted and concern over COVID-19 and its impact from [21, 22] The narrative of fun is the category for captions that present joy

Narrative of fun

and happiness—or optimism, even—during hard times

Narrative of sadness

The narrative of sadness is the category for captions that talk about missing elements and the manifestation of sad emotions

Narrative of responsibility

The narrative of responsibility is the category for captions that call upon the community to play their role as responsible citizens and community members by, for example, staying at home and practising proper hygiene

Narrative of The narrative of encouragement is the category for captions that encouragement call out to the community to stay positive. Self-encouragement also falls under this category

What People Post During the Movement Control Order (MCO)

87

3 Research Method This research used qualitative content analysis (QCA), and Instagram was chosen as the platform to be studied, due to the simplicity of its content presentation. Hashtags of ‘#mco’ and ‘#cmco’, were used to identify relevant posts through Instagram’s search function. The search process was completed with default settings, and posts were merely suggested by Instagram without any modifications. Business accounts were excluded, so only personal accounts were considered for this study. The top 100 posts of each hashtag were extracted for further analysis. Instagram ‘Stories’ and ‘Reels’ were not considered for analysis, due to the difficulty of collecting their ad-hoc data. The 200 posts excluded irrelevant posts such as Orlando International Airport in the United States that uses the same code as well. The time frame for the data collection process was from March to June, 2020. The data obtained was dated from 29 March, 2020, to 29 June, 2020; where only 4 of the posts fell under the months of March and April, while 60 of the posts were posted in May and 136 were posted in June. The content of the images was then coded into the relevant categories listed in Table 1, while captions were coded based on the sentiments in Table 2. Figure 1 shows the elements of the Instagram posts that were analysed in this research.

Fig. 1. Post elements for content analysis

Screenshots of the posts were used to determine the types of content Instagram users shared during the MCO. Captions were the main component used to analyse the sentiments displayed in the posts. Emojis and hashtags were used as references when analysing the sentiments. In order to minimise the subjectivity of the categorisation of each caption’s narrative, a second coder was trained to classify 100 posts, or 50% of the data, and the results showed a coefficient of >0.90, which demonstrates substantial agreement between the coders [23]. Table 2 shows examples of the image content

88

H. L. Goh et al.

categories, while Table 3 shows examples of the captions’ content categories, identified from the collected posts. Table 2. Examples of image content analysis using QCA Category

Example

Remark

Food

Images that have food as their main subject

Gadgets

Images that have products as their main subject

Captioned photos

Images that have embedded text

Pets

Images that have domestically kept animals

Activities

Images that show an activity being carried out by the subject

Selfies

Selfies can include multiple-self-portraits and mirror selfies

Fashion

Interesting clothes or costumes worn by the subject of the photograph

Landscapes

Images that stress on landscape, scenery or architecture

(continued)

Results from the image categories’ analysis show that nearly half of the total posts were portraits, which had the highest rate of occurrence (48.5%), followed by selfies

What People Post During the Movement Control Order (MCO)

89

Table 2. (continued) Category

Example

Remark

Portraits

Images that are photographed with a theme in mind

Videos

Pictures in motion that are easily identifiable

Table 3. Examples of caption content, analysed using QCA Category

Example

Narrative of fear life feels stuck recently’

Remark ‘My The author is showing concern for his general state of life, which has been negatively affected by the prolonged isolation of the MCO

Narrative of fun

‘Happy Sunday ’, or hashtags such as, ‘#cute’, ‘#love’, ‘lovely’ and ‘#happy’

The words and hashtags indicate a joyful sentiment

Narrative of sadness

‘#Throwback to when we look forward to wake up to our breakfast buffer everyday during a vacation praying the crisis goes away soon so we could resume doing what we love. In the meantime, stay safe everyone’

Expresses that the author misses the ‘good old days’. The usage of the sad face emoji reveals sad sentiments

Narrative of responsibility

‘Day 47: After 46 days of quarantine I’m going out to work and I promise it won’t take too long and will follow the rules’

The author is doing his part as a responsible citizen during this challenging time

Narrative of encouragement ‘Let’s face it with positive attitude’

The author is encouraging the community and the public to deal with the pandemic and MCO in a positive manner

(24.5%), activities (7.5%), food (5.5%) and videos (4.5%). Notably, most of the videos

90

H. L. Goh et al.

showcased acts like dancing and singing. The list continues with gadgets and landscapes, both of which had a 2.5% rate of occurrence, and then captioned photos, pets and fashion, which each took up 1.5% of the total percentage, respectively. The total sum of all the captions recorded was 215, although there were only 200 posts included in this study. This shows that some of the posts expressed more than one sentiment in their captions; long captions may have multiple sentiments attached to them. The narrative of fun had the highest frequency among all the categories, where there were 144 (67.0%) posts that had identifiably happy and joyful emotions attached to their captions. Following that is the narrative of sadness, which had 25 posts (11.6%), and then the narrative of encouragement, which had 22 posts (10.2%). The narrative of responsibility comes next with 16 posts (7.5%), and the narrative of fear is ranked last, with only 8 posts (3.7%). The results of the analysis are shown in Table 4. Table 4. Content types and narratives of collected Instagram images and captions Image category Food

Frequency

Percentage

11

5.5

Gadgets

5

2.5

Captioned photos

3

1.5

Pets

3

1.5

Activities

15

7.5

Selfies

49

24.5

Fashion

3

1.5

Landscapes

5

2.5

97

48.5

9

4.5

Total

200

100

Caption sentiments

Frequency

Percentage

Portraits Videos

Narrative of fear

8

Narrative of fun

144

3.7 67.0

Narrative of sadness

25

11.6

Narrative of responsibility

16

7.5

Narrative of encouragement

22

10.2

215

100

Total

4 Discussion The results show that the dominant form of content shared by Malaysians on Instagram during the MCO are portraits and selfies. This aligns with previous studies, which have

What People Post During the Movement Control Order (MCO)

91

shown that designed photos (portraits) are one of the most common types of pictures on Instagram [24]. Content creators using selfies as endorsement tools for product promotion [25], and the general inability of the public to travel during lockdown, may also have contributed to the high number of selfies that were posted. The results of the caption analysis show that the narrative of fun was the dominant sentiment shared in posts during MCO. This result is consistent with the study done by [26], wherein it was found that Malaysians were generally optimistic about battling the COVID-19 pandemic and positive attitudes were frequently presented by them on Instagram [27]. The narrative of sadness being the next most dominant result may be due to people missing their loved ones, and missing the freedom to perform activities, and could also be a reflection of the stress felt due to prolonged isolation [28, 29]. The narrative of encouragement was ranked third, which shows that Malaysians did not forget to spread positivity during the MCO and were dedicated to motivating their community to start new projects, which is consistent with the study of [9]. The results of this study suggest that Malaysian Instagram users are generally favoured towards fun, optimistic and positives posts, combined with well-designed portraits and attractive selfies, as these are the types of posts which garnered the most likes during the MCO and were the most frequently suggested to users by Instagram’s basic algorithm. This is consistent with [30], which states that social media platforms are more receptive towards positive emotions than negative ones. Therefore, businesses and marketers should strategize their social media communication accordingly by posting fun promotional content that is also aesthetically pleasing [31–33].

5 Conclusion The paper provides preliminary results on the sentiments reflected in the top Instagram posts posted during lockdown. Through QCA, it was revealed that the narrative of ‘fun’ was noticeably dominant in the captions of these top posts, while portraits and selfies were the favoured style of images to be posted, indicating that the dominant attitudes of people during the pandemic were those of resilience and optimism. These sentiments were also seen in the Instagram posts themselves, where positive, happy atmospheres and smiling faces, seen through the use of portraiture, gained more attention than any other type of post during this time. Hence, business entities should design their promotional material with the results of this study in mind, emphasising the narrative of fun in their advertisements and online content. This paper’s scope is limited by its data collection method, which was limited to the default search function of Instagram, personal accounts, and top posts (ad-hoc data, such as Instagram ‘Stories’, were excluded). The data collected in this study might also have been affected by back translation, as many of the posts analysed were originally written in languages that were not English, such as Chinese and Malay. Coding of the content (for both images and captions) was confined to frameworks established in prior studies, and future studies might be able to identify new categories for a more sophisticated interpretation of the presented narratives. Future research may also want to examine content posted during a different time frame or on platforms that are not Instagram. This could further enrich the existing discussion surrounding social media and provide deeper insight into the shifting sentiments

92

H. L. Goh et al.

of people, perhaps shedding light onto new methods marketers could use to capitalize on these changing attitudes [34–37]. Despite its limitations, this study was able to examine, identify and classify the content (both images and captions) posted on Instagram during the pandemic and was able to establish a connection between the categories and sentiments reflected in these posts.

References 1. Cham, T.H., Cheng, B.L., Lee, Y.H., Cheah, J.H.: Should I buy or not? Revisiting the concept and measurement of panic buying. Curr. Psychol. 1–21 (2022). https://doi.org/10.1007/s12 144-022-03089-9 2. Cham, T.H., Lim, Y.M., Sigala, M.: Marketing and social influences, hospital branding, and medical tourists’ behavioural intention: before-and after-service consumption perspective. Int. J. Tour. Res. 24(1), 140–157 (2022) 3. Cham, T.H., Cheng, B.L., Ng, C.K.Y.: Cruising down millennials’ fashion runway: a crossfunctional study beyond Pacific borders. Young Consum. 22(1), 28–67 (2020) 4. Low, M.P., Cham, T.H., Chang, Y.S., Lim, X.J.: Advancing on weighted PLS-SEM in examining the trust-based recommendation system in pioneering product promotion effectiveness. Qual. Quant. 1–30 (2021) 5. Guidry, J.P., Jin, Y., Orr, C.A., Messner, M., Meganck, S.: Ebola on Instagram and Twitter: How health organizations address the health crisis in their social media engagement. Pub. Relat. Rev. 43, 477–486 (2017) 6. Elengoe, A.: COVID-19 Outbreak in Malaysia. Osong Pub. Health Res. Perspect. 11(3), 93–100 (2020). https://doi.org/10.24171/j.phrp.2020.11.3.08 7. Tang, K.H.: Movement control as an effective measure against Covid-19 spread in Malaysia: an overview. J. Pub. Health (2020). https://doi.org/10.1007/s10389-020-01316-w 8. Prime Minister’s Office of Malaysia.: Restriction of Movement Order (2020). Retrieved from https://www.pmo.gov.my/2020/03/movement-control-order/ 9. Grover, S., Sahoo, S., Mehra, A., Avasthi, A., Tripathi, A., Subramanyan, A., … Reddy, Y.J.: Psychological impact of COVID-19 lockdown: an online survey from India. Indian J. Psychiat. 62(4), 354–362 (2020).https://doi.org/10.4103/psychiatry.indianjpsychiatry_427_20 10. Nabity-Grover, T., Cheung, C.M., Thatcher, J.B.: Inside out and outside in: How the COVID19 pandemic affects self-disclosure on social media. Int. J. Inform. Manag. 55 (2020). https:// doi.org/10.1016/j.ijinfomgt.2020.102188 11. Wiederhold, B.K.: Using social media to our advantage: alleviating anxiety during a pandemic. Cyberpsychol. Behav. Soc. Netw. 23, 197–198 (2020) 12. Salloum, S.A., Al-Emran, M., Shaalan, K.: Mining text in news channels: a case study from Facebook. Int. J. Inform. Technol. Lang. Stud. 1(1), 1–9 (2017) 13. Statsica.com: Global social networks ranked by number of users (2022). https://www.statista. com/statistics/272014/global-social-networks-ranked-by-number-of-users/ 14. Statsica.com: Share of Instagram users in Malaysia as of March 2021, by gender (2022). https://www.statista.com/statistics/997071/share-of-instagram-users-by-gender-malaysia/ 15. Neuendorf, K.A., Kumar, A.: Content analysis. Int. Encycl. Polit. Commun. 1–10 (2016). https://doi.org/10.1002/9781118541555.wbiepc065 16. Aue, A., Gamon, M.: Customizing sentiment classifiers to new domains: a case study. In Proceedings of Recent Advances in Natural Language Processing (RANLP), vol. 1, no. 3.1, pp. 1–2 (2005)

What People Post During the Movement Control Order (MCO)

93

17. Wang, R., Kim, J., Xiao, A., Jung, Y.J.: Networked narratives on humans of New York: a content analysis of social media engagement on Facebook. Comput. Hum. Behav. 66, 149–153 (2017). https://doi.org/10.1016/j.chb.2016.09.042 18. Huang, F., Zhang, X., Zhao, Z., Xu, J., Li, Z.: Image–text sentiment analysis via deep multimodal attentive fusion. Knowl.-Based Syst. 167, 26–37 (2019). https://doi.org/10.1016/j.kno sys.2019.01.019 19. Dorsch, I., Zimmer, F., Stock, W.G.: Image indexing through hashtags in Instagram. Proc. Assoc. Inform. Sci. Technol. 54(1), 658–659 (2017). https://doi.org/10.1002/pra2.2017.145 05401105 20. Hu, Y., Manikonda, L., Kambhampati, S.: What We Instagram: a first analysis of Instagram photo content and user types. ICWSM (2014) 21. Mohamad, S.M.: Creative production of ‘COVID-19 Social Distancing’ narratives on social media. Tijdschr. Econ. Soc. Geogr. 111(3), 347–359 (2020) 22. Zhan, M., Tu, R., Yu, Q.: Understanding readers: conducting sentiment analysis of Instagram caption. In: Proceedings of the 2018 2nd International Conference on Computer Science and Artificial Intelligence - CSAI ’18, pp. 33–40 (2018). https://doi.org/10.1145/3297156.329 7270 23. Neuendorf, K.A.: The Content Analysis Guidebook. Sage, Thousand Oaks (California) (2002) 24. Manovich, L.: Instagram and Contemporary Image. manovich.net, New York (2016) 25. Cheah, J.-H., Ting, H., Cham, T.H., Memon, M.A.: The effect of selfie promotion and celebrity endorsed advertisement on decision-making processes: a model comparison. Internet Res. 29(3), 552–577 (2019) 26. Azlan, A.A., Hamzah, M.R., Sern, T.J., Ayub, S.H., Mohamad, E.: Public knowledge, attitudes and practices towards COVID-19: a cross-sectional study in Malaysia. PLOS One 15(5) (2020). https://doi.org/10.1371/journal.pone.0233668 27. Noris, A., Cantoni, L.: COVID-19 outbreak and fashion communication strategies on Instagram: a content analysis. In: Soares, M.M., Rosenzweig, E., Marcus, A. (eds.) HCII 2021. LNCS, vol. 12781, pp. 340–355. Springer, Cham (2021). https://doi.org/10.1007/978-3-03078227-6_25 28. Rother, R., Zakaria, H., Gani, F.A.: The Malaysian perspective of the COVID-19 outbreak from a social psychiatric lens. World Soc. Psychiat. 2(2), 139–141 (2020) 29. Reis, H.T., Maniaci, M.R., Caprariello, P.A., Eastwick, P.W., Finkel, E.J.: Familiarity does indeed promote attraction in live interaction. J. Pers. Soc. Psychol. 101(3), 557–570 (2011) 30. Waterloo, S.F., Baumgartner, S.E., Peter, J., Valkenburg, P.M.: Norms of online expressions of emotion: Comparing Facebook, Twitter, Instagram, and WhatsApp. New Media Soc. 20(5), 1813–1831 (2018) 31. Cham, T.H., Easvaralingam, Y.: Service quality, image and loyalty towards Malaysian hotels. Int. J. Serv. Econ. Manag. 4(4), 267–281 (2012) 32. Cham, T.H., Cheah, J.H., Cheng, B.L., Lim, X.J.: I Am too old for this! Barriers contributing to the non-adoption of mobile payment. Int. J. Bank Market. (2021). https://doi.org/10.1108/ IJBM-06-2021-0283 33. Tan, J.X., Cham, T.H., Zawawi, D., Aziz, Y.A.: Antecedents of organizational citizenship behavior and the mediating effect of organization commitment in the hotel industry. Asian J. Bus. Res. 9(2), 121–139 (2019) 34. Lacap, J.P.G., Cham, T.H., Lim, X.J.: The influence of corporate social responsibility on brand loyalty and the mediating effects of brand satisfaction and perceived quality. Int. J. Econ. Manag. 15(1), 69–87 (2021) 35. Fam, K.S., Cheng, B.L., Cham, T.H., Tan, C.Y.M., Ting, H.: The role of cultural differences in customer retention: evidence from the high-contact service industry. J. Hosp. Tour. Res. 10963480211014944 (2021)

94

H. L. Goh et al.

36. Cheng, B.L., Shaheen, M., Cham, T.H., Dent, M.M., Yacob, Y.: Building sustainable relationships: service innovation at the pinnacle of touristic achievement. Asian J. Bus. Res. 11(1), 80–90 (2021) 37. Cheing, A., Hong, E.H., Kuek, T.Y., Chai, B.H.B., Cham, T.H.: Social media effectiveness indicators of microenterprise strategic planning. Asian J. Bus. Res. 10(1), 150–166 (2020)

A Comparative Study on the Recognition of English and Arabic Handwritten Digits Based on the Combination of Transfer Learning and Classifier Bashar Al-Saffar1,2(B) , Amjed R. Al-Abbas1 , and Selma Ay¸se Özel3 1 Department of Computer Techniques Engineering, Al Salam University College, Baghdad,

Iraq [email protected] 2 Department of Electrical and Electronic Engineering, Çukurova University, Adana, Turkey 3 Department of Computer Engineering, Çukurova University, Adana, Turkey

Abstract. In recent days, recognizing handwritten digits in Arabic and English has been useful for several applications. This paper presents an efficient method to recognize the unlimited variation in human handwriting. We have used freely available datasets, MNIST and MADBase, for English and Arabic handwritten digits, respectively. Each dataset involves enough number of images with ten classes from 0 to 9, so that there are 70,000 images in total, 60,000 images are used for training and 10,000 images are used for testing the models. A Deep Learning-based methodology is suggested for recognizing handwritten digits by using various transfer learning types such as; AlexNet, ResNet-18, GoogleNet, and DensNet-201 aimed at deep feature extractions. Moreover, we utilized three types of classifiers: Decision Tree (DT), k-nearest neighbors (KNN), and Support Vector Machine (SVM) and compared their performances. The results show that the AlexNet features with SVM classifiers provide the best results for both datasets, with error rates of 0.96% and 0.9997% for Arabic and English datasets, respectively. Keywords: MNSIT digits handwritten recognition · SVM · CNN · DT · KNN · DL

1 Introduction Handwritten Digits Recognition (HWR) has been an important area due to its applications in several fields [1]. Recognition is an area that covers various fields such as face recognition, image recognition, fingerprint recognition, character recognition, and numerals recognition [2]. Handwriting recognition means the ability of a computer or device to take as input handwriting from a source such as printed physical documents, pictures, and other devices. It also uses handwriting as a direct input to a touch screen and then interprets it as text, e.g., smartphones, tablets, and PDA to a touch screen through a finger [3]. This is convenient since it enables the user to swiftly enter numbers and © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 95–107, 2023. https://doi.org/10.1007/978-3-031-20429-6_10

96

B. Al-Saffar et al.

text into the devices [3]. Handwritten digit recognition is an important component in many applications: check verification, office automation, business, postal address reading, printed postal codes, and data entry applications [2]. Deep learning (DL) [4] is a hierarchical structure network that simulates the human brain’s structure to extract the internal and external input data features. Deep learning is based on algorithms using multilayer networks such as deep neural networks, convolutional deep neural networks, deep belief networks, recurrent neural networks, and stacked auto encoders. These algorithms allow computers and machines to model our world well to exhibit intelligence [5]. In this paper, the Deep Learning based methodology is proposed for recognizing Arabic and English handwritten digits. In order to recognize the variation in the human’s handwriting, different types of transfer learning techniques that are AlexNet, ResNet-18, GoogleNet, and DensNet201 are used. On the other hand, the freely available datasets, MNIST and MADBase, are used for Arabic and English, respectively. The main purpose of this study is to make a comparative evaluation of deep learning and machine learning techniques for handwritten digit recognition for Arabic and English. The rest of this paper is organized as follows: In Sect. 2, related work on handwritten digit recognition is summarized. Section 3 describes the datasets used in the study. Section 4 explains the methods applied in this study for handwritten digit recognition. Section 5 briefly explains implementation platforms. Experimental results and their evaluations are presented in Sect. 6. Finally, Sect. 7 concludes our study.

2 Related Work Handwritten digit recognition is a difficult task that has been extensively researched for many years. The study in [6] have proposed a new model which is a hybrid of Convolutional Neural Network (CNN) and Support Vector Machine (SVM) using the MNIST digit datasets. This model uses CNN as an automatic feature extractor and applies SVM as the output predictor. The suggested model’s efficiency and practicality were assessed on two levels: recognition accuracy and reliability performance. To improve the handwritten digit recognition, the last two layers of the AlexNet structure are replaced with an SVM classifier. In the new method, the SVM acting as the output predictor and the AlexNet working as a feature extractor automatically, the stochastic diagonal LevenbergMarqurdt algorithm is introduced to accelerate the network’s convergence speed. The results based on the MNIST datasets showed that the proposed method could outperform both SVMs and AlexNet. Moreover, the method gets a faster convergence speed (a round 25–30 epochs) in the training process. The best results in this model have been achieved 94.4% recognition rate without rejection in [7]. In [8] presents a novel perceptual shape decomposition strategy to handwritten digit recognition. Where a new handwritten digit recognition technique that works very correspondingly to human perception, has been proposed. The key feature of the proposed method is the representation of deformed digits using four distinct visual primitives. For geometric modeling and efficient recognition, the proposed approach only requires a few shape parts and their obvious spatial relationships. The proposed system’s performance is evaluated on five digit datasets of four popular scripts, Odia, Bangla, Arabic, and English, with the author’s MNIST dataset achieving 99.11% accuracy for English digit recognition and 97.96% accuracy for Arabic digit recognition.

A Comparative Study on the Recognition of English

97

The Arabic digits recognition described by [9], a large amount of handwritten digits dataset had been collected to examine and demonstrate a powerful Deep Convolution Neural Network (DCNN) utilized for classification. Moreover, A vital CNN parameter is carefully chosen and modified to produce the final concrete model. The obtained dataset is trained using an efficient CNN model, and the model is thoroughly analyzed by carefully picking their parameters and demonstrating their resilience for handling the collected dataset. The detection performance reached 95.7%. A unique CNN deep learning architecture has been presented in [10] for the recognition of handwritten Multilanguage digits (mixed numbers belonging to various languages). It was created for manually writing Eastern Arabic and Persian numerals, while studies are also done with other languages like Urdu, Devanagari, and Western Arabic. The results of the experiment of this approach provide superior accuracy for datasets of separate languages as well as datasets of combined languages with the same geometrical features or with combined geometrical features, respectively. With use of CNN one layer the accuracies were 99.21% and 99.13% for Eastern Arabic and Western Arabic (English) respectively. A unique Local Feature Extraction method is also proposed in [11] for designing an unifying multi-language handwritten numeral recognition system using various languages (namely Arabic Western, Arabic Eastern, Persian, Urdu, Devanagari, and Bangla) with different number of digits. Using an RF classifier, the proposed technique is evaluated on six different well-known datasets of various languages, the Eastern Arabic digits recognition accuracy was 98.1%, for Western Arabic (English) accuracy reached to 95.3%, and the average accuracy of all used languages were 96.73%. In the study given by [12], a Decision Tree (DT) classifier approach is used to recognize handwritten English digits from the standard Kaggle digits dataset. To identify handwritten numerals, they assessed the model’s precision against each digit from 0 to 9. With 720 columns and 42000 rows in the Kaggle dataset used for training and testing, the method’s accuracy was 83.4%. The use of ResNet on a standard ISI Kolkata Handwritten Oriya numeral datasetse on 4970 handwritten samples was used to evaluate a deep learning strategy for Odia numeral digit recognition that was introduced in [13]. According to the statistics, ResNet achieves a true recognition rate of 99.20%, which is the highest accuracy for Odia numeral digit recognition when compared to other approaches. To avoid the complicated ensemble (classifier combination) expensive feature extraction, and sophisticated pre-processing, Convolutional neural network approaches to classic recognition systems have been evaluated in [14], and has shown the importance of several hyper-parameters. The error rate of the algorithm reached 0.11% with the Adam optimizer for the MNIST dataset. In our study, we utilized MNIST and MADBase datasets to experimentally evaluate the performance of hybrid of transfer learning with SVM, KNN and DT classifiers on recognizing English only, Arabic only, and English–Arabic mixed handwritten digits recognition.

98

B. Al-Saffar et al.

3 Data-Set This paper has utilized freely available datasets: MNIST and MADBase for English and Arabic handwritten digits, respectively [15, 16]. Each dataset contains 70,000 images in 10 classes, from which 10,000 images are used for testing and the remaining 60,000 images are used for training the models. The images in the dataset have low resolution, with size 28 × 28 × 1 pixels in grayscale. An example from each dataset is presented in Fig. 1.

Fig. 1. Examples from MNIST (for English) and MADbase (for Arabic) datasets.

4 Method Figure 2 displays the workflow of the proposed method. At first, CNN and Transfer Learning is used to extract features from the images, then Decision Tree (DT), k-nearest neighbors (KNN), and Support Vector Machine (SVM) classifiers are utilized to classify handwritten digit images into 10 classes. 4.1 Dataset Usage In this paper, we did four ways of handwritten digit identification as described in the followings: • English (10 classes): The classification is done for only English digits, and each digit has one class label. We have 10 classes from 0 to 9. • Arabic (10 classes): The classification is done only for Arabic digits, and each digit has one class label. We have 10 classes from 0 to 9. • English and Arabic (10 classes): We classify the combination of English and Arabic digits. Each digit represents one class in English and Arabic digits. We mixed each English digit with an equal Arabic digit to be all zeros in one file (class), and other digits have the same procedure. Finally, the number of classes will be 10 from 0 to 9. • English and Arabic (20 classes): The classification of English and Arabic digits are done together, but each digit is classified as English or Arabic separately. That means the zero digits in Arabic have a different class label than zero in English; as we described, we have 10 Arabic digits classes, and 10 English digits classes, totally we have 20 classes.

A Comparative Study on the Recognition of English

99

Fig. 2. The workflow of the proposed method.

4.2 The Stages of the Workflow 4.2.1 Pre-processing In order to improve the accuracy of digits recognition, the solution to the digits identification problem requires beating some main problems, like differing image size and quality of image, different levels of illumination, and employing a high number of images. Therefore, it is important to use pre-processing stages of the images before processing them. This paper’s pre-processing stage involve Histogram Equalization (HE), image resize techniques. • Histogram Equalization (HE): For the illumination normalization, we used the Histogram Equalization (HE) technique, which reduces the light effect and luminosity of unity of all images in the dataset. This stage will positively affect the performance of the proposed model. Histogram Equalization (HE) [17] is a fast, easy, and effective image enhancement that can effectively confirm the image density information in all regions. • Image Resizing: The size of the cropped images is a very important step in making the test image sizes the same as the size of train images, and the more important thing is to be the same size as the input layer of the used CNN. We have used image sizes in this experiment as 28 × 28 × 1 pixels, however we resized the images according to input size of the CNN architectures for example the input layer of AlexNet was 227 × 227 × 3 and in DensNet-201, ResNet-18, and GoogleNet the input sizes were 224 × 224 × 3. 4.2.2 Convolution Neural Network (CNN) When dealing with huge datasets, it is very difficult for a human, to determine which label or class a data point belongs to. Instead, a classifier, which is a supervised machine

100

B. Al-Saffar et al.

learning method, is used. A classifier is trained from a dataset with known class labels, then it is used on data of the same type and feature length to assign class label to new test data. To train a classifier, features from the data needs to be extracted and CNN is used for feature extraction purpose in this study. Multiple abstract data representations can be learned by multi-layered computational models using Deep Learning (DL). From training to testing, the success of deep artificial neural networks varies. Conventional learning assumes that the model family’s properties or the regularization techniques used in training have minor generalization errors. Convolutional neural networks (CNN) are a type of DL [18, 19]. Despite the fact that neural networks were not popular at the time, they have achieved a great deal in practice, and the computer vision community has recently embraced them. CNN are powerful visual models that generate hierarchies of features. Full color RGB image composed of three two-dimensional arrays including pixel intensities in the three color channels, for example, is intended for the processing of data in the form of multiple arrays. In this study, handwritten digit images are given as input to a CNN model, then it produces the feature vector for each image where the length of the feature vector depends on the CNNs model and feature layer. These traits are entered into a classifier to identify handwritten digits. 4.2.3 Classifiers In the current study, we utilized three classifiers: Decision Tree, KNN, and Support Vector Machine (SVM) to check the performance, and the best classifier will be used to identify the handwritten digits. • Support Vector Machine (SVM): is a supervised machine learning algorithm that finds some soft hyperplanes that separate each class or group by employing a training algorithm [20]. • Decision Tree (DT): is a classification algorithm commonly utilized in data mining. The objective is to create a model that detects the label based on the input variables. In each node or stage of a DT classifier, we try to form a pattern condition that separates all classes involved in the dataset to the most accurately [21]. • k-Nearest-Neighbors (K-NN): Is a simple supervised machine learning algorithm can be used for classification and regression, It is easy to understand and implement, but the main hurdle is that it slows down significantly with the size of the huge data. In k-NN classification, an object is classified by majority votes of its k nearest neighbors (k is an odd positive integer, usually small). If k is equal to one, the object is simply assigned to the class of that single nearest neighbor [22]. We design the architecture of the hybrid of various types of transfer learning models with various types of classifiers such as SVM, DT, and KNN, by replacing the fully connected layer (FC) of the CNNs model with a classifier. Several different classifiers and transfer learning models have been applied because of the No Free Lunch (NFL) theorem which means that there is no unique machine learning algorithm that performs best for all problems [23]. If an algorithm does particularly well on one class of problems,

A Comparative Study on the Recognition of English

101

then it is highly probable that it does more poorly over the remaining problems. That’s why we need to test several different classification and transfer learning models to see their performances. For example in the AlexNet model, we have three fully connected layers as layers 6–8th in each layer we have different features and lengths as displayed in Table 1. In our study, we combine the deep features with a classifier and will test each transfer learning model with different fully connected layer and classifier in order to select the best performance. The output units of the fully connected layer in a CNN’s network are the estimated probabilities of the input samples. An activation function computes each output probability. The activation function’s input is that the output of the previously hidden layer has trainable weights plus a bias term. It is pointless to look at the output values of the hidden layers just for the CNN network itself; however, these values can be considered as features of any other classifier. In this paper we presented four type of transfer learning namely AlexNet, DensNet 201, ResNet18, and GoogleNet combined with various type of classifiers such as SVM, DT, and KNN.

5 Implementations All the requirements in this study were implemented using Transfer Learning, image processing, and classification applications in MATLAB 2018-b and a computer equipped with an Intel (R) Core™-i7-7500U (2.9 GHz) processor. 12 GB of RAM, which helped us get good performance and good processing times.

6 Results and Discussion The experiment tested the handwritten digits recognition in many ways, such as the outcome obtained by the MNIST dataset for English digits only. In this case, we have 10 classes, one class for each digit. MADbase dataset contains Arabic digits only. In this case, we classify data into 10 Arabic digits classes. After that both datasets are mixed and all the images with related classes for Arabic and English are merged. As an example, all Arabic zero (0) images and English zero images are combined to form zero class images. Therefore we have 10 classes from 0 to 9. Finally, we classify the digits separately with respect to either being Arabic or English. Again, we combine the two datasets without combining the classes. Each class has digits either in Arabic or English (here, we have two classes of each digit, one for Arabic and the other for English). We have a total of 20 classes, 10 for Arabic and 10 for English digits. We have utilized Transfer Learning for feature extraction and various classifiers such as DT, KNN, and SVM and compared their performances. The experimental results are displayed in Tables 1, 2, 3 and 4. As well as the examples of confusion matrix obtained by a combination of Alex-Net and SVM are displayed in Figs. 3, 4, 5, 6 and 7. The results show that the combination of AlexNet-6th fully connected layer (‘Fc6’) with the SVM classifier provides best results with less than 1% error in both datasets. The hybrid of CNN with a Decision Tree (DT) as well as CNN with KNN classifiers also has been done in our study, however the accuracy was lower than SVM classifier;

102

B. Al-Saffar et al.

DT was better than KNN. About running time-consumption, the KNN was faster than SVM and DT classifiers; for example in order to test 10000 samples SVM takes 2349 s, DT takes 14.4 s, and KNN takes 1.38 s. According to the results, the hybrid of CNN with KNN was faster than other classifier but lower accuracy. The Transfer Learning (AlexNet-fc6) with SVM classifier shows the best results in all cases (Tables 1, 2, 3 and 4). The accuracy reached more than 99% in Arabic, and English datasets shows the proposed method is successful and can be used in many applications to recognize the digits. Table 1. The classification results of the handwritten English digits by using various types of CNN and classifiers. DL name

Feature layer

Size of input layer

Total layer

Feature length

Accuracy of classifiers (%) SVM

DT

KNN

AlexNet

fc6

227 × 25 227 × 3

4096

99.0003

98.393

98.002

AlexNet

fc7

227 × 25 227 × 3

4096

98.7785

97.98

95.68

AlexNet

fc8

227 × 25 227 × 3

1000

97.7279

96.9476

93.84

DensNet201

fc1000

224 × 709 224 × 3

1000

96.375

95.8159

93.85

ResNet18

fc1000

224 × 72 224 × 3

1000

95.9125

94.8789

86.13

GoogleNet

loss3-classifier

224 × 144 224 × 3

1000

95.095

92.93

84.423

Figure 5 shows that some numbers in the two languages have high similarity, such as the numbers one and nine in English are similar to the numbers one and nine in Arabic, which can improve detection performance, especially when using handwriting digits. And some different numbers have similar properties, for example, the number five in Arabic has a high similarity to the number zero in English language, and they have the same properties as displayed in Fig. 1, which can reduce the classification performance. As shown in Fig. 5. The maximum error was in numbers 0 and 5. The overall performance (accuracy) of AlexNet-fc6 with SVM was 98.23%. In this method, the handwritten numbers were classified on the basis of the different numbers for each language, where the English numbers were separated from the Arabic numbers, and each dataset contains 10 classes, and the classification will be done for 20 classes in total. As shown in Fig. 6, it was noted that the error rate occurred in the numbers with similar features in both languages, such as the numbers 1, 9, 0, and 5 as shown in Fig. 1, where the error rate was the largest, as it was noted that the error rate in the mentioned numbers is about 2%. The overall performance reached to 98.56%

A Comparative Study on the Recognition of English

103

Fig. 3. Confusion matrix of handwritten English digits only (10 classes) by using AlexNet-fc6 with SVM. Table 2. The classification results of the handwritten Arabic digits by using various types of CNN and classifiers. DL name

Feature layer

Size of input layer

Total layer

Feature length

Accuracy of classifiers (%) SVM

DT

KNN

AlexNet

fc6

227 × 227 × 3

25

4096

99.04

98.93

98.66

AlexNet

fc7

227 × 227 × 3

25

4096

99.02

98.66

97.72

AlexNet

fc8

227 × 227 × 3

25

1000

98.72

98.48

96.84

DensNet201

fc1000

224 × 224 × 3

709

1000

98.25

98.16

95.7

ResNet18

fc1000

224 × 224 × 3

72

1000

97.84

97.43

95.02

GoogleNet

loss3-classifier

224 × 224 × 3

144

1000

97.59

97.32

94.82

by AlexNet-fc6 with SVM, it is better than the mixing of both dataset. In the MNIST dataset the number of images was unbalanced but in average was 10000 test images in each class. According to experimental evaluation, we have observed that, for all 4 types of dataset usage, the best handwritten digit recognition is performed by the combination of the AlexNet-fc6 with SVM classifier.

104

B. Al-Saffar et al.

Fig. 4. Confusion matrix of handwritten Arabic digits only (10 classes) by using AlexNet-fc6 with SVM. Table 3. The classification results of the combined handwritten English and Arabic digits (10 classes) by using AlexNet and GoogleNet as automatic feature extraction and SVM, DT, and KNN as classifiers. DL name

Feature layer

Size of input layer

Total layer

Feature length

Accuracy of classifiers (%) SVM

DT

KNN

AlexNet

fc6

227 × 227 × 3

25

4096

98.23

97.43

96.87

GoogleNet

loss3-classifier

224 × 224 × 3

144

1000

94.39

89.516

86.57

Results show that the proposed method gives low error rate and high accuracy of the handwritten digits recognition as compared to previous studies. When we compare our results with the previous studies we observed that, we improved the results of the study presented in [12] by 15.6%, our results are 0.94% better than that of results in [11], for Arabic digits we improved results of [9] by about 3.34%, and again for Arabic digits we improved results of [8] by 1.08%. Our results are better than many previously reported accuracies.

7 Conclusion The supervised machine learning algorithm in the application field is so vast that it depends on the theorem (no free lunch) [23], so we cannot choose a unique machine learning approach that can give us a good performance in all fields. Moreover, the accuracy of a machine learning algorithm is based on the input features and the required

A Comparative Study on the Recognition of English

105

Fig. 5. Confusion matrix of the combination of English and Arabic digits handwritten (10 classes) by using AlexNet-fc6 with SVM.

Table 4. The classification results of the handwritten English and Arabic digits (20 classes) only by AlexNet and GoogleNet as automatic feature extraction and SVM, DT, and KNN as classifiers. DL name

Feature layer

Size of input layer

Total layer

Feature length

Accuracy of classifiers (%) SVM

DT

KNN

AlexNet

fc6

227 × 227 × 3

25

4096

98.56

97.25

96.14

GoogleNet

loss3-classifier

224 × 224 × 3

144

1000

94.86

90.63

84.74

class name in the output. Hence, various machine learning algorithms should be checked to know which method will provide us with better accuracy. Therefore, we have implemented, tested, and trained three different classifiers accuracies in this project: DT, KNN, and SVM. To calculate the accuracy of each model and select the best one, as well as four CNN architectures (AlexNet, GoogleNet, DensNet201, and ResNet18) were used to features extraction. The difference in the performance of CNNs model came from the design of CNNs layers and size of convolution window etc. The SVM classifiers introduced the best results in all used CNNs, the DT classifier was the second best, and KNN performed worst. However, DT run faster than SVM, while KNN required the longest classification time as it is lazy method. The results show that the deep feature obtained by (AlexNet) from 6th fully connected layer with features length 4096 values, combined with the SVM classifier shows the best results in all cases. The error reached is less than 1% in Arabic and English Datasets which shows the proposed method is successful and can be used in many applications

106

B. Al-Saffar et al.

Fig. 6. Confusion matrix of English and Arabic digits handwritten separately in each (20 classes) by using AlexNet-fc6 with SVM.

to recognize the digits. Our future work will be highlighted improving the accuracy of handwriting Arabic and English digit recognition by utilizing other classifiers, and changing the parameters as well as an improved CNNs model.

References 1. El-Sawy, A., El-Bakry, H., Loey, M.: CNN for handwritten Arabic digits recognition based on LeNet-5. In: International Conference on Advanced Intelligent Systems and Informatics, pp. 566–575. Springer, Cham (2016) 2. Loey, M., El-Sawy, A., El-Bakry, H.: Deep learning autoencoder approach for handwritten Arabic digits recognition (2017). arXiv preprint arXiv:1706.06720 3. Hamid, N.A., Sjarif, N.N.: Handwritten recognition using SVM, KNN and neural network (2017). arXiv preprint arXiv:1702.00723 4. Rodrigues, I.R., da Silva Neto, S.R., Kelner, J., Sadok, D., Endo, P.T.: Convolutional extreme learning machines: a systematic review. InInformatics (MDPI) 2(8), 33 (2021) 5. Rawat, W., Wang, Z.: Deep convolutional neural networks for image classification: a comprehensive review. Neural Comput. 9(29), 2352–2449 (2017) 6. Niu, X.X., Suen, C.Y.: A novel hybrid CNN–SVM classifier for recognizing handwritten digits. Pattern Recogn. 45(4), 1318–1325 (2012) 7. Yu, N., Jiao, P., Zheng, Y.: Handwritten digits recognition base on improved LeNet5. In: The 27th Chinese control and decision conference, CCDC, pp. 4871–4875. IEEE (2015) 8. Dash, K.S., Puhan, N.B., Panda, G.: Unconstrained handwritten digit recognition using perceptual shape primitives. Pattern Anal. Appl. 21(2), 413–436 (2016). https://doi.org/10.1007/ s10044-016-0586-3 9. Alwzwazy, H.A., Albehadili, H.M., Alwan, Y.S., Islam, N.E.: Handwritten digit recognition using convolutional neural networks. Int. J. Innov. Res. Comput. Commun. Eng. 4(2), 1101– 1106 (2016) 10. Latif, G., Alghazo, J., Alzubaidi, L., Naseer, M.M., Alghazo, Y.: Deep convolutional neural network for recognition of unified multi-language handwritten numerals. In: 2nd International

A Comparative Study on the Recognition of English

11. 12. 13.

14. 15. 16. 17. 18. 19.

20. 21. 22.

23.

107

workshop on Arabic and derived script analysis and recognition (ASAR), pp. 90–95. IEEE (2018) Alghazo, J.M., Latif, G., Alzubaidi, L., Elhassan, A.: Multi-language handwritten digits recognition based on novel structural features. J. Imaging Sci. Technol. 63, 1–10 (2019) Assegie, T.A., Nair, P.S.: Handwritten digits recognition with decision tree classification: a machine learning approach. Int. J. Electr. Comput. Eng. (IJECE) 9(5), 446–4451 (2019) MrinmoySen, S.B., Ray, P., Sasmal, M., Mukherjee, R.: Handwritten Odia digits recognition using residual neural network. Turk. J. Comput. Math. Educ. (TURCOMAT) 11(1), 567–574 (2020) Ahlawat, S., Choudhary, A., Nayyar, A., Singh, S., Yoon, B.: Improved handwritten digit recognition using convolutional neural networks (CNN). Sensors 20(12), 3344 (2020) The English Handwritten Digits Dataset (MINST). http://yann.lecun.com/exdb/mnist/. Accessed 21 Mar 2022 The Arabic Handwritten Digits Dataset (MADBase). https://datacenter.aucegypt.edu/sha zeem. Accessed 16 May 2022 Dey, E.K., Khan, M., Ali, M.H.: Computer vision based gender detection from facial image. LAP LAMBERT Academic Publishing (2013) Rakotomamonjy, A.: Applying alternating direction method of multipliers for constrained dictionary learning. Neurocomputing 106, 126–136 (2013) Miikkulainen, R., Liang, J., Meyerson, E., Rawal, A., Fink, D., Francon, O., Raju, B., Shahrzad, H., Navruzyan, A., Duffy, N., Hodjat, B.: Evolving deep neural networks. In: Artificial Intelligence in the Age of Neural Networks and Brain Computing, pp. 293–312. Academic Press (2019) Rong, Y., et al.: Post hoc support vector machine learning for impedimetric biosensors based on weak protein–ligand interactions. Analyst 143(9), 2066–2075 (2018) Vens, C., Struyf, J., Schietgat, L., Džeroski, S., Blockeel, H.: Decision trees for hierarchical multi-label classification. Mach. Learn. 73(2), 185–214 (2008) Guo, G., Wang, H., Bell, D., Bi, Y., Greer, K.: KNN Model-based approach in classification. In: Meersman, R., Tari, Z., Schmidt, D.C. (eds.) OTM 2003. LNCS, vol. 2888, pp. 986–996. Springer, Heidelberg (2003). https://doi.org/10.1007/978-3-540-39964-3_62 Wolpert, D.H., Macready, W.G.: No free lunch theorems for optimization. IEEE Trans. Evol. Comput. 1(1), 67–82 (1997)

Predictive Analytics for Oil and Gas Asset Maintenance Using XGBoost Algorithm Norshakirah Aziz(B) , Mohd Hafizul Afifi Abdullah , Nurul Aida Osman , Muhamad Nabil Musa, and Emelia Akashah Patah Akhir Computer and Information Sciences Department, Universiti Teknologi PETRONAS, 32610 Seri Iskandar, Perak, Malaysia {norshakirah.aziz,nurulaida.osman,muhamad.nabil_24608, emelia.akhir}@utp.edu.my

Abstract. One of the most important aspects of the oil and gas industry is asset management at their respective platforms. Without proper asset management, it will lead to various unexpected scenarios including an increase in plant deterioration, increased chances of accidents and injuries, and breakdown of assets at unexpected times which will lead to poor and hurried maintenance. Given the significant economic contribution of the oil and gas sector to oil-producing countries like Malaysia, accurate asset maintenance prediction is essential to ensure that the oil and gas platform can manage its operations profitably. This research identifies the parameters affecting the asset failure on oil and platform that will be interpreted using the XGBoost gradient boosting model from machine learning libraries. The model is used to predict the asset’s lifetime based on readings collected from the sensors of each machine. From result, our prediction method using XGBoost for asset maintenance has presented a 6.43% increase in classification accuracy as compared to the Random Forest algorithm. Keywords: Predictive analytics · XGBoost · Random forest · Data analytics

1 Introduction Malaysia is one of the leading oil and gas production companies in the Asia-Pacific region with an estimated oil barrels output of more than e1.7 million each day in 2018. The total commercial reserves in the country are valued at more than 5 billion barrels of oil equivalent found in more than 400 fields, with a gas comprising around threefourths of the mix. This describes the important contribution the industry has made to the Malaysian economy, contributing up to 20% to the gross domestic product (GDP) and being the second-largest export after electric and electronic products in recent years [1]. Thus, asset management needs to be introduced in the industry to ensure greater significance in enhancing operating efficiencies [2]. Asset management will provide a systematic approach to deliver cost-effectiveness while reducing the environmental effects of the activities involved in the oil and gas industry. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 108–117, 2023. https://doi.org/10.1007/978-3-031-20429-6_11

Predictive Analytics for Oil and Gas Asset Maintenance

109

The recent emergence of sensors for data recording in exploration, drilling, and production activities has made the oil and gas industry to become more data-centric [3]. The data collected from these sensors are considered big data as they fulfill its main characteristics which include high volume, variety, velocity, unknown veracity, and valuable. Data collected from sensors coming from the oil and gas platform could be used for predictive maintenance to anticipate issues that are associated to plant maintenance. Asset management consists of activities and practices that monitor the assets of a company and use them to gain value effectively. In general, the operation of oil and gas industry involves heavily invested assets as depicted in Fig. 1.

Fig. 1. Assets involved in in the operations of oil and gas industry.

Businesses that are involved in the oil and gas niche need to enhance their performance and efficiency through excellent assets management techniques to level with the high cost of assets required in oil and gas production [3]. This includes asset lifetime monitoring to ensure measures are taken before the asset fails. However, difficulty in monitoring assets has caused the operational cost to increase by a huge margin. When assets on platforms are not effectively overseen and observed, the likelihood of failure increases and the assets will eventually come to a halt, stopping all operations and resulting in significant financial losses. Many works related to the efficiency of the XGBoost algorithm in terms of prediction [4–7], yet there is no one-for-all model suitable to be used for predicting oil and gas asset maintenance purposes. It is virtually impossible to estimate the lifetime of oil and gas reserves with 100% precision [8]. Consequently, predictive analytical approaches must always boost the accuracy of data sets. Therefore, this study aims to utilize sensors data using predictive analytics method in determining time to fail for oil and gas machines.

2 Literature Review The oil and gas industry is an asset-intensive sector [9], spanning from pipelines and LNG equipment in the mid-to downstream sectors to offshore platforms to upstream pipelines. Such assets are complex and costly. Due to massive investments in reserves, oil and gas firms are continually under pressure to better handle them [10]. It entails the use of various management methods to minimize costs and improve asset performance. Asset management and control in the modern plant processes include a wide range of tasks with varying timeframes and complexities, including the reconciliation and melting of information, fault detection, isolation, and accommodation. Many research studies have proposed various combinations of theoretical systems and artificial intelligence

110

N. Aziz et al.

techniques [11–13] in addressing the question of asset management. Big data analysis applied in the industry to capture, process, and analyze data may improve an understanding of the irregular behavior of a production system. It helps to create the Creative-Based Maintenance (CBM) paradigm, in which decisions are made using massive, diversified, and dynamic data to reduce operational costs. 2.1 Big Data in the Oil and Gas Industry Big Data is a popular term that refers to large, fast, and highly complex data that is considered difficult to process using traditional learning methods. Big Data may have a complex structure, thus introducing challenge for storage, analysis, and visualization. However, studying these large amount of data helps us to reveal hidden patterns [14], which allows organizations to gain insights and market advantage. Big Data is useful for managing our society as the solution of a complex urban problem creates new problems [16]. However, the strategic solutions require a large amounts of data. Thus, Big Data can be used as there is enough correlation to fix the problems. Predictive analytics consider potential predictions based on near past observations or process observations [17]. The data collected for predictions use statistics integrated into mathematical programming and operational methods of research. With the rapid expansion of data, businesses strive to explore methods to develop predictive analytics, data mining, and expand their data and resources infrastructure. The challenges of managing data growth, integrating diversified business intelligence tools, and analyzing data to provide useful insights are met with these new strategies. Oil production has become more challenging to supply despite a firm growth of petroleum products [18]. There are now more critical issues due to the increase in data volume in the oil and gas industry, and analytics will be necessary to provide opportunities for enhanced decision-making [19]. Predictive analytics impacts the oil and gas industry’s digital adoption at three levels, where companies utilize the data they have, train data models, perform predictive analytics using the trained model, and transform the output into dashboards. Predictive analytics have been applied for forecasting crude oil prices [20]. Apache Corporation used predictive analysis to identify the root cause of a mechanical fault that affected 10,000 barrels per day, and they discovered 40 steps necessary to minimize unplanned downtime and costs [21]. 2.2 Predictive Analytics in Oil and Gas Industry Preventive maintenance is another area where quick decision-making will save huge amounts of money from oil and gas firms. Failure of equipment, especially when not expected, can be extremely costly [22]. Predictive maintenance is a popular technique to tackle maintenance problems, despite the increasing need to reduce downtime and related costs. This module needs to be typically adapted to the problem at hand, thereby supporting the existence of several different approaches to predictive maintenance in the literature [23]. In the oil and gas industry, many maintenance problems are investigated through predictive maintenance modules including detection of lost circulation in drilling wells [24], predicting failure rates of equipment in safety instrumented systems [25], and predicting dynamic fuel oil consumption on ships [25].

Predictive Analytics for Oil and Gas Asset Maintenance

111

Machine condition monitoring provides significant cost savings, protection, and reliability benefits to businesses by providing an early indicator of potential system failure during the process operating cycle. Thus, condition monitoring has received a lot of interest from businesses and researchers [26]. In [27], monitoring the condition of offshore infrastructure is a crucial component of the oil and gas industry. Monitoring work performance using modern technology unlocks the element of dependability and reliability of details through source information [28]. For many reasons, the most current generation of industrial monitoring technologies and creative techniques satisfy numerous requirements in the design of these industrial systems. This enables the execution of an efficient maintenance strategy for decision-making and allows the quick and precise identification of their shortcomings [29]. Various condition monitoring activities have already been implemented in the oil and gas industries. Monitoring of machines and equipment is important, as complexity and value increase. It generates many criteria which can be difficult for humans alone to navigate and understand. Analytical techniques can greatly enhance the value of the information generated by the monitoring devices. Intelligent structuring, screening, and measuring of these details will allow the management team to make effective machine-based decisions [30]. In the oil and gas industry, Big Data Analytics solutions through condition monitoring are proved necessary. There is an increase in the number of wells, sensors, processing units, and data parameters as well as the sophistication of certain technologies. The potential for creativity is to construct an effective Big Data approach for all these types, independent of the supplier or part, gather data, track, and manage the appropriate elements in the field [31]. 2.3 Machine Learning (ML) ML attempts to devise algorithms that develop through experience automatically. Nonetheless, ML techniques usually select the best-suited algorithm from a pool of algorithms based on defined performance criteria [32]. The oil and gas sector is challenging and complex as there are many processes and stakeholders involved, each producing an enormous amount of data. Given the global and distributed nature of the business, it is a tough task to process and manages this data. For many sectors in recent times, including the oil and gas industries, artificial intelligence, and ML have become very well-established to help people tackle these complex tasks [33]. To achieve an efficient maintenance strategy, the need for a cost-effective, reliable, and secure operation of the machinery requires accurate identification and classification of faults. Furthermore, fault prediction is critical in strategic sectors such as the oil and gas industries to extend the component lifetime and reduce unplanned equipment, thereby preventing cost-cutting and plant shutdowns [34]. There have been various efforts and success stories to forecast a probable time-to-failure so that maintenance solutions can be planned and deployed at the most appropriate moment [34]. Various issues encountered in the industry are solved through ML such as pipeline defect detection [35–37] drilling operations [38–40] and forecasting purposes [41–43]. Gradient boosting is a prediction and learning algorithm that incorporates the effects of several basic predictors to create an efficient committee with enhanced performance

112

N. Aziz et al.

over single members. It is well known for producing cutting-edge results in a few complicated data sets [44]. This algorithm’s randomization processes can greatly increase speed and accuracy as well as its ability to counteract basic learners’ adaption [45]. Gradient boosting tuning is more intuitive and flexible than optimizing a neural network. For example, an increased number of trees per one enables the careful model to change while increasing the number of neural network nodes per one, in small datasets, could lead to significant changes in model performance and possible overfitting. The gradient boosting value frequently includes a feature-related analysis, enabling a better understanding of algorithm output and its parameters. Gradient boosting [46] is widely used for prediction and classification purposes in the oil and gas industry which include hydraulic fracturing efficiency-boosting, oil production monitoring [47], and damage prediction purposes [48]. eXtreme Gradient Boosting (XGBoost) [49] is a popular algorithm for solving a variety of ML and data mining problems [50, 51]. It is considered a scalable tree-boosting ML framework and is available as an open-source package. The package is extensible to allow users to achieve their computational goals [52]. It features a tree learning approach and an effective linear model solver. The outputs are correlation, grouping, and sorting. XGBoost’s concept is to dynamically add a CART tree and break features to expand a tree. The fact that one tree is introduced at the time is a new function. It is a combination of the first and second derivatives, whilst the algorithm uses the tree model’s complexity as the standard term in its objective function to avoid overfitting [53]. In linear regression, the model is outfitted exponentially. Thus, XGBoost uses a regularized model to avoid over-fitting and simplifies the parallelization process [54]. The longest process is to sort the data in a gradient tree boosting framework—thus, XGBoost saves data in memory blocks or frames to lower the computational cost.

3 Methodology This study uses XGBoost classification and regression methods to predict the lifetime of oil and gas assets on offshore platforms for maintenance purposes. The experimental processes include (i) data gathering, (ii) data exploration and pre-processing, (iii) data splitting, (iv) data training, and (v) testing and evaluating the model. 3.1 Data Acquisition Simulation data from assets at the offshore A1 plant will be used for the tasks. From this data, two sets of data are created—one for a classification task and the other for a regression task. The data for the classification task contains a total of 220,321 rows and 9 columns with a class label, where the measurements from 6 sensors are set as columns. Meanwhile, the data for the regression task contains a total of 15,157 rows and 3 columns with a target value. Each sensor represents a piece of equipment in the A1 plant and the sensor data are recorded in an hourly manner, as shown in Fig. 2.

Predictive Analytics for Oil and Gas Asset Maintenance

113

Fig. 2. The snippet of the (a) A1 classification dataset and (b) A1 regression dataset

3.2 Data Exploration and Pre-processing During the data understanding phase, the data file is loaded and examined in a Python environment to ensure that each column’s data types match the supplied values. All Not a Number (NaN) values are replaced with an average value during this step. 3.3 Data Splitting A data split ratio of 70% for training and 30% for testing was chosen to create a robust XGBoost classification model that is trained from 70% of the entire dataset. The remaining 30% of the data will be used for validating the trained model. Figure 3 below shows the data split for classification and regression datasets.

Fig. 3. Test and train split strategy for the (a) classification dataset and (b) regression dataset.

3.4 Data Training The XGBoost classifier is used for executing the classification method to predict asset failure of the machines at oil and gas plants using the first dataset. Meanwhile, the second dataset utilizes XGBoost’s regressor to predict the asset failure of the machines at the oil and gas plant for the next day. The dataset is then fitted into each model using the supposed parameters to achieve the highest prediction accuracy. Table 1 shows the parameter settings for the XGBoost classifier and regressor model. 3.5 Data Testing and Evaluation The prediction was conducted using the XGBoost classifier and regressor models. The parameters of the gradient boosting model are examined and revised by training the model iteratively in order to achieve an optimum parameter, hence, improving the accuracy of the trained model. According to the findings, the models can reliably anticipate asset failure. The performance of the models is evaluated based on the conditions predicted on the six sensors (Normal = 1.0, Recovering = 0.5, and Broken = 0.0). Table 2 summarizes the result obtained from the data of equipment A1.

114

N. Aziz et al.

Table 1. Parameter settings for creation of the XGBoost’s classifier and regression model. Classifier parameters

Regressor parameters

Objective function: linear regression, Max depth: 3, Verbosity: 0, Learning rate: 0.3, N estimators: 200

Objective function: squared error regression, Base score: 0.5, Booster: gbtree, Max depth: 6, Min child weight: −1, Missing: NaN. Verbosity: 0, Learning rate: 0.3, N estimators: 40, Random state: 10

Table 2. Prediction accuracy for equipment A1. Plant code

Prediction result details Method

Accuracy (XGBoost)

Accuracy (random forest)

A1

Classification

0.99834 (99.8%)

0.93409 (93.4%)

A1

Regression

2.173 (RSME)

2.170 (RSME)

4 Conclusion and Recommendations XGBoost model has shown a successful application to predict the oil and gas asset lifetime with high accuracy. Asset management strategies are not subjected to size constraints or geographical differences and can be implemented from a centrally situated position to supervise several institutions. Thus, a well-designed asset management system that is part of an asset management strategy is critical to ensure that all plant activities run smoothly. In conclusion, the XGBoost algorithm is proven to be a reliable and suitable model to predict machine failure as the accuracy obtained is higher than the Random Forest algorithm in the classification method. Acknowledgment. This research was funded by Yayasan Universiti Teknologi PETRONAS (YUTP) with a Cost Center 015LC0-277 for the Centre of Research in Data Science (CeRDaS).

References 1. Suppramaniam, S.U.K., Ismail, S., Suppramaniam, S.: Causes of delay in the construction phase of oil and gas projects in Malaysia. Int. J. Eng. Technol. 7, 203–209 (2018) 2. Calixto, E.: Gas and Oil Reliability Engineering: Modeling and Analysis, 2nd edn. Gulf Professional Publishing, Cambridge, USA (2016) 3. Mohammadpoor, M., Torabi, F.: Big Data analytics in oil and gas industry: an emerging trend. Petroleum 6(4), 321–328 (2018) 4. Li, W., Yin, Y., Quan, X., Zhang, H.: Gene expression value prediction based on XGBoost algorithm. Front. Genet. 10, 1077 (2019) 5. Gumus, M., Kiran, M.S.: Crude oil price forecasting using XGBoost. In: 2017 International Conference on Computer Science and Engineering (UBMK), pp. 1100–1103. IEEE, Antalya, Turkey (2017)

Predictive Analytics for Oil and Gas Asset Maintenance

115

6. Torlay, L., Perrone-Bertolotti, M., Thomas, E., Baciu, M.: Machine learning–XGBoost analysis of language networks to classify patients with epilepsy. Brain Inf. 4(3), 159–169 (2017). https://doi.org/10.1007/s40708-017-0065-7 7. Bikmukhametov, T., Jäschke, J.: Oil production monitoring using gradient boosting machine learning algorithm. IFAC-PapersOnLine 52(1), 514–519 (2019) 8. Toutounchian, S., Abbaspour, M., Dana, T., Abedi, Z.: Design of a safety cost estimation parametric model in oil and gas engineering, procurement and construction contracts. Saf. Sci. 106, 35–46 (2018) 9. Njualem, L., Smith, M.: Exploring the effects of enterprise resource planning systems on direct procurement: an upstream asset-intensive industry perspective. Int. J. Supply Oper. Manag. 5(4), 396–402 (2018) 10. Hartmann, J., Inkpen, A.C., Ramaswamy, K.: Different shades of green: global oil and gas companies and renewable energy. J. Int. Bus. Stud. 52(5), 879–903 (2020) 11. Aremu, O.O., Palau, A.S., Parlikad, A.K., Hyland-Wood, D., McAree, P.R.: Structuring data for intelligent predictive maintenance in asset management. IFAC-PapersOnLine 51(11), 514– 519 (2018) 12. Mihet, R., Philippon, T.: The economics of Big Data and artificial intelligence. In: Disruptive Innovation in Business and Finance in the Digital World, vol. 20, pp. 29–43. Emerald Publishing Limited (2019) 13. Aziz, S., Dowling, S.: Machine learning and AI for risk management. In: Disrupting Finance, pp. 33–50. Palgrave Pivot, Cham (2019) 14. Eissa, H.: Unleashing Industry 4.0 opportunities: big data analytics in the midstream oil & gas sector. In: International Petroleum Technology Conference. OnePetro (2020) 15. Sun, Z., Strang, K., Li, R.: Big data with ten big characteristics. In: Proceedings of the 2nd International Conference on Big Data Research, pp. 56–61, Association for Computing Machinery, New York, USA (2018) 16. Lu, H., Guo, L., Azimi, M., Huang, K.: Oil and Gas 4.0 era: a systematic review and outlook. Comput. Ind. 111, 68–90 (2019) 17. Al-Alwani, M.A., Britt, L.K., Dunn-Norman, S., Alkinani, H.H., Al-Hameedi, A.T., Al-Attar, A.M., Alkhamis, M., Al-Bazzaz, W.H.: From data collection to data analytics: how to successfully extract useful information from big data in the oil & gas industry? In: SPE/IATMI Asia Pacific Oil & Gas Conference and Exhibition, OnePetro (2020) 18. Khodnenko, I., Ivanov, S., Perets, D., Simonov, M.: Detection of lost circulation in drilling wells employing sensor data using machine learning technique. Procedia Comput. Sci. 156, 300–307 (2019) 19. Aliguliyev, R.M., Imamverdiyev, Y.N.: Big data strategy for the oil and gas industry: general directions. Probl. Inf. Technol., 31–42 (2017) 20. Aziz, N., Abdullah, M.H.A., Zaidi, A.N.: Predictive analytics for crude oil price using RNNLSTM neural network. In: 2020 International Conference on Computational Intelligence (ICCI), pp. 173–178. IEEE (2020) 21. Lamb, K.: Challenges of digitalisation in the offshore oil and gas sector. In: CDBB Research Bridgehead Report, CDBB (2018) 22. Jaf, M.F.: Using big data technologies in petroleum industry. In: AIP Conference Proceedings 2207, p. 040009. AIP Publishing LLC (2020) 23. Carlin, D., O’Kane, P., Sezer, S.: Dynamic analysis of ransomware using opcodes and opcode categories. Int. J. Cyber Situat. Aware. 3(1), 84–97 (2018) 24. Jiang, H., Liu, G., Li, J., Zhang, T., Wang, C., Ling, X.: An innovative diagnosis method for lost circulation with unscented Kalman filter. J. Petrol. Sci. Eng. 166, 731–738 (2018) 25. Xie, L., Håbrekke, S., Liu, Y., Lundteigen, M.A.: Operational data-driven prediction for failure rates of equipment in safety instrumented systems: a case study from the oil and gas industry. J. Loss Prev. Process Ind. 60, 96–105 (2019)

116

N. Aziz et al.

26. Ahlgren, F., Mondejar, M.E., Thern, M.: Predicting dynamic fuel oil consumption on ships with automated machine learning. Energy Procedia 158, 6126–6131 (2019) 27. Tang, X., Wang, X., Cattley, R., Gu, F., Ball, A.D.: Energy harvesting technologies for achieving self-powered wireless sensor networks in machine condition monitoring: a review. Sensors 18(12), 4113 (2018) 28. Topchiev, A.: Design and monitoring of oil and gas industry facilities based on the use of ultra-light aviation and digital technologies. In: IOP Conference Series: Materials Science and Engineering, vol. 734, no. 1, p. 012005. IOP Publishing (2020) 29. Djaidir, B., Guemana, M., Kouzou, A., Hafaifa, A.: Failure monitoring of gas turbine based on vibration analysis and detection. In: 6th International Conference on Systems and Control (ICSC), pp. 397–402. IEEE (2017) 30. Egbunike, C.F., Okerekeoti, C.U.: Macroeconomic factors, firm characteristics and financial performance. Asian J. Account. Res. (2018) 31. Lee, I.: Big data: dimensions, evolution, impacts, and challenges. Bus. Horiz. 60, 293–303 (2017) 32. Sousa, A.L., Ribeiro, T.P., Relvas, S., Barbosa-Póvoa, A.: Using machine learning for enhancing the understanding of bullwhip effect in the oil and gas industry. Mach. Learn. Knowl. Extr. 1(3), 994–1012 (2019) 33. Hanga, K.M., Kovalchuk, Y.: Machine learning and multi-agent systems in oil and gas industry applications: a survey. Comput. Sci. Rev. 34, 100191 (2019) 34. Orrù, P.F., Zoccheddu, A., Sassu, L., Mattia, C., Cozza, R., Arena, S.: Machine learning approach using MLP and SVM algorithms for the fault prediction of a centrifugal pump in the oil and gas industry. Sustainability 12(11), 4776 (2020) 35. Layouni, M., Hamdi, M.S., Tahar, S.: Detection and sizing of metal-loss defects in oil and gas pipelines using pattern-adapted wavelets and machine learning. Appl. Soft Comput. 52, 247–261 (2017) 36. Kanin, E., Osiptsov, A., Vainshtein, A., Burnaev, E.: A predictive model for steady-state multiphase pipe flow: machine learning on lab data. J. Petrol. Sci. Eng. 180, 727–746 (2019) 37. Flammini, M., Mauro, M., Tonelli, M., Vinci, C.: Inequity aversion pricing in multi-unit markets. In: 24th European Conference on Artificial Intelligence (ECAI), 321, pp. 91–98. IOS Press (2020) 38. Noshi, C.I., Schubert, J.J.: The role of machine learning in drilling operations; a review. In: SPE/AAPG Eastern Regional Meeting. OnePetro (2018) 39. Hegde, C., Gray, K.: Evaluation of coupled machine learning models for drilling optimization. J. Natl. Gas Sci. Eng. 56, 397–407 (2018) 40. Gurina, E., et al.: Application of machine learning to accidents detection at directional drilling. J. Petrol. Sci. Eng. 184, 106519 (2020) 41. Xue, L., Liu, Y., Xiong, Y., Liu, Y., Cui, X., Lei, G.: A data-driven shale gas production forecasting method based on the multi-objective random forest regression. J. Petrol. Sci. Eng. 196, 107801 (2021) 42. Andrianov, N.: A machine learning approach for virtual flow metering and forecasting. IFACPapersOnLine 51, 191–196 (2018) 43. Beyca, O.F., Ervural, B.C., Tatoglu, E., Ozuyar, P.G., Zaim, S.: Using machine learning tools for forecasting natural gas consumption in the province of Istanbul. Energy Econ. 80, 937–949 (2019) 44. Biau, G., Cadre, B., Rouvière, L.: Accelerated gradient boosting. Mach. Learn. 108(6), 971– 992 (2019). https://doi.org/10.1007/s10994-019-05787-1 45. Subasi, A., El-Amin, M.F., Darwich, T., Dossary, M.: Permeability prediction of petroleum reservoirs using stochastic gradient boosting regression. J. Ambient. Intell. Humaniz. Comput. 13, 3555–3564 (2020). https://doi.org/10.1007/s12652-020-01986-0

Predictive Analytics for Oil and Gas Asset Maintenance

117

46. Makhotin, I., Koroteev, D., Burnaev, E.: Gradient boosting to boost the efficiency of hydraulic fracturing. J. Pet. Explor. Prod. Technol. 9(3), 1919–1925 (2019). https://doi.org/10.1007/s13 202-019-0636-7 47. Schmidt, K.G., Jørgensen, T.M.: Oil production monitoring and optimization from produced water analytics; a case study from the Halfdan chalk oil field, Danish North Sea. IFACPapersOnLine 51(8), 203–210 (2018) 48. Song, M., Zhou, X.: A casing damage prediction method based on principal component analysis and gradient boosting decision tree algorithm. In: SPE Middle East Oil and Gas Show and Conference. OnePetro (2019) 49. Osman, A.I.A., Ahmed, A.N., Chow, M.F., Huang, Y.F., El-Shafie, A.: Extreme gradient boosting (Xgboost) model to predict the groundwater levels in Selangor Malaysia. Ain Shams Eng. J. 12(2), 1545–1556 (2021) 50. Cherif, I.L., Kortebi, A.: On using eXtreme gradient boosting (XGBoost) machine learning algorithm for home network traffic classification. In: 2019 Wireless Days (WD), pp. 1–6, IEEE (2019) 51. Gohiya, H., Lohiya, H., Patidar, K.: A survey of Xgboost system. Int. J. Adv. Technol. Eng. Res. 8, 25–30 (2018) 52. Fan, J., et al.: Comparison of support vector machine and extreme gradient boosting for predicting daily global solar radiation using temperature and precipitation in humid subtropical climates: a case study in China. Energy Convers. Manag. 164, 102–111 (2018) 53. Wang, S., et al.: A new method of diesel fuel brands identification: SMOTE oversampling combined with XGBoost ensemble learning. Fuel 282, 118848 (2020) 54. Gul, S., van Oort, E.: A machine learning approach to filtrate loss determination and test automation for drilling and completion fluids. J. Petrol. Sci. Eng. 186, 106727 (2020)

Event Detection and Information Extraction Strategies from Text: A Preliminary Study Using GENIA Corpus Mohd Hafizul Afifi Abdullah(B) , Norshakirah Aziz , Said Jadid Abdulkadir , Emelia Akashah Patah Akhir, and Noureen Talpur Computer Information Sciences Department, Universiti Teknologi PETRONAS, 32610 Seri Iskandar, Perak, Malaysia {mohd_20002084,norshakirah.aziz,saidjadid.a,emelia.akhir, noureen_19001744}@utp.edu.my

Abstract. In the world we live today, data is the new oil. Data can reveal hidden knowledge that gives us an advantage over our competitors. However, data that are present in an unstructured form such as text documents are difficult to be processed by conventional machine learning algorithms. Therefore, in this study, we attempted to perform information extraction from textual data using current and state-of-the-art models to understand their working mechanisms. To perform this study, we have chosen the GENIA corpus for evaluating the performance of each model. These selected event extraction models are evaluated based on specific measures which are precision, recall, and F-1 measure. The result of our study shows that the DeepEventMine model has scored the highest for trigger detection with a precision of 79.17%, recall at 82.93%, and F-1 measure at 81.01%. Similarly, for event detection, the DeepEventMine model has scored highest among other models with a precision of 65.24%, recall at 55.93%, and F-1 measure at 60.23% based on the selected corpus. Keywords: Text information extraction · NER · Event extraction

1 Introduction Over the years, businesses and organizations have amassed vast volumes of data; yet, only a few of them have been able to effectively exploit it because a large amount of data is costly to maintain and difficult to manage [1]. Furthermore, data is collected in a variety of formats and forms including structured, semi-structured, and unstructured data formats [2]. Different data structures employ different encoding mechanisms and cannot be merged easily due to incompatibility issues. As a result, structured and unstructured data must be processed to extract the information contained within them. The first step in processing data in multiple formats is to extract information from unstructured data, which is necessary for businesses to combine with other structured data for analytics. However, the current data mining strategies face challenges to process data with high heterogeneity [3]. The issue of data heterogeneity can be generalized as © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 118–127, 2023. https://doi.org/10.1007/978-3-031-20429-6_12

Event Detection and Information Extraction Strategies

119

a variety of data types, structures, formats, forms, units, or inconsistent terminologies usage. As data are gathered in various formats and sources, there is a challenge to extract information from these different structures, encoding them into a single structure, and maintaining the trustworthiness of the data source and data healthiness for data wrangling purposes [4]. In regard to IE from text, there is difficulty to extract event occurrences and their parameters from an unstructured text body, especially data curated by humans due to the big data’s variety problem [1, 2, 5]. The focus of this study is to explore event extraction methods for extracting event information from text data. There are several approaches that can be used to reduce the problem which includes providing additional knowledge to build the training model [6] and utilizing the capability of deep learning or deep neural network to learn the hidden patterns within the unstructured data [7, 8]. Thus, this study aims to evaluate the performance of the existing event extraction models to extract event-related information from the GENIA corpus, which consists of text coming from biomedical publications. The remaining of this paper is organized as follows. The next section (Sect. 2) presents the literature studies regarding the existing event extraction models as well as their challenges in terms of the ‘variety’ issue in text data. Section 3 presents the methodology to execute simulations of event extraction as presented in previous studies. Section 4 discusses the results of the simulations, while Sect. 5 concludes the study.

2 Literature Review Information extraction (IE) from unstructured textual data is a very crucial part of a data pipelining process as textual data is not numerical, thus it is difficult to be processed directly by machines. Therefore, several IE techniques and models are designed to extract the information from text data and create a representation of the text in numerical values, which can be processed by machines. Generally, Natural Language Processing (NLP) is used to identify and extract information from text data based on 3 common strategies which are Named Entity Recognition (NER), Relation Extraction (RE), and Event Extraction (EE) [5]. NER is used to identify and extract named entities from text, RE is used to identify the relation between two or more named entities from a text, and EE is used to extract events-related information from a text [2, 5]. Various methods can be used to extract useful information from text such as named entities, relations, and events occurrence. For instance, other machine learning models such as neural networks, backpropagation neural networks (BPNN), and spiking neural networks (SNN) have proven to be powerful for classification and regression tasks. Therefore, combining IE techniques for textual data with these machine learning models will be beneficial for creating a more efficient and accurate extraction model. However, the most recent advancement in data science such as deep neural networks (DNN) or deep neuro-fuzzy systems (DNFS) has shown reliable and robust performance in solving classification and regression problems [7, 9]. This situation has prompted researchers to consider implementing methods such as DNN or deep learning models for classifying named entity information, which can then be used to improve event extraction since event detection is dependent on high-quality named entity information. The DNN approach offers the capability of learning further details from the data provided

120

M. H. A. Abdullah et al.

to the model. In addition, DNFS has the reasoning capability advantages that it inherits from the fuzzy inference system with the help of rules as well as learning capability from the deep learning approach [8, 9]. For example, DNN has also been implemented in the detection and classification of COVID-19 medical images [10]. In the end, the extracted information can be used for various purposes, including aiding the business management team to make informed decisions based on data. In the case of event extraction, the system needs to be able to detect the presence of an event from text based on “triggers”, then identify the arguments of the event which are entities that are related to the events [11]. Further explanation regarding event extraction is explained in the next subsection. 2.1 Event Detection and Extraction Generally, an event in the text is identified using triggers, which are verbs (or normalized verbs) that identify the event that exists in the text [12]. To do this, the dataset or a text corpus is initially tokenized (segmented) into words, and each word is assigned with an ID (token). Note that the majority of event extraction models presume that the entities are provided for the event detection task [13]. From here, the trigger words are identified to detect the presence of an event. Once the event is identified, related information (arguments) to the event are identified as well [12]. Each of the extracted information can be stored in dataframes for further processing. Finally, the outcome of the event detection can be scored using evaluation matrices. Figure 1 below presents the general data pipelining process for event extraction.

Fig. 1. General data pipelining process for EE.

The event extraction method is adapted in various methods across published works. However, the fundamental idea remains the same, which is to extract the information about event occurrences from the text [14]. As depicted above, event extraction begins by tokenizing the text data by sentence and words (and in some cases, into sub-words). Then, the next process is to detect the event type, which will then be used to retrieve the event argument table. Once the event type is identified, the event arguments are extracted, and the result of event extraction can be provided. A model scoring can be produced by using precision, recall, and F1 measures as follows: Precision = Recall = F1-measure =

TP TP + FP 

TP TP + FN 

2 × Precision × Recall Precision + Recall

(1) (2) (3)

Event Detection and Information Extraction Strategies

121

Despite event extraction has been implemented in various applications, it faces challenges when presented with social media data compared to formal news articles (due to noise and ambiguity representation) [15]. Other than that, the technique also faces challenges when dealing with a huge amount of unstructured data. The next subsections will explain several event extraction models employed in this preliminary study. 2.2 TEES 2.1 Extraction Model Turku Event Extraction System 2.1, or TEES 2.1 [16] is an extraction system by the Turku Centre for Computer Science (TUCS), University of Turku. The initial version of TEES was developed for the BioNLP’09 Shared Task [17] and BioNLP’11 Shared Task [18]. Due to its generalization, TEES is not limited to biomedical corpora but applies to other corpora automatically [16]. The TEES 2.1 system divides event extraction into four main steps which are parsing, trigger detection, edge detection, and unmerging (see Fig. 2). Initially, entity identification is done for each word in a sentence [18]. The named entities can either be provided or identified using named entity detection methods within the system. Next, the sentences are parsed to create dependency parse for the words before trigger words are searched during the trigger detection phase. To ensure trigger detection is efficient, any entities not given during the phase are sought during the trigger detection phase. Afterward, edge detection identifies the events and arguments between entities. In the end, entity and argument sets are separated into different events by unmerging the created knowledge graph [18]. Finally, event modality is predicted. The arguments are paired with the target protein (event). Figure 2 shows the steps involved in the process of event extraction using the TEES 2.1 extraction model.

Fig. 2. An overview of the TEES event extraction process [18].

2.3 EVEX Text Mining Resource in ST’13 EVEX is a text mining resource that was created based on the PubMed abstracts and PubMed open-access publication’s full-text [19]. EVEX text mining resource was extracted from the publications metadata and full-text using the BANNER named entity detector [20] combined with the TEES event extraction system [18] which were made publicly available. It provides events generalizations thus allowing summarization of knowledge across various documents [21]. The generalizations of events are used to group events of similar types and arguments together [6]. The confidence score was

122

M. H. A. Abdullah et al.

calculated based on the TEES classification procedure for each event, where a higher confidence value means the higher confidence of the predictions made. In their work, Hakala et al. [6] have presented a methodology to improve event extraction performance from biomedical publications by integrating EVEX resources. The goal of this work was to see if an extended event context extracted from documents on a large scale (EVEX text mining resource) could improve the performance of event extraction. Figure 3 presents the methodology for extracting events from ST’13 data based on EVEX resources and TEES 2.1 which are trained based on ST’11 data.

Fig. 3. Overview of the EVEX text mining resource process with TEES 2.1 for extracting events from the GENIA corpus.

Despite EVEX being incorporated into the TEES 2.1 presents a promising methodology to capture events using additional knowledge from the PubMed abstracts and full-text, the model can only improve precision, but not recall. This happens since the model is not capable to create new events after removing events from the TEES output. EVEX implementation to TEES has presented a great method to incorporate additional knowledge for creating a generalized data model. 2.4 DeepEventMine Extraction Model DeepEventMine is a state-of-the-art (SOTA) end-to-end nested event extraction model that capable of extracting many overlapping events from a raw sentence using Bidirectional Encoder Representation from Transformers (BERT). This model was developed using the DNN architecture that has been pre-trained on large-scale corpora and is based on EventMine [22]. This model consists of several deep learning and machine learning modules including BERT [23], Embeddings from Language Models (ELMo) [24], and OpenAI Generative Pre-Training (GPT) [25]—and have presented impressive results on several NLP tasks. Figure 4 visualizes the implementation of the DeepEventMine model which consists of 4 layers: the BERT layer, entity/trigger layer, role layer, and the event layer. For simplicity, the hidden layers are not presented in the visualization. This model employs the pre-trained SciBERT model for the entity identification procedure. For the entity/trigger layer, it employs the model of Sohrab and Miwa [26] which enables the extraction of nested triggers and entities at once. In the role layer, it detects all possible combinations of the trigger and argument pairs (such as triggertrigger and trigger-entity) and assigns a role type or no role to the pairs, provided that the entity/trigger layer can capture all triggers and entities from the text. To create event candidates for each trigger, the event layer computes all allowed combinations of role pairings, including events with no arguments. Next, each candidate is classed as either

Event Detection and Information Extraction Strategies

123

Fig. 4. Overview of the DeepEventMine event extraction model [13].

an event (class 1) or a non-event (class 0). This layer also recognizes event modification like speculation or negation. As a result, the DeepEventMine model has the advantage of detecting nested entities and triggers, roles, nested events, and their modifications in an end-to-end manner [13]. A nested event is an event with another event in its arguments, while a flat event is an event with only its entities in the arguments.

3 Methodology Before we simulate the existing EE models, the first step is to gather and prepare the datasets for the experiments. The datasets selected are identified based on the most commonly used in related studies. In this study, we have selected the GENIA corpus, which consists of annotated text collections from biomedical publications for running the experiment. The corpus is compatible with IE from text tasks, including event extraction from the text. The GENIA task has three subtasks which are the detection of events based on the main arguments (task 1), detection of entities related to the events (task 2), and detection of negation or speculation (task 2). The GENIA task exemplifies the benefits of automatic learning for the event annotation scheme. To ensure the simulation runs well with a proper runtime, we recommend a GPU machine running 64-bit kernels with a minimum of 16 gigabytes of RAM. Thus, this simulation has been executed on a machine running the latest stable version of the Linux Ubuntu (20.04 LTS) operating system with a 64-bit kernel, an Intel i7 processing chip, GPU-enabled, and 16 gigabytes of RAM. The system is equipped with Python version 3.9 to execute the program, with required dependencies pre-installed. Each model is executed using the parameters, settings, optimizer parameters, and experimental procedures as given in Björne and Salakoski [16] (TEES 2.1), Hakala, et al. [6] (EVEX), and Trieu, et al. [13] (DeepEventMine). First, the data is obtained from the resource link http://www.geniaproject.org/geniacorpus. During data preparation, the data is prepared according to each model’s configuration and are placed in the input directory of each model. Next, three different models are configured (TEES 2.1 [16], EVEX [6], and DeepEventMine [13] and trained accordingly. The parameters are optimized by iteratively training the extraction model. Finally, the result obtained after training and testing the extraction models are observed and recorded. Figure 5 presents the flow of experimental procedures using the three different models. The learning models have utilized the Adam optimizer [27], which is simply a stochastic optimization algorithm based on gradient-descent optimization. This is due to

124

M. H. A. Abdullah et al.

Fig. 5. Experimental procedures for setting and running the models.

the straightforward implementation of Adam, enabling it to be very efficient especially in terms of memory consumed during computational process. Other suitable optimization methods that can be explored includes Equilibrium Optimizer (EO) [28] and Rider Optimization Algorithm (ROA) [29].

4 Result and Discussion For the trigger detection and event detection of each model’s simulations, the overall result is measured based on the precision (Eq. 1), recall (Eq. 2), and F1 (Eq. 3). Table 1 presents the outcome of the simulations. Table 1. Overall performance of event detection on GENIA corpus. Model (Author)

Measure

Trigger detection (%) Event detection (%)

TEES 2.1 (Björne and Salakoski [16]) Precision 62.69 Recall 49.40 F-1 55.26

56.32 46.17 50.74

EVEX (Hakala et al. [6])

Precision 64.30 Recall 48.51 F-1 55.30

58.03 45.44 50.97

DeepEventMine (Trieu et al. [13])

Precision 79.17 Recall 82.93 F-1 81.01

65.24 55.93 60.23

Based on the result, it is evident from Table 1 that the DeepEventMine model has received the highest precision rate in terms of trigger detection and event detection. The DeepEventMine model has achieved 16.48% higher precision for trigger detection and 8.92% higher precision for event detection than the TEES 2.1 model; whereas 14.87% higher precision for trigger detection and 7.21% higher precision for event detection as compared to the EVEX model. In terms of recall, the DeepEventMine model has shown the highest recall rate among the three models. It has scored 33.53% higher recall for trigger detection and 9.76% higher recall for event detection than the TEES 2.1 model; while 34.42% higher recall for trigger detection and 10.49% better recall than the EVEX model. This shows that the deep learning methods are capable to provide a better recall as compared to the conventional machine learning models.

Event Detection and Information Extraction Strategies

125

F

82.93

81.01 60.23

55.3 50.97

R

55.93

48.51 45.44

64.3 58.03

55.26 50.74

60

Event Detecon

49.4 46.17

80

Trigger 62.69 56.32

100

79.17 65.24

On the other hand, in the perspective of the F1-measure, the DeepEventMine model has presented the highest F1-measure compared to the other two models. The DeepEventMine model has achieved a 25.75% higher F1-measure for trigger detection and 9.49% better F1-measure for event detection than the TEES 2.1 model; whereas 25.71% higher F1-measure for the trigger detection and 9.26% higher event detection as compared to the EVEX model. Based on the overall result of the event trigger and event detection on the GENIA corpus above, the DeepEventMine model presented by Trieu, et al. [13] has been able to conduct event extraction with the highest precision, recall, and F1 as compared to other models. Figure 6 visualizes the comparison of the performance for the three models based on trigger detection and event detection.

40 20 0 P

R

F

Björne and Salakoski [14]

P

Hakala, et al. [20]

P

R

F

Trieu, et al. [11]

Fig. 6. Performance comparisons for 3 different event extraction models on the GENIA corpus.

Even though the DeepEventMine model has shown the best performance among other SOTA models, there are still some problems that this model cannot resolve. For instance, this model struggles to identify events when roles, triggers, entities, arguments, and nested arguments are not present in the text [13]. The most critical issues are the missing role and missing trigger, which causes the model unable to detect the event from the data. Thus, researchers can focus their efforts to tackle the most critical issue to further improve the performance of event detection in the future.

5 Conclusion and Future Recommendation The techniques for extracting information from textual data can generally be categorized as the extraction of named entities, extraction of relations between entities, and extraction of events and their arguments. In this study, we assessed three existing models for extracting event information from the GENIA corpus. Based on the results, the simulation shows that the DeepEventMine model has scored the best among the three models simulated in terms of precision, recall, and F1-measure for trigger detection and event detection. This presumes that models with a deeper network (or DNN) have a higher capability to reason the association between words to detect the trigger words and event

126

M. H. A. Abdullah et al.

occurrences from a text. However, this does not imply that a DNN’s ability to associate hidden patterns in the data is the sole reason for its success; other factors may contribute to the model’s performance. In the context of improving mechanisms for event extraction in the future, researchers may focus on solving the issues of missing roles and missing triggers, which is a major contributing factor to the performance of these models. In most cases, missing trigger words in a sentence cause the model to miss an event in the text, making the model prone to error. Other than that, the models can be tested with other benchmark datasets suitable for event extraction tasks. Acknowledgment. This research was funded by Yayasan Universiti Teknologi PETRONAS (YUTP) with a Cost Center 015LC0-277 for the Centre of Research in Data Science (CeRDaS).

References 1. Sivarajah, U., Kamal, M.M., Irani, Z., Weerakkody, V.: Critical analysis of Big Data challenges and analytical methods. J. Bus. Res. 70, 263–286 (2017) 2. Adnan, K., Akbar, R.: An analytical study of information extraction from unstructured and multidimensional big data. J. Big Data 6(1), 1–38 (2019). https://doi.org/10.1186/s40537019-0254-8 3. Giudice, P.L., Musarella, L., Sofo, G., Ursino, D.: An approach to extracting complex knowledge patterns among concepts belonging to structured, semi-structured and unstructured sources in a data lake. Inf. Sci. 478, 606–626 (2019) 4. Gandomi, A., Haider, M.: Beyond the hype: big data concepts, methods, and analytics. Int. J. Inf. Manag. 35(2), 137–144 (2015) 5. Adnan, K., Akbar, R., Wang, K.S.: Information extraction from multifaceted unstructured big data. Int. J. Recent Technol. Eng. 8, 398–1404 (2019) 6. Hakala, K., Van Landeghem, S., Salakoski, T., Van de Peer, Y., Ginter, F.: EVEX in ST’13: application of a large-scale text mining resource to event extraction and network construction. In: Proceedings of the BioNLP Shared Task 2013 Workshop, pp. 26–34. Association for Computational Linguistics, Sofia, Bulgaria (2013) 7. Talpur, N., Abdulkadir, S.J., Hasan, M.H.: A deep learning based neuro-fuzzy approach for solving classification problems. In: 2020 International Conference on Computational Intelligence (ICCI), pp. 167–172. IEEE (2020) 8. Talpur, N., Abdulkadir, S.J., Alhussian, H., Hasan, M.H., Aziz, N., Bamhdi, A.: Deep NeuroFuzzy System application trends, challenges, and future perspectives: a systematic survey. Artif. Intell. Rev., 1–49 (2022)https://doi.org/10.1007/s10462-022-10188-3 9. Talpur, N., Abdulkadir, S.J., Alhussian, H., Hasan, M.H., Aziz, N., Bamhdi, A.: A comprehensive review of deep neuro-fuzzy system architectures and their optimization method. Neural Comput. Appl. 34(3), 1837–1875 (2022) 10. Albahri, O.S., et al.: Systematic review of artificial intelligence techniques in the detection and classification of COVID-19 medical images in terms of evaluation and benchmarking: taxonomy analysis, challenges, future solutions and methodological aspects. J. Infect. Public Health 13(10), 1381–1396 (2020) 11. Adnan, K., Akbar, R.: Limitations of information extraction methods and techniques for heterogeneous unstructured big data. Int. J. Eng. Bus. Manag. 11, 1847979019890771 (2019) 12. Miwa, M., Thompson, P., Korkontzelos, Y., Ananiadou, S.: Comparable study of event extraction in newswire and biomedical domains. In: Proceedings of COLING, the 25th International Conference on Computational Linguistics: Technical Papers. Dublin City University and Association for Computational Linguistics, Dublin, Ireland (2014)

Event Detection and Information Extraction Strategies

127

13. Trieu, H.L., Tran, T.T., Duong, K.N., Nguyen, A., Miwa, M., Ananiadou, S.: DeepEventMine: end-to-end neural nested event extraction from biomedical texts. Bioinformatics 36(19), 4910–4917 (2020) 14. Wang, P., Deng, Z., Cui, R.: TDJEE: a document-level joint model for financial event extraction. Electronics 10(7), 824 (2021) 15. Finkel, J.R., Grenager, T., Manning, C.D.: Incorporating non-local information into information extraction systems by Gibbs sampling. In: Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL’05), pp. 363–370 (2005) 16. Björne, J., Salakoski, T.: TEES 2.1: Automated annotation scheme learning in the BioNLP 2013 shared task. In: Proceedings of the BioNLP Shared Task 2013 Workshop, pp. 16–25. Association for Computational Linguistics, Sofia, Bulgaria (2013) 17. Björne, J., Heimonen, J., Ginter, F., Airola, A., Pahikkala, T., Salakoski, T.: Extracting complex biological events with rich graph-based feature sets. In: Proceedings of the BioNLP 2009 Workshop Companion Volume for Shared Task, pp. 10–18 (2009) 18. Björne, J., Salakoski, T.: Generalizing biomedical event extraction. In: Proceedings of BioNLP Shared Task 2011 Workshop, pp. 183–191 (2011) 19. Van Landeghem, S., et al.: Large-scale event extraction from literature with multi-level gene normalization. PLoS One 8(4), e55814 (2013) 20. Leaman, R., Gonzalez, G.: BANNER: an executable survey of advances in biomedical named entity recognition. In: Biocomputing 2008, pp. 652–663. World Scientific (2008) 21. Van Landeghem, S., Ginter, F., Van de Peer, Y., Salakoski, T.: a PubMed-scale resource for homology-based generalization of text mining predictions. In: Proceedings of BioNLP 2011 Workshop, pp. 28–37. Association for Computational Linguistics, Portland, Oregon, USA (2011) 22. Miwa, M., Pyysalo, S., Ohta, T., Ananiadou, S.: Wide coverage biomedical event extraction using multiple partially overlapping corpora. BMC Bioinform. 14(1), 1–12 23. Devlin, J., Chang, M.W., Lee, K., Toutanova, K.: Bert: pre-training of deep bidirectional transformers for language understanding. arXiv preprint (2018) 24. Peters, M.E., Neumann, M., Iyyer, M., Gardner, M., Clark, C., Lee, K., Zettlemoyer, L.: Deep contextualized word representations. In: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pp. 2227–2237. Association for Computational Linguistics, New Orleans, Louisiana (2018) 25. Radford, A., Narasimhan, K., Salimans, T., Sutskever, I.: Improving language understanding by generative pre-training (2018) 26. Sohrab, M.G., Miwa, M.: Deep exhaustive model for nested named entity recognition. In: Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pp. 2843–2849 (2018) 27. Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization. In: Proceedings of the 3rd International Conference on Learning Representations (ICLR). San Diego, CA, USA (2015) 28. Mohammed, Y.R., Basil, N., Bayat, O., Mohammed, A.H.: A new novel optimization techniques implemented on the AVR control system using MATLAB-SIMULINK. Int. J. Adv. Sci. Technol. 29(05), 4515–4521 (2020) 29. Mohamadwasel, N.B.: Rider optimization algorithm implemented on the AVR control system using MATLAB with FOPID. In: IOP Conference Series: Materials Science and Engineering, vol. 928(3), p. 032017. IOP Publishing (2020)

Human Evacuation Movement Simulation Model: Concepts and Techniques Noor Akma Abu Bakar1(B)

, Siew Mooi Lim1 , and Mazlina Abdul Majid2

1 Faculty of Computing and Information Technology, Tunku Abdul Rahman University, College

(TAR-UC), Wilayah Persekutuan Kuala Lumpur, Malaysia {noorakma,siewmooi}@tarc.edu.my 2 Faculty of Computing, Universiti Malaysia Pahang (UMP), Pekan, Malaysia [email protected]

Abstract. An emergency scenario is an unforeseen situation that threatens human life, and it is related to the emergent movements of evacuees, which is a critical challenge to model as their movements are unpredictable However, such practices provide less concern on how humans move, individual behaviours and individual differences, obstacles and other components. Therefore, modelling and simulation (M&S) are one of the methods that can be used to face this issue. Modelling is a method of solving problems that can be replaced by a simple object that describes the real system with its behaviour. A program with a running algorithm of a computer model is called a computer simulation. In order to develop a simulation model, a conceptual model consisting of a few components such as input, output, and techniques to be used is found to be important to be investigated for modelling the human evacuation egress (EE) movements. Therefore, two simulation techniques were found appropriate for modelling human EE, namely Social Force (SF) and Agent-based (AB). AB is autonomous with self-directed agents that pursue a series of predefined guidelines and rules to accomplish the objectives whilst the interaction among agents and the environment. Whereas SF is an approach to representing human behaviour with social-psychological and physical forces. The primary aim of this work is to review previous conceptual models and to propose a preliminary concept for modelling the human EE simulation. The findings reveal that the significant important components, such as the concept of the EE simulation model, have been identified based on the appropriateness and importance of each, such as the simulation techniques, EE movement procedure, and EE movement state. The conceptual model will be designed to assist in the development process of the EE simulation model for future work. Keywords: Simulation model · Social force · Agent-based · Human movements · Evacuation egress · Data analytics

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 128–137, 2023. https://doi.org/10.1007/978-3-031-20429-6_13

Human Evacuation Movement Simulation Model

129

1 Introduction Modelling is defined as replicating a situation to describe a real system and its behaviour by replacing it with a simple object. Furthermore, modelling is important as a method to solve real-world problems efficiently. While a simulation which is related to human movement is a computer-based approach which imitates the movement of a single person and the collective actions of a group [1]. Usually, a model appears with simulation together and running with the computer code or program to signify the complex system. Possible causes to be investigated which relate to the human movements are difficult to model correctly specifically, the human movements during real evacuation practice or training. However, such practices provide less concern on how humans move, individual behaviours and individual differences, obstacles and other components [2]. Research by [3] reported that simulation and modelling with visualization are often used in the stage of planning to improve performance efficiency. On the other hand, a simulation model is important in saving time. This paper is organized into five primary sections; the first section delivers the introduction whereas the second section describes the related work on the conceptual model. Next is the third section which provides the conceptual modelling for EE human simulation model, next section delivers the findings and discussion on the proposed conceptual modelling. The final section provides the conclusion and future works.

2 Related Works 2.1 Evacuation and Human Movements Fire disasters is a harmful scenario that threatens people’s force and lives. When the fire happens, an evacuation emergency will occur. The people started to move to enable them to quickly evacuate from the disaster area. The people involved will try to immediately avoid the hazardous area and move to a safer area. The people involved will try to immediately avoid the hazardous area and move to a safer area. Thus, it is vital to be investigated as giving impacts and implications for the safe evacuation. Furthermore, human movement is difficult to model correctly during real evacuation practice or training because such practices are not concerned with how people move, individual emotions and individual differences as well as obstacles. Figure 1 presents the summary of the human evacuation theory. This theory was adopted from the provided definitions and descriptions of human evacuation by a few past research works. The theory procedure behind evacuation has explained that in a real situation of human evacuation, not all pedestrians or people are able to identify the particular evacuation path [4]. Instead of their behaviours based on self-organisation with spontaneous behaviour of non-linear interaction of many objects other people may also influence them [5]. The previous statement was supported by [6]. They reported the behaviour indicates based on the actions and own judgments are also affected by public behaviours. Some of the pedestrians were moving independently [7] while some of them prefer to get along with other people, specifically friends or family. Human movement is the motion and flow of people from the first place to the destination. The movement or motion involves the changes of a person in a place, position

130

N. A. A. Bakar et al.

Fig. 1. Human evacuation movements

or posture relative to some point in the environment. This movement seemed to be a solid definition that satisfied all necessary conditions. Nevertheless, expanded this definition by claiming the movement is not uniform and intensely stratified and needed to be identified and analyzed thoroughly at its many layers. The analysis of human motion originates as far back as the fifth century BC when Aristotle and his colleagues developed a model of the human musculoskeletal system involving levers, forces and a center of gravity [8]. While egress is the action of going out of or leaving a place. Egress of occupants from a facility is generally straightforward under normal conditions. Problems arise when an emergency happens and many occupants are attempting to egress as quickly as possible. People will always try to find the shortest and easiest way to reach their destination. If possible, they will avoid obstacles and heading to the shortest way got crowded. The basic principle is the “least effort principle”, which means everyone tries to reach their goal as fast as possible to spend the least amount of energy and time. Thus, studying egress behavior in emergencies is difficult since it often requires real people to be exposed to the actual scenarios, possibly the dangerous scenarios [9]. 2.2 Modelling and Simulation (M&S) The significance of simulations enables the researchers to predict human flows for the evaluation of architectural designs and operational plans. The simulation modelling commonly used in operational research as a tool to represent the real conditions in a system and to study the characteristics of the simulation models [10]. The simulation model with multiple types of applications also had been applied in different fields such as human movement and behaviour, health and medical. Besides, simulation model is important due to some cases could not run the actual experiments in real-life because of the technical difficulties. Moreover, a simulation model is needed because of the ethical and risk as well. Research by [11] reported that simulation and modelling with visualization are often used in the stage of planning to improve performance efficiency. On the other hand, a simulation model is important in saving time and cost as well as to ensure the effectiveness of the model in solving the problems of modelling the human movement (operations) that occurred in the real world. Recently, there are various existing simulation techniques with different research scopes of various study areas which to overcome several issues. However, a simulation model based on the study area

Human Evacuation Movement Simulation Model

131

usually requires specific components such as the simulation technique and conceptual modelling. 2.3 Techniques: Agent-Based and Social Force One of the well-known techniques in the evacuation model is social force-based and known as Social Force Model (SFM) or Social Force Simulation (SFS) or Social Force (SF) which is related to human motions or movements with the behaviour or directions. SFS technique is a human behaviour model based on social-psychological and physical forces. There are three forces in SFS, namely as desired force, interaction force between human. Therefore, these three types of forces can be applying for the proposed work as necessary for modelling the human movement in evacuation. A simulation model with SFS technique consisted of the queue principle. The queueing element is significant to use for predicting the waiting behaviour in the systems. Queue needed if the demand for a service is more considerable than the service offered or available. Queue system also assists in understanding the system behaves with numerous environments for a simulation mode. AB has the characteristics of a decentralised control system. The decentralised control system is realistic and faithful compared to the centralised. The autonomy of individuals and free will in taking decisions is natural to many social scenarios or phenomena. In addition, ABS also covers the individual-centric approach. Individual-centric lays the people (or refer as agent in simulation model) in a specific environment. AB model with agent; which is referring to a set of agents with the characteristics (attributes) and behaviour. Then, an agent’s relationship and interaction among the agents is presented as proactive behaviour. The third element is the agents’ environment which comprises the connection between the environment and the agents. Furthermore, the interactions between the various behaviours of people and global behaviour are also covered in ABS. There are three advantages of ABS compared to other modelling techniques; AB is able to model the emergent phenomena, using deductive and inductive reasoning [12]. Then, ABS offers an accurate depiction of a system, and lastly, it is flexible for a simulation model. ABS found suitable to use for modelling the human movement in crowd evacuation. Conclusively, the single simulation technique such as AB or SF is not adequate to be used individually to resolve EE simulation model issues. The limitations from both simulation models can be solved by combining both models’ characteristics in one simulation model to improve simulation results and gain closer real scenario representation. In addition, the solution of combing two techniques into one is in order to achieve consistent simulation results, with affordable computational costs while mimicking individual movements in better execution time. Aside from focusing on simulation results, a combined technique or also known as a hybrid technique is proposed due to human behavioural complexities and computer resource limitations. These hybrid techniques are the extension of singele crowd technique to overcome problems faced in existing techniques i.e., AB and SF and to combine the capabilities of both.

132

N. A. A. Bakar et al.

2.4 Existing Conceptual Models Conceptual modelling for a simulation model described the scope and level of detail [13]. For instance, the previous work by [14] (as depicted in Fig. 2) discussed the important component of their simulation model is the conceptual modelling. It was used for integrating crowd modelling and energy analysis as well.

Fig. 2. Evacuation conceptual model by [14]

The conceptual modelling for a simulation model is vital to define the components, descriptive variables and interactions (logic) which constitute the system.

Fig. 3. Conceptual framework for evacuation simulation by [15]

The preliminary design for the simulation models is a conceptual model, and the researchers have designed it as an initial step which intends for integrating crowd modelling and energy analysis. Furthermore, [16] described that developing a preliminary model or concept either graphically (e.g. block diagram or process flowchart) is required to define the components, descriptive variables and interactions (logic) which constitute the system. Figure 3 depicts the component for a simulation model of the crowd evacuation (CE) simulation model. Their conceptual modelling for the simulation model covered the factors of socio-demographic character and environment of the hazard. Moreover, their research focused on the spatiotemporal of the population risk using GIS which is appropriate for outdoor instead of closed space building. Other than that, previous researchers proposed their evacuation model as depicted in Fig. 4; the conceptual model for the terminal passenger experience simulation model [17]. They designed the

Human Evacuation Movement Simulation Model

133

pedestrian experience for the airport layout and the human movement. However, it has not described as the details for the evacuation simulation model. Conclusively, there are few designs or conceptual models as a preliminary design for a running simulation model based on the various research domain and area. Thus, the conceptual model is found as one of requirements and it is important for simulation specifically for EE simulation model (this research work).

Fig. 4. Conceptual model of terminal passenger experience [18]

3 Evacuation Simulation Model The statistics from 20 fire reports describe that more than 50% emergency evacuation happened in commercial buildings, while the rest occur in residential areas. Thus, it motivates the researchers to investigate the human movement during evacuation egress and focuses on close area such as office buildings. The standard evacuation procedure by [19] found it as the appropriate step to apply any evacuation simulation model as it is involved the pedestrian or people, and the evacuation is happen in a close area. Figure 5 presents the illustration of evacuation procedure. When the signal appears, then the emergency evacuation needs to be determined. The people will respond to the signal, moving to the exits. Then, the waiting and queuing element needed for the regrouping and congestion during the evacuation. Lastly, the evacuation process completed.

Fig. 5. Illustration of the evacuation procedure [20]

Figure 6 presents the flow chart for developing a simulation model by [1]. This set of steps focused on the simulation of fire crowd evacuation (CE) in a building. It focused on the building layout to be simulated concerning the elements of human movement,

134

N. A. A. Bakar et al.

path selection and evacuation actions as the requirements for the EE simulation model. The output was the report and graphics visualizer as the researchers used the 2D or 3D simulation software. However, the steps ignored the validation and verification, experimentation and simulation result analysis. Thus, researchers intend to adopt these simulation steps with additional steps which is including validation and verification for CE or EE simulation model.

Fig. 6. Steps for a CE/EE Simulation model [21]

Instead of simulation steps and procedures, the movement state of each people is an essential to identify as well. The normal state describes the regular activities by the human or people or also known as agent. A simulation model provides a formal description of the physical and social aspects of the real world for presenting the critical parts of a system (under investigation) as well as the interaction between these parts. Meanwhile, people moving out from fire by stating that “a conceptual of behavioural model is a composite of existing theories and data that has been drawn together to represent some portion of people behaviour”. The outlines of a conceptual model consist of aims, model input, model output, scenario, scopes or level of details Robinson 2013. Therefore, from the above reviews, the appropriate techniques have been chosen and the concepts of human EE simulation model has been described such as movement procedure, human behaviour during EE and the evacuation movement state.

4 Discussion, Conclusion and Future Works A conceptual model is defined as a “persistent artefact” which describes the composition of the simulation model and is known as the model abstract. Then, the model abstract described the activity for “what to model” and “what not to model”. One of the critical elements for a simulation model is the concepts and the researchers found and described the concepts of the human EE simulation model. As one of future works, researchers intend to extend the works on the designing the conceptual modelling for EE/CE simulation model. In addition, the concepts determine the preliminary steps of a conceptual model; described the scope and level of detail. One of the findings of this research work is the comparison of the features of ABS and SFS. This comparison of SFS versus ABS is intended to provide an insight into the SFS and ABS in considering the significant features for modelling human evacuation and is useful for future researchers as well.

Human Evacuation Movement Simulation Model

135

Thus, Table 1 presents the advantages and drawbacks of two techniques; ABS and SFS. In conclusion, the human EE movement simulation model will be developed based on the identified concepts and simulation techniques which are already described in this paper. The significant components were identified such as the simulation techniques, and the concepts such as EE movement procedure, human behaviour during EE and also the human EE movement state. Table 1. Advantages and drawbacks of SFS and ABS Simu-lation tech-nique Advantage

Limitation

SFS

• Effective memory usage • Consider high-pressure characteristics • Occupant moves in an arbitrary direction • Simple form and its small number of parameters • Capable of modelling the localized motion and interactions of passengers • Process flow with the queue system • Egress element for the emergency evacuation • Emphasize in the obstacle avoidance • Good represent of crowd evacuation

Lack of realism for high density • Oversimplified the process of wayfinding of the pedestrian through the traffic flow • Time consumptions • The simulation complexity using equations • No proactive behaviour • Based on the process flow and no intelligent–independent object/entity

ABS

Capture emergent phenomena • Provide a natural description of a system • Presumption of equilibrium is not required • Flexible • Intelligent- independent entity • Natural mapping among the agents • Agent as a self-contained object or entity (set of characteristic or attribute) • Proactive behavior • Decentralized control system

Consider as highly sophisticated cognitive models • The demand for memory and processor of the computer • Computer resources required • Not emphasize on the rules in modelling how the agent is working • ABS focused on the decision-making and less concern on movement representing the foundation • Difficulties in developing a simulation model • High skills in computation • Not cover the queue system

Therefore, the EE simulation model will be developed using two appropriated techniques namely SF and AB techniques. Then, for future work, the researchers propose to

136

N. A. A. Bakar et al.

investigate which simulation technique is better and most effective to model the human EE movements by implement the simulation runs/experimentation. Other than that, the impactful factors, and their impact on the simulation results which are referred to as performance measurement parameters (PMP) can be revealed in future work. The further research work is aimed to provide a significant impact for the EE simulation models explicitly as a baseline to the researcher; specifically, the developed simulation models will be tested during experimentation based on the chosen factors and performance measures (PMP). Acknowledgement. This research is financially supported by Faculty of Computing Information Technology, Tunku Abdul Rahman University College (TAR-UC). Appreciation to the reviewers and editors for considering this paper.

References 1. Edrisi, A., Lahoorpoor, B., Lovreglio, R.: Simulating metro station evacuation using three agent-based exit choice models. Case Stud. Transp. policy 9(3), 1261–1272 (2021) 2. Hassanpour, S., Gonzalez, V., Liu, J., Zou, Y., Cabrera-Guerrero, G.: A hybrid hierarchical agent-based simulation approach for buildings indoor layout evaluation based on the postearthquake evacuation. Adv. Eng. Inf. 51, 101531 (2022) 3. Kano, T., Iwamoto, M., Ueyama, D.: Decentralised control of multiple mobile agents for quick, smooth, and safe movement. Phys. A Stat. Mech. Appl. 572, 125898 (2021). https:// doi.org/10.1016/j.physa.2021.125898 4. Li, M.X., Zhu, S.B., Wang, J.H., Zhou, Z.: Research on fire safety evacuation in a university library in Nanjing. Procedia Eng. 211, 372–378 (2018). https://doi.org/10.1016/j.proeng.2017. 12.025 5. Giordano, G., Sjöberg, J.: Black- and white-box approaches for cascaded tanks benchmark system identification. Mech. Syst. Signal Process. 108, 387–397 (2018). https://doi.org/10. 1016/j.ymssp.2018.01.008 6. Zhang, H., Liu, H., Qin, X., Liu, B.: Modified two-layer social force model for emergency earthquake evacuation. Phys. A Stat. Mech. Appl. 492, 1107–1119 (2018). https://doi.org/10. 1016/j.physa.2017.11.041 7. Eliyan, L.F.: Modeling family behaviors in crowd simulation, no. January, 2017 8. Dutt, A., Ismail, M.A.: A Systematic Review on Educational Data Mining, vol. 5, pp. 15991– 16005 (2017) 9. Noor Akma, A.B., Ainil Fahsha, A.N., Majid, M.A., Adam, K., Younis, Y.M., Fakhreldin, M.: Vehicular traffic flow simulation model using discrete event simulation: a case study. In: 3ICT Bahrain 2018, pp. 1–6. IEEE (2018) 10. Yuan, Z., Guo, R., Tang, S., He, B., Bian, L., Li, Y.: Simulation of the separating crowd behavior in a T-shaped channel based on the social force model. IEEE Access 7, 13668–13682 (2019). https://doi.org/10.1109/ACCESS.2019.2894345 11. Zankoul, E., Khoury, H., Awwad, R.: Evalutation of agent-based and discrete-event simulation for modeling construction earthmoving operations. In: Proceedings of the 32nd International Symposium on Automation and Robotics in Construction and Mining (ISARC 2015), no. June (2017). https://doi.org/10.22260/isarc2015/0014 12. Marshall, D.A., et al.: Selecting a dynamic simulation modeling method for health care delivery research part 2: report of the ISPOR dynamic simulation modeling emerging good practices task force. Value Heal. 18(2), 147–160 (2015). https://doi.org/10.1016/j.jval.2015. 01.006

Human Evacuation Movement Simulation Model

137

13. Robinson, S., Arbez, G., Birta, L.G., Tolk, A., Wagner, G.: Conceptual modeling: definition, purpose and benefits. In: Proceedings of the Winter Simulation Conference, vol. 2016-Febru, pp. 2812–2826 (2016). https://doi.org/10.1109/WSC.2015.7408386 14. Wu, S., Del Zendeh, E.: A conceptual framework to simulate building occupancy using crowd modelling techniques for energy analysis, no. October 2016 (2017) 15. Jumadi, S., Quincey, D.: A conceptual framework of volcanic evacuation simulation of Merapi using agent-based model and GIS. Procedia Soc. Behav. Sci. 227, 402–409 (2016). https:// doi.org/10.1016/j.sbspro.2016.06.092. No. November 2015 16. Page, S.: Systems Simulation: Topics in Descriptive Simulation Modeling Techniques for Sensitivity Estimation Simulation-based Optimization Techniques, pp. 1–45 (2013)

A Data Mining Approach to Determine Prospective Debtor of Unsecured Credit (Case Study: Bank XYZ in Indonesia) Devi Fitrianah1(B) , Anita Ratnasari2 , S. Ayu Eka Fuji2 , and Siew Mooi Lim3 1 Computer Science Department, School of Computer Science, Bina Nusantara University,

11530 Jakarta, Indonesia [email protected] 2 Faculty of Computer Science, Universitas Mercu Buana, 11650 Jakarta, Indonesia 3 Faculty of Computing and Information Technology, Tunku Abdul Rahman University College, Kuala Lumpur, Malaysia

Abstract. The motivation of this research is to provide alternative way for determining prospective debtor of unsecured credit using a data mining approach. In credit provision, especially for unsecured credit, bank needs to analysis the need of prospective debtor, since the high level of credit in Bank and the large amount of prospective customer applying for loan make the credit analyst find it difficult to determine the creditworthiness. By using appropriate credit assessment model, bank may evaluate whether prospective debtor is worthy to be approved or not. So far, the credit assessment model utilized in banking mostly based on human expertise as the credit analyst. Not often the evaluation of determining the prospective debtor turn out incorrect result and lead to bad credit. A solution to solve the existing problem is to use data mining approach that create a classification model C4.5 algorithm. This model supports the decision making based on classification of customer’s creditworthiness. The classification model generated 9 rules in classifying the data. Utilizing the k-fold cross validation, the accuracy level reach 94.78% and it is considered as excellent findings. We can understand that the C4.5 classification model can be used to assist in determining prospective debtor. Keywords: Classification · C4.5 algorithm · Unsecured credit · Prospective debtor · Decision tree · Data mining

1 Introduction Bank, as one of financial institutions, has significant role for the economic sustainability in Indonesia, through its function as public fund collector as well as an institution allocating funds to numerous parties, by re-allocating such funds to the society through credit. Good management data is one of resources that may be used to increase the said excellency, since it is correlated with the high level of bad credit in Bank [1, 2]. Based on Law of Republic of Indonesia Number 10 of 1998 on the Amendment to Law Number © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 138–149, 2023. https://doi.org/10.1007/978-3-031-20429-6_14

A Data Mining Approach to Determine Prospective Debtor

139

7 of 1992 on the Banking, as Bank refers to business entity that collects public funds in the form of deposit and allocates such funds to society in the form of credit and others to improve society’s living standard [3]. There is also other study in banking application utilizing classification method C4.5 and K-NN [4]. In allocating its credit, Bank XYZ must be significantly able in assessing the customers in the future. Activities related to credit is affected by many factors, both internal and external, in which one of the inhibiting factors is the issue on the decision of creditworthiness such as revenue, dependents and receivables owned by the prospective debtor. This factor is very essential for banking parties since this will show the worthiness of any business or individual for credit provision in-order to minimize bad credit. The difficulties in determining the prospective debtor lie to the integration of required data and the limitation of human analysis in making the final decision whether to approve the loan or not. A data driven method is required to assist the loan analyst to gain a right decision based on the data. Based on the problem stated above, a machine learning algorithm, C4.5 will be implemented as the model to learn the provision data and gain the best model to determine the prospective debtor in an unsecured credit. Using the model, the banking parties may easily identify and figure out the relationship among the influencing factors of any problem and may discover the best solution from them. According to Fitrianah and Fahmi [5], decision tree is one of classification methods that uses the representation of tree structure, whereas the uppermost is called as root and each node represents attribute, while its branches represents the value of attribute, and its leaf represents the class. Decision tree is powerful and popular classification and prediction method [6]. The method of decision tree changes huge factors to be decision tree representing rule. Rule may be easily understood by natural language and, they may also be expressed in the form of basic data language such as Structured Query Language to figure out record in particular category [7]. The use of Algorithm C4.5 in the previous researches have been applied to different fields. Hijriana and Rasyidan [8] conducted a research and the methodological accuracy of Decision Tree of Algorithm C4.5 as one of the alternatives to support decision in selecting the prospective awardees in University XYZ so that grant of scholarship fund may be accurate. Relating this study for predicting the academic success by Mesalic and Sebalj [9] and predicting student performance also using C4.5 [10]. Other research that applied the classification method C4.5 is to provide the recommendation of sales partner for PT. Atria Artha Persada [11]. Another data mining research in education has been conducted in [12] to conduct prediction study upon the score of final projects obtained by the students of diploma program of information management. In a study conducted a predictive study of the determination of student majors by classification model [13]. Furthermore, implemented algorithm C4.5 to predict the performance of middle school students [14]. Maurina and Fanani [15] conducted prediction study to the implementation of scholarship recommendation in SHS of Muhammadiyah Gubug School. Subsequently, in the field of health, researches using the C4.5 algorithm show excellent result [16–18]. In addition, another study using C4.5 in health was conducted by Angkasa and Fitrianah [19] to determine the illness risk level for Indonesia Health Insurance. The feature data used is the data transaction of all insurance patient.

140

D. Fitrianah et al.

Using this model, the accuracy obtained is 99.87%. The promising results for implementing C4.5 algorithm for education fields also can be seen in [20–25]. Based on the above studies and the nature of the data in this study, it is decided to utilize the C4.5 algorithm to solve the problem in this study since the algorithm yield better accuracy and can generate rules based model based on the data.

2 Method This study uses quantitative research combined with the experimental research. The study was conducted based in the research design as presented in Fig. 1.

Fig. 1. Diagram block of the study design

The study design is described as in Fig. 1. There are five stages in this study. The first stage is the data collection. In this stage, the dataset is being collected from 5 years transactions. The dataset contains the customer data and being analyze against their credit data. Next stage is data pre-processing. In this stage, there are two sub activities: data cleaning and data transformation. The third stage is the implementation of C4.5 algorithm. This stage is to implement the algorithm using the dataset. Next is model validation. The model that is generated using the C4.5 algorithm is being validated using the k-fold cross validation method with the confusion matrix to find the accuracy. The last stage is the result analysis, where the paper will discuss the results of the method applied.

3 Result and Discussion In this section we will discuss the implementation of our research method on how to implement algorithm C4.5 using the data. 3.1 Data Collection In this study, the data used is gathered from the Bank XYZ for five years transactions. The data is prepared in order to obtain the required information to achieve the research objectives. Such information may be generated through relevant consultations and interviews with the parties concerned.

A Data Mining Approach to Determine Prospective Debtor

141

3.2 Data Pre-processing In the data pre-processing stage, the authors perform two-step activities: data cleaning and data transformation. • Data cleaning Once the problem is identified, the next phase is to do a data selection which involves decision variables determination. In this case, decision variable determinations were the risk level of a loan and the determinant attribute in decision tree formation. The attribute was selected by considering that the value of the attribute is not much. In other words, the attribute that has a big value will not be selected. The attributes that will be taken were the receivable, income, and dependents. – Receivables: Receivables are based on a customer’s credit record or history that is registered on Bank Indonesia. – Income: Represents the value of the debtor’s income on each month. – Dependents: The number of a family members which have a fixed expenditure on each month. For more details, the assessment criteria engaged in this research can be seen in Table 1 below. Table 1. Assessment criteria No

Attribute

1

Receivables – Secure – In concern – Doubtful

2

1 2 3

Income – Small – Medium – Big

3

Criteria

2.000.000–5.000.000 5.100.000–9.999.999 >10.000.000

Dependents – Few – Many

1–3 people >3 people

• Data transformation Data Transformation is the process of transforming or converting data into the appropriate form in order to be processed with C4.5 Algorithm calculation. The table to

142

D. Fitrianah et al.

be transformed was a table that contains customer data information as shown in this following Table 2. Table 2. Customer data No

Name

Addr

Receive

Income

Dep

Result

1

xxx

xxx

Doubtful

17,000,000

3

Yes

2

xxx

xxx

Secure

17,000,000

1

Yes

…

…

…

…

…

…

…

…

…

…

…

…

…

…

229

xxx

xxx

Doubtful

15,000,000

1

Yes

230

xxx

xxx

Secure

15,000,000

2

Yes

After that Table 2 was transformed to facilitate the calculation process by using the C4.5 algorithm. Table 3 is the requirement for data transformation to be performed. Table 3. Data transformation No 1

Variable Income

Value transformation Initial

Result

Nominal

Low Medium High

2

Receivables

Secured In concern Doubtful

3

Dependents

Nominal

Few Many

After determining the value for the data transformation, Table 2 which contains the customer data of XYZ Bank will change as presented in Table 4. The formation of algorithm decision tree was carried out with 230 research data in which the data used had passed the data transformation phase. Figure 2 shows the data that the categories of the attributes and labels have been specified. This will be used as a reference for making the decision tree. The data presented in the label are the result of the data while the others are attributes.

A Data Mining Approach to Determine Prospective Debtor

143

Table 4. Overall data after transformation No

Name

Addr

Receive

Income

Dep

Result

1

xxx

xxx

Doubtful

Big

Few

Yes

2

xxx

xxx

Secure

Big

Few

Yes

…

…

…

…

…

…

…

…

…

…

…

…

229

xxx

xxx

Doubtful

Big

Few

Yes

230

xxx

xxx

Secure

Big

Few

Yes

Fig. 2. Data on rapid miner

3.3 C4.5 Implementation To convert the data into a decision tree, the authors used C4.5 Algorithm. The data in Table 4 were classified based on the attributes of receivables, income, and debts in accordance with Table 3. The result is divided into 2 indicating that the credit is Secure to be approved and is Risky to be rejected. From the table, it is known that there are 230 data. 146 data are in the Secure category while the other 84 data are in the Risky category. Then, an entropy calculation was done as followed:         146 84 84 146 × log 2 + × log 2 = 0.946928968 Entropy(S) = − 230 230 230 230 After the entropy of the whole case was obtained, an analysis to each attribute and its values was performed. It was continued with a calculation of the entropy as shown in Fig. 3 which also became the result of the attribute calculation.

144

D. Fitrianah et al.

Fig. 3. Attribute calculation result

Once the entropy was calculated, the gain of each attribute was also examined. Entropy and Gain of all attributes can be seen in Fig. 4.

Fig. 4. Entropy and gain calculation result

From the results, the attribute of information gain was selected as the best classifier with the largest value of 0,212941273 in the Income attribute. Further, income was engaged to expand the tree or become the root of the decision tree. In the next sub-node, income cannot be used again for tree expansion. Based on the data in Table 4 (node 1), the data is reclassified with Small Income based on Accounts Receivable and Deeds. The result is divided into 2, which is Secured and Risky for the debtor. From Table 5, it is known that there are 230 data; 146 data show Yes and 84 show No. To determine the attribute that is the best classifier and acts as Subtree, the information gain needs to be calculated for all those attributes. Therefore, the total entropy for the overall attribute is calculated:          146 84 84 146 × log 2 + × log 2 = 0.94692896 Entropy(P) = − 230 230 230 230

A Data Mining Approach to Determine Prospective Debtor

145

Table 5. Data classification Node 1.1

Attribute

Value

Sum (Value)

Sum (Yes)

Sum (No)

Total

Income

230

146

84

Receivable

Secure

97

68

29

In concern

80

65

15

Doubtful

53

13

40

Few

122

88

34

Many

108

58

50

Dependents

After the result entropy and gain calculation was obtained, the decision tree will be shown as on Fig. 5. From the decision tree, the rule model obtained is as follows: 1. If the income is “low”, the result is = No In the first Rule, if the income has a low value, the answer will be No and the prospective debtor will not get a credit. 2. If the income is “high”, the receivables are “secure”, the result is = Yes In the second rule, if the income is high and has secured receivables, the answer will be Yes answer and the prospective debtor will get a credit. 3. If the income is “high”, the receivables are “doubtful”, the dependents are “high”, the result is = No In the third rule, if the income is high, has doubtful receivables and has so many number of dependents, the answer will be No and the prospective debtor will not get a credit. 4. If the income is “high”, the receivables are “doubtful”, the dependents are “few”, the result is = Yes In the fourth rule, if the income is high, has doubtful receivables and few dependents, the answer will be Yes and the prospective debtor will get a credit. 5. If the income is “high”, the receivables are “in concern”, the result is = Yes In the fifth rule, if the income is big and the receivables are in concern, the answer will be Yes and the prospective debtor will get a credit. 6. If the income is “medium”, the receivables are “high”, the result is = No In the sixth rule, if the income is medium and the receivables are high, the answer will be No and the prospective debtor will not get a credit.

146

D. Fitrianah et al.

7. If the income is “medium”, the receivables are “low”, and the dependents are “secure”, the result is = Yes In the seventh rule, if the income is medium, the receivables are low, and the dependents are secure, the answer will be Yes and the prospective debtor will get a credit. 8. If the income is “medium”, the receivables are “low”, and the dependents are “doubtful”, the result is = Yes In the eighth rule, if the income is medium, the receivables are low, and the dependents are doubtful, the answer will be Yes and the prospective debtor will get a credit. 9. If the income is “medium”, the receivables are “low”, and the dependents are “in concern”, the result is = Yes In the ninth rule, if the income is medium, the receivables are low, and the dependents are in concern, the answer will be Yes and the prospective debtor will get a credit.

Fig. 5. Decision tree result

3.4 Model Validation The final task is validating the model. It was performed with k fold cross-validation method which is a statistical method used to perform evaluation and comparison to a set of data by dividing the data into two parts such as training data and data testing. For the training we use 80% of data and for the testing we use 20% k is assigned to 10. The calculation was carried out by using the Confusion Matrix Model which will produce true positive (TP) and true negative (TN) result. The following Fig. 7 is a result of Confusion Matrix calculation by using C4.5 algorithm:

A Data Mining Approach to Determine Prospective Debtor

147

Fig. 7. The result of C4.5 algorithm model

Sensitivity = Specificity = Accuracy = PPV = Precision =

144 TP = = 0.986 TP + FN (144 + 2) TN 74 = = 0.880 TN + FP 74 + 10 144 + 74 TP + TN = = 0.947 TP + TN + FP + FN 144 + 74 + 10 + 2 144 TP = = 0.935 TP + FP 144 + 10 TN 74 = = 0.974 TN + FN 74 + 2

3.5 Result Analysis The explanation of the C4.5 implementation results can be seen through the validation of the model used. Based on the dataset used in training, a decision tree model was obtained with 9 rules and the accuracy reached 94.78% and it is considered as excellent findings. For other validation models, such as specificity, sensitivity, and precision, all of them reached the best values. This shows that the resulting model for the dataset works well. This is possible because with the decision tree model, all the generated rules managed to classify the test data very well, resulting in a high accuracy value.

4 Conclusion The research was conducted by using the method of data mining- classification with a C4.5 algorithm to determine the eligibility of prospective debtor had resulted in an accuracy value that was equal to 94.78%. This study also shows the data required to generated good classification model, they are customer profile data, customer ownership data, receivables, income and dependent. The model generated from this study comprises of 9 rules to be used as references in designing and making the appropriate application. Before implementation, we also performed data pre-processing to guarantee the usability of the data. From the result of the research, the C4.5 algorithm has a good accuracy to determine the creditworthiness without any collateral. We can conclude that the model generated from this research can be utilized to assist in providing recommendation for determining the prospective debtor in Bank XYZ.

148

D. Fitrianah et al.

References 1. Hidayatsyah, M.R.: Penerapan Metode Decision Tree Dalam Pemberian Pinjaman Kepada Debitur Dengan Mhd. Rido Hidayatsyah 5, 22 (2013) 2. Badawi, A.: Effect of credit risk, liquidity risk, and market risk banking to profitability bank (Study on Devised Banks in Indonesia Stock Exchange). Eur. J. Bus. Manag. 9, 1–8 (2017) 3. Rifqo, M.H., Arzi, T.: Implementasi Algoritma C4.5 untuk Menentukan Calon Debitur dengan Mengukur Tingkat Risiko Kredit pada Bank Bri Cabang Curup. J. Pseudocode. III, 83–90 (2016) 4. Chitra Devi, J.: Binary decision tree classification based on C4.5 and KNN algorithm for banking application. Int. J. Comput. Intell. Inf. 4, 125–131 (2014) 5. Fitrianah, D., Fahmi, H.: The identification of determinant parameter in forest fire based on feature selection algorithms. SINERGI 23, 184–190 (2019) 6. Dai, W., Ji, W.: A MapReduce implementation of C4.5 decision tree algorithm. Int. J. Database Theory Appl. 7, 49–60 (2014) 7. Mantas, C.J., Abellán, J.: Credal-C4.5: decision tree based on imprecise probabilities to classify noisy data. Expert Syst. Appl. 41, 4625–4637 (2014). https://doi.org/10.1016/j.eswa. 2014.01.017 8. Hijriana, N., Studi, P., Informatika, T., Islam, U.: Penerapan Metode Decision Tree Algoritma C4.5 Untuk Seleksi. Al-Ulum: Jurnal Sains Dan Teknologi 3(1), 9–13 (2017) 9. Mesari´c, J., Šebalj, D.: Decision trees for predicting the academic success of students. Croat. Oper. Res. Rev. 7, 367–388 (2016). https://doi.org/10.17535/crorr.2016.0025 10. Adhatrao, K., Gaykar, A., Dhawan, A., Jha, R., Honrao, V.: Predicting students’ performance using Id3 and C4.5 classification algorithms. Int. J. Data Min. Knowl. Manag. Process. 3, 39–52 (2013). https://doi.org/10.5121/ijdkp.2013.3504 11. Arifin, M.F., Fitrianah, D.: Penerapan Algoritma Klasifikasi C4.5 Dalam Rekomendasi Penerimaan Mitra Penjualan Studi Kasus: PT Atria Artha Persada. InComTech. 8, 87–102 (2018). https://doi.org/10.22441/incomtech.v8i1.2198 12. Paramita Mayadewi, E.R.: Prediksi Nilai Proyek Akhir Mahasiswa Menggunakan Algoritma Klasifikasi Data Mining. Sist. Inf. 11, 1–7 (2015) 13. Swastina, L.: Penerapan Algoritma C4.5 Untuk Penentuan Jurusan Mahasiswa. Gema Aktual. 2, 6 (2013) 14. Pambudi, R.H., Setiawan, B.D.: Penerapan Algoritma C4.5 Untuk Memprediksi Nilai Kelulusan Siswa Sekolah Menengah Berdasarkan Faktor Eksternal. 2, 2637–2643 (2018) 15. Maurina, D., Zainul Fanani, A.: Penerapan Data Mining Untuk Rekomendasi Beasiswa Pada Sma Muhammadiyah Gubug Menggunakan. 1–5 (2016) 16. Ihsan, Wajhillah, R.: Penerapan Algoritma C4.5 Terhadap Diagnosa Penyakit Demam Tifoid Berbasis Mobile. Swabumi. III, 1–8 (2015) 17. Viswanath, P., Suresh Babu, V.: Rough-DBSCAN: a fast hybrid density based clustering method for large data sets. Pattern Recognit. Lett. 30, 1477–1488 (2009). https://doi.org/10. 1016/j.patrec.2009.08.008 18. Mirqotussa’adah, M., Muslim, M.A., Sugiharti, E., Prasetiyo, B., Alimah, S.: Penerapan Dizcretization dan Teknik Bagging Untuk Meningkatkan Akurasi Klasifikasi Berbasis Ensemble pada Algoritma C4.5 dalam Mendiagnosa Diabetes. Lontar Komput. J. Ilm. Teknol. Inf. 8, 135 (2017). https://doi.org/10.24843/LKJITI.2017.v08.i02.p07 19. Angkasa, A., Fitrianah, D.: The implementation of classification algorithm C4.5 in determining the illness risk level for health insurance company in Indonesia. Int. J. Comput. Appl. 177, 44–50 (2020). https://doi.org/10.5120/ijca2020919883 20. Suhada, S., Setiawan, E.: Classification Needs Teachers, pp. 408–414

A Data Mining Approach to Determine Prospective Debtor

149

21. Harryanto, F.F., Hansun, S.: Penerapan Algoritma C4.5 untuk Memprediksi Penerimaan Calon Pegawai Baru di PT WISE. Jatisi. 3, 95–103 (2017) 22. Wirawan, V., Hansun, S., Nugroho, H.T.: Implementasi Algoritma Squeezer dan Term Frequency Ranking dalam Pembangunan Sistem Rekomendasi Tempat Makan. Ultim. Comput. VI, 25–32 (2014) 23. Putri, G.A., Maryono, D., Liantoni, F.: Implementation of the C4.5 algorithm to predict student achievement at SMK Negeri 6 Surakarta. Indones. J. Inf. Educ. 4, 10–19 (2021) 24. Wang, X., Zhou, C., Xu, X.: Application of C4.5 decision tree for scholarship evaluations. Procedia Comput. Sci. 151, 179–184 (2019). https://doi.org/10.1016/j.procs.2019.04.027 25. Damiaza, R., Fitrianah, D.: Prediction analysis of Kartu Jakarta Pintar (KJP) awardees in vocational high school XYZ using C4.5 algorithm. Int. J. Mach. Learn. Comput. 10, 44–50 (2020). https://doi.org/10.18178/ijmlc.2020.10.1.896

Date Palm Leaves Discoloration Detection System Using Deep Transfer Learning Shamma Alshehhi(B) , Shamma Almannaee, and Maad Shatnawi Higher Colleges of Technology, Abu Dhabi, UAE {H00366793,maad.shatnawi}@hct.ac.ae

Abstract. Most countries are concerned about the quality of their agricultural products. All facilities involved in the agricultural food production cycle must be examined. So, there are many challenging are facing this field, and the most prominent of these are crops diseases. In order to keep UAE one of the leading countries in producing dates, a date palm leaves discoloration detection system based on deep learning is proposed. This system uses three different convolutional neural network (CNN) models, which are SqueezeNet, GoogleNet, and AlexNet. All of the three models show a high validation & testing accuracy with a convenient training time. SequeezeNet achived the best test accuracy, which is 98%. The proposed system helps dates producers and farm owners to reduce the risk of diseases, and in turn, production and workforce expenses specially in remote areas and large farms. Keywords: Date palm tree diseases detection · Leaves discoloration detection · Crop diseases · Leave classification · Convolutional neural networks · Transfer learning · Deep learning

1 Introduction The date palm is a highly profitable crop that is mostly grown in tropical and subtropical countries [1]. As well, date palm is traditionally grown throughout the Arabian Peninsula, including the United Arab Emirates (UAE), which is currently one of the leading date palm producers in the world. In 2010 and 2016, the UAE produced 825,300 and 671,891 tons of dates, respectively, a decline of 18% in production levels [2]. This can be assigned due to the disease tensions, hostile climatic changes, and increasing production and workforce expenses. In order to keep UAE one of the leading countries in producing and exporting dates, addressing the current and ongoing challenges to date palm production in the UAE and around the world is critical. Therefore, it is significant to implement a system that defines if the palm tree has an illness from the leaf color. Machine learning is a rapidly developing field of artificial intelligence which simulates how people learn from their surrounding environment, while deep learning is a sort

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 150–161, 2023. https://doi.org/10.1007/978-3-031-20429-6_15

Date Palm Leaves Discoloration Detection System

151

of machine learning that imitates how the human brain processes data such as objects detection, speech recognition, languages translation, and decisions making [3]. Convolutional Neural Networks, commonly referred to as deep-CNNs, are used to analyze input photos for object classification and detection. They can also be used to split and compress images [4]. Besides, transfer learning is a technique for improving a learner from one domain by transferring knowledge from a related domain [5]. Transfer learning approaches use machine learning algorithms applied to learn tasks and can often be viewed simply as extensions of those algorithms [6]. Pre-trained models are commonly applied to express transfer learning. A pre-trained model is one that has been trained on a huge benchmark dataset to tackle a problem that is comparable to the one we are working on [7]. The most important characteristics of a CNN model are network precision, speed, and size [8]. In this study, we have examined three CNN pre-trained models which are AlexNet, SqueezeNet, and GoogleNet. AlexNet is one of the most popular CNN models, and it was the first convolutional Network that used the GPU to increase performance [9]. It has five convolutional layers, three maxpooling layers, two normalization layers, two fully connected layers, and one Softmax layer in its design. As well, it accepts an image as an input of size is 27 × 27 × 3 and generates a label for each object in the image. SqueezeNet is a smaller network that was created to be a more compressed alternative to AlexNet. It has over 50 times fewer parameters than AlexNet but performs three times faster [9]. Also, it consists of 18 deep layers. The network can classify photos into 1000 different object categories, which it has learned on a variety of rich feature representations for a variety of images [10, 11]. GoogLeNet uses a CNN design that is substantially deeper and wider than typical DCNNs, at the tradeoff of a slight increase in evaluation time [12, 13]. The network structure has 22 layers deep, plus 5 pooling layers, and it uses a 1 × 1 convolution filter [8]. As well, it accepts an image as an input of size is 24 × 24. This work presents a date palm leaves discoloration detection system based on deep learning convolutional neural networks (CNN). We employ transfer learning to train several pre-trained CNNs on our palm tree leave image data that consists of three different categories (healthy, unhealthy, and dead). The system will help users to check the discolored tree leaves if is it infected with any diseases, pests, fungi, or insects, since most diseases symptoms start with the discoloring the leaf and changing in shape. This system can be utilized by the use of Unmanned Aerial Vehicles (UAVs). As it is proved the best and most efficient data gathering technology in the field since it is able to scan the whole field within a specific period set out [14]. UAVs use cameras to collect images and sensors to create a set of data that can be used to aid in farm monitoring and decision-making [15, 16]. It also provides instant visual information about the wide area to help farmers make fast decision-making [14]. These real-time photos can be analyzed to provide information that can affect farm management decisions for urgent treatment from pre-planting to post-harvest phases. The shape, height, texture, color, and growth rate of vegetation can all be used to create a pattern that can be used to mimic farmland [14].

152

S. Alshehhi et al.

2 Related Works Crops diseases identification has been investigated by numerous researchers. One of the key solutions [14] using remote sensing techniques, a database of healthy, unhealthy, and partially unhealthy crop photos was compiled, and the images were processed to construct a model for determining the best treatment strategies for maize plantation. Although, different strategies have been used to improve the processes, including the usage of neural networks, support vector machines, and fuzzy logic. But the deep learning approach of convolutional neural networks for picture classifications has shown rapid and good results. The model was supplied with features derived from vegetative photos, which were then categorized, segmented, and fed back onto it. The deep convolutional neural network (VGG16) is utilized to estimate plant health on a maize plantation with an average prediction accuracy of 99.58%. One of the applications proposed by Bari et al. [17] uses the Faster R-CNN model to provide a high-performing rice leaf infection detection system capable of diagnosing the most frequent rice illnesses in real-time. The method employs an enhanced RPN architecture that generates candidate regions by addressing the object position exceptionally accurately. Training the Faster R-CNN model with publicly accessible online and own real-field rice leaf datasets improves its resilience. The program correctly identified a healthy rice leaf 99.25% of the time. Aravind, K. R., & Raja, P [18] proposed a solution using deep learning algorithms, which is an automated disease diagnostic system. The system classifies ten different diseases that have symptoms on the leaves for four different crops, which are Eggplant, Hyacinth beans, Lime, and Ladies finger. The dataset was created by collecting leaf samples, and the study has been evaluated using six pre-trained deep learning models, namely AlexNet, VGG16, VGG19, GoogLeNet, ResNet101 and DenseNet201. The data augmentation technique was used to enhance the dataset by adding distortion, which improves generalization and minimizes overfitting difficulties. The system shows that the best network for the data validation is GoogLeNet, which was the highest accuracy of 97.3%. While in the test situation, VGG16 produced the best accuracy. Using computer vision and machine learning approaches, Kulkarni et al. [19] proposed a smart and efficient method for detecting agricultural illness. The suggested method can detect 20 distinct illnesses in five common plants (apple, corn, grapes, potato, and tomato) with 93% accuracy. The authors in [20] provided a novel method for detecting kiwifruit images utilizing an RGB-D sensor aligned RGB and NIR images and deep convolutional neural networks (CNN). A total of 39,678 fruits were classified in the RGB photos of the entire dataset, which included 1000 images of RGB and near-infrared (NIR) modalities. After that, the VGG16 network was fine-tuned to handle RGB and NIR parameters individually but simultaneously. The researchers utilized RGB-Only, NIR-Only, Image-Fusion, and Feature-Fusion as detecting modes. Image-Fusion was the best option, with a 90.7% accuracy and rapid detection.

Date Palm Leaves Discoloration Detection System

153

The proposed work in [21] contains a system that detects three of the most frequent diseases impacting palms today: Leaf Spots, Blight Spots, and the lethal insect red palm weevil using both image processing and machine learning approaches. These diseases are diagnosed by obtaining normal and thermal photos of palm trees, which are then processed using image processing algorithms. To distinguish between Leaf Spots and Blight Spots diseases, CNN (VGG) was utilized, and an SVM model was developed using the scikit-learn library for the Red Palm Weevil insect. The accuracy ratio success rates for the CNN and SVM algorithms were 97.9% and 92.8%, respectively. The proposed work in [22] describes a method for detecting the (Sudden Decline Syndrome) in four different stages as it affects date palms in Pakistan specifically and the world in general. The disease is clearly recognized, allowing necessary procedures to begin to reduce yield losses. On the basis of texture and color extraction methods, a customized CNN deep learning methodology is utilized to identify diseases. The system’s additional duty is to identify the stage of the disease when it has been detected. The dataset of 1200 date palm disease photos were created to test the suggested disease diagnosis technique, and it attained an overall accuracy of 89.4%. As mentioned, multiple times through this document that date palm trees play an important role in North-Africa and Middle East regions. One study we looked into further used the U-Net deep neural network model to distinguish individual date palm treetops from photos taken above the “Targa N’Touchka” Oasis plantations in Morocco’s southwest. They acquired a total of 420 images split into 70% for training and 30% for testing which left them with a classification efficiency of 96.94%. Moreover, an issue that seems to be significant is areas with dense plantation since it’s hard to identify [23]. Another research that considers palm trees to be important is the Al-Shalout and Mansour approach. However, they experimented in Jordan as appose to Morocco using convolutional neural networks (CNN). At the beginning stage, they focused on diseases found in date palm trees in Jorden and used this information to train their model with 139631 images for various diseases. The final classification efficiency is 80% [24]. We decided to implement a deep transfer learning system that aimed to save the date palm trees from pests and diseases from leaf discoloration. Since most diseases symptoms of the palm trees can be figured from the changing in the leaf color to the yellow and brown, and its shape. Additionally, the date palm tree is a very valuable plant in the middle east, and especially in the Arabia desert. Therefore, our work is distinguished from other systems, since none of the recent and stated diseases detection approaches in the related work were applied to date palm trees in the Arabian Gulf region.

3 Method In order to use pre-trained networks, we have to go through several steps. These steps are data acquisition and preprocessing, training and validation, and testing. For all the three transfer systems we implement, we started off by extracting our training data which consisted of 3 classes with a total of 216 images (95 healthy date palm tree images,

154

S. Alshehhi et al.

71 unhealthy date palm tree images, and 50 dead palm tree images) and dividing them into training and validation sets; 70% and 30% respectively. Approximately 80% of the images were taken using iPhone 13 pro camera, which is 12MP, ƒ/2.8 aperture from different locations in Abu Dhabi, UAE. Whereas 20% was collected from the available sources on the internet. We also set aside total of 50 images for testing, with thus they shouldn’t include in the training process. A sample of our dataset is shown in Fig. 1.

Fig. 1. Sample of date palm leave dataset

The photos have been modified to fit the CNN model’s input dimensions, since each network has a distinct input size. 227x227 is used by SqueezeNet and AlexNet, whereas 224x224 is used by GoogleNet. Moreover, all of the images were taken in RGB color, which makes it easier to extract the necessary details into a singular format (JPG). We employed data augmentation in our trained networks by acquiring many photos from various perspectives, surroundings and conditions. Furthermore, this allows us to increase our limited data pool without increasing cost and time which in return allows the model to perform better and more accurately [25]. We used three sorts of data augmentation which are reflection, rotation, and scaling. On each picture, a random rotation with an angle ranging from -40 to 40 degrees is applied. As well as a random scaling factor in the range of 0.5 to 2 is applied followed by flipping the image with around the y-axis. Prior to beginning training, the proper training time, learning rate, number of epochs, and validation frequency must all be configured. The training process may get halted if the initial rate is too low. Whereas the training process may become unstable or learn a sub-optimal set of weights too rapidly if the rate is too high. As a result, we set the initial learning rate to 0.0001, the validation frequency to 3, and the maximum epochs of 30

Date Palm Leaves Discoloration Detection System

155

since we intended to avoid as much as possible the failure of the training halt based on error rates. Then, the minimum batch size is 11 was added to the training settings.

4 Results and Discussion Modelling a network for training requires a very large amount of data, high computational power, and complex mathematics that takes a long time to produce results, while transfer learning removes the last layer of the trained network from the neural network [5]. This technique is applied here using AlexNet, SequeezeNet, and GoolgleNet to store what is learned with weights and fine-tune the weights learned from existing architectures. MATLAB 2021a was used to do the deep learning data classification. The AlexNet was trained using 20 epochs with 13 iterations per epoch, after several variations on the number of epochs to ensure that the data was well-trained and validated. The network achieved a validation accuracy of 93.75% after 260 iterations. The network worked 15 min and 2 s to finish the training, which is the shortest training period of time compared to the other networks. In SqueezeNet, the training used 30 epochs, with 13 iterations each epoch. It attained a validation accuracy of 93.75% after 390 iterations. It took 23 min and 1 s to complete the training. The obtained accuracy is exactly the same as the AlexNet. For GoogleNet to well train and validate the data, the network required 30 epochs, with 13 iterations each epoch. It achieved a validation accuracy of 95.31% after 390 iterations. The training took the network 28 min and 26 s to complete, which is the longest training period of time compared to the other networks. For the three networks, we applied the same starting learn rate, iteration validation, and one CPU. However, the number of epochs was variable. Several variations on the number of epochs have been tried to ensure that the data was well-trained and validated at the maximum accuracy rate. We have tried 20, 30, 60 and 90 epochs, and all of them gave approximately the same accuracy percentage. However, we have selected the best ones, which are for 30 epochs and stated in our paper. As presented in Figs. Fig. 2, 3 and 4, that after a certain number of epochs the validation accuracy stays constant. As well, a 3-iteration validation approach was used to guarantee that the system was well-trained while not overfitting the data. As a result, in terms of highest validation accuracy, GoogleNet is the best network, and it takes an acceptable time to train. Then, AlexNet and SqueezeNet gave the second-best options. The long training time can be assigned due to the large number of parameters. Table 1 summarizes the training and validation outcomes of the three networks. After the training process is completed, the trained models were tested on unseen date palm tree images. The output of each image is the predicted class along with the confidence level. All networks have been tested using same images with a total of 50 (20 healthy date palm tree images, 20 unhealthy date palm tree images, and 10 dead palm tree images). A sample of testing results of the three networks are demonstrated below in (Figs. 5, 6 and 7). As well, a confusion matrix has been created of each model (AlexNet, GoogleNet, and SequeezeNet) to find the overall testing accuracy and determine the best

156

S. Alshehhi et al. Table 1. Training and validation results of the CNN models.

Model

Learning rate

Epochs

Validation accuracy

Elapsed time

Hardware resources

Max iterations

AlexNet

0.0001

20

93.75%

15 min 2 s

Single CPU

260

GoogleNet

0.0001

30

95.31%

28 min 26 s

Single CPU

390

SqueezeNet

0.0001

30

93.75%

23min 1 s

Single CPU

390

Fig. 2. AlexNet training and validation results

Fig. 3. SqueezeNet training and validation results

network as displayed in Tables 2, 3 and 4. To elaborate, the following formula is applied to calculate the accuracy from the confusion matrix: Accuracy =

True positive + True negative (1) True positive + True negative + False positive + False negative

Date Palm Leaves Discoloration Detection System

157

Fig. 4. GoogleNet training and validation results

Fig. 5. Sample of AlexNet testing results

Fig. 6. Sample of SqueezeNet testing results

Fig. 7. Sample of GoogleNet testing results

A comparison of existing date palm trees disease detection approaches in terms of machine learning technique used and accuracy is demonstrated in Table 5. Our approach achieved an accuracy of 98%, which is clearly visible that outperforms other approaches as shown in Fig. 8. Our system focuses on leaf discoloring and shape changing, as well as the appearance of the palm tree in general. Our created system can be considered as the first implemented date palm system in the Arabian Gulf region. In addition, it is easy for non-expert users since it uses simple words (healthy, unhealthy, and dead). Besides, it achieved a higher accuracy compared to all other approaches although they used higher dataset size than ours. As Alaa et al. (2020) is a competitor to our approach, but we achieved higher accuracy with less dataset compared to them. As well, our approach is different from them as it’s evaluated for the Gulf region, where the environment is harsh, the temperature is high, and the humidity is high. Hence, selecting the pre-trained

158

S. Alshehhi et al. Table 2. Confusion matrix of the AlexNet model Actual

Predicted

Healthy date palm tree

Unhealthy date palm tree

Dead date palm tree

Healthy date palm tree

20

0

1

Unhealthy date palm tree

0

20

3

Dead date palm tree

0

3

6

Table 3. Confusion matrix of the GoogLeNet model Actual

Predicted

Healthy date palm tree

Unhealthy date palm tree

Dead date palm tree

Healthy date palm tree

20

0

0

Unhealthy date palm tree

0

19

1

Dead date palm tree

0

1

9

Table 4. Confusion matrix of the SqueezeNet model Actual

Predicted

Healthy date palm tree

Unhealthy date palm tree

Dead date palm tree

Healthy date palm tree

20

0

0

Unhealthy date palm tree

0

19

0

Dead date palm tree

0

1

10

training with proper settings (learning rate, number of epochs, max batch size, etc.) help to attain high accuracy.

Date Palm Leaves Discoloration Detection System

159

Table 5. Comparison of date plant diseases detection approaches Approach

Geographic region Dataset size

Method/technique

Accuracy

Alaa et al. (2020)

Egypt

7,650

VGG & Digital image processing

97.9%

Magsi et al. (2020)

Pakistan

1,200

CNN

89.4%

Rhinane et al. (2021)

Morocco

420

Deep Neural Network (U-Net)

96.94%

Al-Shalout and Mansour (2021)

Jordan

139,631

Convolutional Neural Networks

80%

Our approach

Abu Dhabi (Gulf region)

266

CNN-GoogLeNet

98%

Fig. 8. Comparison of date palm trees disease detection approaches

5 Conclusion This study provides a date palm tree healthiness system that based on deep learning, where we used 3 pre-trained CNN models and trained them to fit our classifications task. With that, we compared all three methods from validation accuracy to training time and testing accuracies, where AlexNet and SqueezeNet had a validation accuracy of 93.75%. Whereas the GoogleNet is 95.31%, which is the best-obtained validation accuracy. While in the test situation, SequeezeNet achived the best accuracy, which is 98%. The proposed technique can help farmers to maintain track of their crops, lowering labor costs. This research could be expanded by looking at more transfer learning pretrained CNN models. This work can be also extended by including more date palm image data for training and validation.

160

S. Alshehhi et al.

References 1. Alwahshi, K., et al.: Molecular identification and disease management of date palm sudden decline syndrome in the United Arab Emirates. Int. J. Mol. Sci. 20(4), 923 (2019). https:// doi.org/10.3390/ijms20040923 2. Food and Agriculture Organization of the United Nations (FAO). Date Palm Production; FAOSTAT Database; FAO: Rome, Italy (2016) 3. Shatnawi, M., Almenhali, N., Alhammadi, M., Alhanaee, K.: Deep learning approach for masked face identification. Int. J. Adv. Comput. Sci. Appl. (IJACSA) 13(6) (2022). https:// doi.org/10.14569/IJACSA.2022.0130637 4. Basil, N., Raad, M., Wazzan, A.N., Marhoon, H.M.: Face Recognition with Real-Time Framing Based on Multi Task Convolutional Neural Network: A Case Study 5. Weiss, K., Khoshgoftaar, T.M., Wang, D.: A survey of transfer learning. J. Big Data 3(1), 1–40 (2016). https://doi.org/10.1186/s40537-016-0043-6 6. Torrey, L., Shavlik, J.: Handbook of Research on Machine Learning Applications and Trends. IGI Global, Hershey, PA (2010) 7. Marcelino, P.: Transfer learning from pre-trained models, 28 Oct. 2018. https://towardsdatas cience.com/transfer-learning-from-pre-trained-models-f2393f124751#:~:text=In%20comp uter%20vision%2C%20transfer%20learning,that%20we%20want%20to%20solve 8. Alhanaee, K., Alhammadi, M., Almenhali, N., Shatnawi, M.: Face recognition smart attendance system using deep transfer learning. Procedia Comput. Sci. 192, 4093–4102 (2021). https://doi.org/10.1016/j.procs.2021.09.184 9. “Great Learining Team” (24 Jun. 2020). “AlexNet: The First CNN to win Image Net”. AlexNet: The First CNN to win Image Net | What is AlexNet? (mygreatlearning.com) 10. Kurama, V.: A Review of Popular Deep Learning Architectures: ResNet, InceptionV3, and SqueezeNet (2020). A Guide to ResNet, Inception v3, and SqueezeNet | Paperspace Blog 11. Deng, W., Wu, R.: Real-time driver-drowsiness detection system using facial features. IEEE Access PP(99), 1–1 (2019). https://doi.org/10.1109/ACCESS.2019.2936663 12. Gupta, S.: Classify any object using pre-trained CNN model, 7 Jun. 2020. https://towardsda tascience.com/classify-any-object-using-pre-trained-cnn-model-77437d61e05f 13. pawangfg: Understanding GoogLeNet Model–CNN Architecture, 18 Nov. 2021. https://www. geeksforgeeks.org/understanding-googlenet-model-cnn-architecture/ 14. Abdullahi, H.S., Sheriff, R., Mahieddine, F.: Convolution neural network in precision agriculture for plant image recognition and classification. In: 2017 Seventh International Conference on Innovative Computing Technology (INTECH), vol. 10, pp. 256–272. Ieee (Aug. 2017) 15. UAVs and Precision Agriculture #15, 24 Mar. 2014. http://aerialfarmer.blogspot.co.uk/2014/ 03/uavs-and-precision-agriculture-15.html 16. Wright, D., Rasmussen, V., Ramsey, R., Baker, D., Ellsworth, J.: Canopy reflectance estimation of wheat nitrogen content for grain protein management. GISci. Remote Sens. 41(4), 287–300 (2004) 17. Bari, B.S., et al.: A real-time approach of diagnosing rice leaf disease using deep learningbased faster R-CNN framework. Peer J. Comput. Sci., San Diego, 7 Apr. 2021. https://doi. org/10.7717/peerj-cs.432 18. Aravind, K.R., Raja, P.: Automated disease classification in (Selected) agricultural crops using transfer learning, Automatika, 61:2, 260–272 (2020). https://doi.org/10.1080/00051144. 2020.1728911 19. Kulkarni, P., Karwande, A., Kolhe, T., Kamble, S., Joshi, A., Wyawahare, M.: Plant Disease Detection Using Image Processing and Machine Learning (2021). arXiv:2106.10698 20. Liu, Z., et al.: Improved kiwifruit detection using pre-trained VGG16 With RGB and NIR information fusion. IEEE Access 8, 2327–2336 (2020). https://doi.org/10.1109/ACCESS. 2019.2962513

Date Palm Leaves Discoloration Detection System

161

21. Alaa, H., Waleed, K., Samir, M., Tarek, M., Sobeah, H., Salam, M.A.: An intelligent approach for detecting palm trees diseases using image processing and machine learning. Int. J. Adv. Comput. Sci. Appl. 11(7), 434–441 (2020) 22. Magsi, A., Mahar, J.A., Razzaq, M.A., Gill, S.H.: Date palm disease identification using features extraction and deep learning approach. In: 2020 IEEE 23rd International Multitopic Conference (INMIC), pp. 1–6. IEEE (Nov. 2020) 23. Rhinane, H., Bannari, A., Maanan, M., Aderdour, N.: Palm trees crown detection and delineation from very high spatial resolution images using deep neural network (U-Net). In: 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, pp. 6516–6519 (2021)https://doi.org/10.1109/IGARSS47720.2021.9554470 24. Al-Shalout, M., Mansour, K.: Detecting date palm diseases using convolutional neural networks. In: 2021 22nd International Arab Conference on Information Technology (ACIT), pp. 1–5 (2021). https://doi.org/10.1109/ACIT53391.2021.9677103 25. Takimoglu, A.: What is data augmentation? Techniques & examples in 2022. In: AIMultiple (2021). https://research.aimultiple.com/data-augmentation/. Accessed 17 Jul. 2022

Solving Drinking-Water Challenges: Supply and Temperature in a Smart Poultry Monitoring System Using IoT System Ahmed Y. Mohammed, Harith A. Hussein, and Moceheb Lazam Shuwandy(B) Collage of Computer and Math Sciences, Tikrit University, Tikrit, Iraq [email protected]

Abstract. Poultry faces many problems that affect the life and productivity of chickens, including the traditional water supply. When the drinking water temperature for chickens increases, the consumption of drinking water by chickens decreases, which leads to health problems resulting in the death of chickens. The main goal is to create an intelligent system based on the Internet of Things that increases production by maintaining a tank’s drinking water temperature and level. In this paper, an ESP32 is used as a microcontroller with a BLYNK application, a DS18B20 sensor for drinking water temperature measurement, and a JSN-SR04T sensor for drinking water level measurement. The microcontroller maintains the drinking water temperature by controlling the cooling and heating system and maintaining the drinking water level through a water pump when the water level drops. Through testing the system, the results showed that the system could increase productivity by providing appropriate conditions, which indicates that the system is effective and suitable for farmers. Keywords: Internet of Things · Monitoring systems · Sensors · Smart poultry

1 Introduction In the modern chicken business, feeding system practices are one of the most crucial elements for poultry farmers, particularly in regions with high temperatures, since feeding systems can lower the temperature. In tropical and subtropical areas, heat stress is one of the significant variables impacting poultry performance [1]. The most crucial nutrient for poultry is water. Water intake during extreme heat stress in poultry environments may easily treble [2]. For grown birds to drink, the water’s temperature should be between 10 °C and 15 °C [3]. Birds drink less water when the temperature is one or two degrees higher than body temperature, which causes them to be highly thirsty [4]. It has been demonstrated that broilers and layers perform better when given chilled water. It will be advantageous to have a water cooler than the bird’s body temperature since drinking it will assist the bird’s body temperature drop [5]. Several areas in the smart poultry business need technical advancements and applications, including the control of drinking water supplies. Since water makes up 80% of © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 162–168, 2023. https://doi.org/10.1007/978-3-031-20429-6_16

Solving Drinking-Water Challenges: Supply and Temperature

163

a chicken’s nutritional requirements, water supply management is crucial for preserving poultry output [6]. However, there are several issues with a traditional water delivery system, including maintaining the water supply level and keeping the water temperature at an optimal level. Several research studies create models for managing the water supply based on the abovementioned issues. An Arduino and an Ultrasonic sensor are used to build the water level monitoring system [7, 10, 13]. The water pump will start when the water level dips, and important notices will be posted on the website. An Arduino UNO uses a microprocessor, a water level sensor to track changes in water volume, and a flow sensor to track leaks. A water supply management system was suggested in [6]. The work of the system is controlling the water level and determining if a leak is there or not. The Arduino likewise manages the water extension system on the urbanized chicken coop monitoring system using IoT [8, 11]. The water level is determined using the float switch. The high and low markers are placed at two different levels. The valve will turn on to allow water to be fed into the container when the water level is low. The valve will then close, and the water supply will be cut off if the water reaches the high-level mark. The control module in [9] was an Arduino, and the NodeMCU ESP8266 was used to link it to the Internet of Things. Water levels are monitored using the HC-SR04 Ultrasound Sensor, and water is dispensed using a Plastic Water Solenoid Valve. Once the sensor determines that the water bowl is empty, water is continually provided to the chicken. A system is in place to alert users when the water tank is running low. The design and implementation of a drinking-water supply management system App are the topics of this paper. The system is designed to keep track of the container’s water level and alert the owner when the water supply runs low. The water cooling and heating system keep the water at the ideal temperature. The controller is an ESP32, and the system’s interface is the BLYNK cloud.

2 Proposed System Overview In the proposed system, low-cost sensors and the Internet of Things are used to monitor and adjust chickens’ drinking water temperature and level. As shown in Fig. 1, the suggested system supports monitoring and controlling temperature and water levels. This system improves the health of chickens and their growth, and it is useful for poultry operations of all sorts, from small to big production. 2.1 Proposed Hardware Cost and detection range were taken into consideration while choosing the sensors. The controller was an ESP32 (see Fig. 2. B), a low-cost, low-power microcontroller with integrated Wi-Fi and dual-mode Bluetooth. The DS18B20 sensor (see Fig. 2. A), a waterproof digital sensor with a temperature range of −10 °C to +85 °C and an accuracy of 0.5 °C, is used to measure the water’s temperature. The water level in the tank was determined using the Waterproof Ultrasonic JSN-SR04T (see Fig. 2. C). It is a waterproof sensor that transmits and receives ultrasonic waves to detect distance or

164

A. Y. Mohammed et al.

Fig. 1. Monitor drinking water for chickens.

sense objects remotely. The water was heated and cooled using the water cooler. The switch for the load portion is a single channel 5V unit relay.

Fig. 2. A. DS18B20 Sensor. B. ESP32. C. Waterproof Ultrasonic JSN-SR04T.

2.2 Proposed Software To monitor data in the poultry houses, utilize BLYNK. BLYNK is an iOS and Android software that allows the owner to control Arduino modules [12], Raspberry Pi, ESP32, NodeMCU ESP8266, WEMOS D1s, and other internet-connected devices. The BLYNK application is used in this study to display the water temperature and the volume of water in the tank, as shown in Fig. 3. In the case of a specific problem, warning notices are provided to the application [14, 15]. 2.3 Implementation The ESP32 gets data through a digital connection from the DS18B20 sensor, which measures the temperature of the water. The ESP32 reaches the received value if the

Solving Drinking-Water Challenges: Supply and Temperature

165

Fig. 3. Monitor drinking water for chickens.

temperature exceeds the threshold value of 15 °C; ESP32 works the cooling system until the water temperature is within the optimal range and warns the farm owner by sending a warning message to the application, as shown in Fig. 4. When the temperature falls below 10 °C, the ESP32 runs the heating system until the value reaches the desired level and sends an application warning alert to the farm owner. When the distance grows above 20 cm, that is determined by the water level in the tank using the JSN-S04T sensor. This suggestion indicates a shortage of drinking water; thus, the ESP32 turn on the water pump until the distance value falls below 20 cm, at which point the farm owner is communicated by sending a notice to the application.

3 Results The prototype on the ESP32 keeps the water’s temperature and level within the optimal range in both the summer and the winter. The controller receives the values from the temperature and distance sensors and sends them in real-time to the farm owner through the BLINK app. The required action is conducted, and a warning message is delivered when the value deviates from the threshold value. In the experiment, broiler hens were separated into two experiments, with the first experiment receiving care according to the suggested system and the second experiment receiving care according to the conventional system. The selected findings of the suggested system in Table 1 represent most of the cases that were made as a result of changing environmental conditions. Sensors’ readings were taken in case of high temperature and low water level in the tank during summer. Furthermore, a group of other readings was selected in the winter. As a result of the

166

A. Y. Mohammed et al.

Fig. 4. A warning notice about high water temperature.

selected cases from the experiments, it was noted that in the winter season through the morning and afternoon, the sensor reading of the drinking-water temperature was high from 20 to 21 °C. Therefore, the system turned on the drinking water cooling system. In addition, the water level in the tank was normal in the morning and low in the afternoon, so the system activated the water pump. However, a drop in temperature case was observed in the winter during the afternoon and night, when the temperature was 6–8 °C. Nevertheless, the tank’s water level was normal at night and low in the afternoon; therefore, the system turn-on the water pump. In the only cases in which the water level in the tank was normal and the temperature was perfect, it was noticed that the heating and cooling systems and the water pump deactivated, as in Table 1. The experiment results showed that the first experiment of chickens had grown more weight than the second experiment by the time it was done. The demonstration of the experiment had a positive outcome.

4 Conclusion The paper aimed to increase productivity by providing the right conditions for chickens with less human intervention. This paper presents a water supply management system based on the Internet of Things. The system controls the water temperature to avoid heat stress that leads to reduced productivity and maintains the water level in the tank to prevent a lack of water from reaching the chickens. The system improved effectiveness and intelligence for remote monitoring and control during testing, addressing the factors affecting productivity. As a result, farmers benefit from increased productivity and profits.

Solving Drinking-Water Challenges: Supply and Temperature

167

Table 1. The effects of environmental changes. Environment

Water level (cm)

Water temperature (°C)

Water pump (ON/OFF)

Cooling system (ON/OFF)

Heating system (ON/OFF)

Status WL*

WT*

Groups

Cases

Group 1 summer season

Morning

15 cm

20 °C

Off

On

Off

Perfect

High

Noon

23 cm

21 °C

On

On

Off

Low

High

Group 2 winter season

Afternoon

25 cm

13 °C

On

Off

Off

Low

Perfect

Night

18 cm

12 °C

Off

Off

Off

Perfect

Perfect

Morning

26 cm

11 °C

On

Off

Off

Low

Perfect

Night

19 cm

6 °C

Off

Off

On

Perfect

Low

Afternoon

25 cm

8 °C

On

Off

On

Low

Low

Noon

18 cm

14 °C

Off

Off

Off

Perfect

Perfect

*WL: Water level; WT: Water temperature

References 1. Akbarian, A., Michiels, J., Degroote, J., Majdeddin, M., Golian, A., De Smet, S.: Association between heat stress and oxidative stress in poultry; mitochondrial dysfunction and dietary interventions with phytochemicals. J. Anim. Sci. Biotechnol. 7(1), 1–14 (2016) 2. Abdel-Moneim, A., et al.: Nutritional manipulation to combat heat stress in poultry–a comprehensive review. J. Therm. Biol 98, 102915 (2021) 3. Farghly, M., et al.: Implementation of different feed withdrawal times and water temperatures in managing turkeys during heat stress. Poult. Sci. 97, 3076–3084 (2018) 4. Jones, F., Watkins, S.: How Does Taste Influence Water Consumption in Broilers? www.the poultrysite.com (2009) 5. Fairchild, B., Ritz, C.: Poultry Drinking Water Primer. University of Georgia (2009) 6. Tjoa, G., Aribowo, A., Putra, A.: Design of automatic drinking water supply system for poultry cage. In: 2019 5th International Conference on New Media Studies (CONMEDIA) (2019) 7. Sitaram, K., Ankush, K., Anant, K., Raghunath, B.: IoT based smart management of poultry farm and electricity generation. In: 2018 IEEE International Conference on Computational Intelligence and Computing Research (ICCIC) (2018) 8. Ramadiani, R., Widada, D., Widiastuti, M., Jundillah, M.: Temperature and humidity control system for broiler chicken coops. Indones. J. Electr. Eng. Comput. Sci.22, 1327–1333 (2021) 9. Batuto, A., Dejeron, T., Cruz, P., Samonte, M.: e-Poultry: an IoT poultry management system for small farms. In: 2020 IEEE 7th International Conference on Industrial Engineering and Applications (ICIEA) (2020) 10. Arvindan, A., Keerthika, D.: Experimental investigation of remote control via Android smart phone of Arduino-based automated irrigation system using moisture sensor. In: 2016 3rd International Conference on Electrical Energy Systems (ICEES) (2016) 11. Wang, Y., Chi, Z.: System of wireless temperature and humidity monitoring based on Arduino Uno platform. In: 2016 Sixth International Conference on Instrumentation and Measurement, Computer, Communication and Control (IMCCC) (2016) 12. Nagendra Reddy, P., Kumar Reddy, K., Kumar Reddy, P., Kodanda Ramaiah, G., Kishor, S.: An IoT based home automation using android application. In: 2016 International Conference on Signal Processing, Communication, Power and Embedded System (SCOPES) (2016)

168

A. Y. Mohammed et al.

13. Abd Karim, N.: Urbanized chicken coop monitoring system using IoT. Int. J. Adv. Trends Comput. Sci. Eng. 8, 450–454 (2019) 14. Jouda, A., Sagheer, A., Shuwandy, M.: MagRing-SASB: Static Authentication of Magnetism Sensor Using Semi-Biometric Interaction Magnetic Ring. 2021 IEEE 11th International Conference on System Engineering and Technology (ICSET). (2021) 15. Shuwandy, M., Zaidan, B., Zaidan, A.: Novel authentication of blowing voiceless password for android smartphones using a microphone sensor. Multimedia Tools and Applications. (2022)

Offline Marker-Less Augmented Reality Application for Exploring Threatened Historical Places Hasan Badir Althewaynee(B) , Maytham M. Hamood, and Harith A. Hussein College of Computer and Math Sciences, Tikrit University, Tikrit, Iraq [email protected]

Abstract. Integrating technology such as augmented reality into the tourism industry has led to e-tourism and smart tourism development. In the most recent few years, there has been observed a meteoric rise in the usage of mobile devices that can swiftly retrieve information has been noticed. As a result, Mobile augmented reality applications have proven globally successful in serving different fields like historical site tourism. Despite this, tourism at historical sites in Iraq continues to suffer from different problems, including a decline in popularity as well as a lack of use of modern technologies to document, preserve, and display sites to attract visitors. This paper aims to develop an offline mobile application for exploring Iraq’s most important tourism site. Based on the unity3D engine and ARcore SDK, Marker-less augmented reality application was created to display the minaret and the Great Mosque of Samara. The application functionality was tested, and it was found that it works well and has good interactivity and ease of use, and can provide new solutions for tourism of the historical sites in Iraq. Keywords: Augmented reality · Cultural heritage · Tourism · Marker-less · Historical places · Mobile application

1 Introduction Tourism encompasses a wide range of actions, services, and industries that people can do, including using modes of transportation, lodging and enjoyment, visiting sports centers, restaurants, stores, historical and cultural sites, and so on [1]. For most nations, historical and heritage tourism is crucial [2]. In general, historical tourism is dependent on historical sites that are special, old, and present in a variety of nations and are constantly an attraction to tourists [3]. Throughout history, the tourism sector in Iraq has been plagued by indifference, vandalism, and extinction for a variety of reasons, including the overall security situation, failure to implement policies, and a lack of investment in infrastructure and essential services, even though Iraq has vast tourist and cultural potential, as well as holy sites for various religions dating back thousands of years [4]. For more than 25 years, the tourism sector has been significantly impacted by information and communication technology (ICT) [5]. Tourists, investors, companies, © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 169–180, 2023. https://doi.org/10.1007/978-3-031-20429-6_17

170

H. B. Althewaynee et al.

and countries that own parks, monuments, and other cultural sites have turned to etourism and smart tourism as a distinct and growing field that uses various technologies to serve tourism [6]. Augmented reality (AR) is one of these technologies that are adopted in tourism [7]. AR allows virtual elements to be projected in real-time onto a real-world environment, enhancing the real-world experience by adding those objects and giving human interaction a new twist [8]. Virtual objects provide information that assists a user’s real-time interest. In recent years, the employment of such technology has served as a link between the digital and real worlds [9]. In recent years, Mobile device usage has been rapidly growing [10]. With the use of a mobile device such as a phone or tablet, augmented reality technology allows you to add text, pictures, animations, and other information to the camera view of these devices. This method communicates the information to the users interactively [11]. The basic reason for highlighting tourism of historical sites in augmented reality technology is the engagement between real-world conditions and virtual buildings [12]. Samarra Archaeological City is one of the most important historic sites consisting of the great cultural heritage of the mosque and its minaret [13], which is called (Moulouya); it was created by the caliph Al-Mu’tasim Billah ibn Harun Al-Rashid in 221H/836AD [14]. It is located 125 km north of Baghdad/Iraq, on both sides of the Tigris River [15]. From 2007 to the present, the city has been on UNESCO World Heritage Sites (WHS) in danger and prevalence of threats [16]. It is threatened with gradual collapse due to the neglect of those concerned with providing the necessary maintenance of Societal ignorance and military operations [17]. Additionally, governmental organizations ignore to use of cutting-edge technology that promotes tourism and attracts tourists to such locations [18]. Due to these fears, the main objective of our study is to shed light on historical and cultural sites in Iraq which is at risk by developing a Mobile Augmented Reality app to help internal and international tourists learn more about Samarra great mosque, which is difficult to access due to security conditions. The next section of the paper includes a review of the literature that concerned the development of augmented reality systems for tourism of historical and archaeological sites. The rest of the paper includes the methodology, implementation and testing of the proposed application.

2 Related Work In tourism of historical and archaeological areas, AR has been used to meet actual needs such as (revival of the archaeological regions, restoration or reconstruction by default, study and definition, attraction and guidance). Due to worries over the destruction and damage to globally recognized heritage assets in Afghanistan, Syria, Iraq, and Brazil, digital documenting and the protection of historical and cultural monuments have recently gained international attention [19]. Thus [20] developed a system that provides locationbased querying in historic multimedia collections and adds an augmented reality-based user interface that enables the overlay of historical images and the current view without showing any 3D models or interactive objects. Also, [21] took care of reviving the heritage related to the Roman Theater at Byblos and displayed only missing parts of it when the user was on site [22]. Develop AR App to assist in navigating the old Famosa

Offline Marker-Less Augmented Reality Application

171

fortress site in Melaka, Malaysia, but without user interface and interaction. Also, [12] Proposed a method for developing an AR system that shows the Portuguese Malacca, but a limitation is that it lacked a practical application. It was interested in simulating a building and paying attention to how light affected it. In [23] and [24], They developed a geo-location-based AR application to make the Roman Bath of Ankara/turkey a much more attractive place to attract more visitors. But with the limitations of needing to use a GPS and no interaction with the 3d model, there is no user interface. In [25] They designed an app to explore augmented reality influence on identifying historical sites in Medellin’s Cisnero Square in Colombia, but it requires GPS and the tourist’s presence on site. So, with the restrictions above, A new method that can present several objects simultaneously, detect flat surfaces for a much more realistic scenario, and present characteristics that the old way of giving 3d objects in augmented reality couldn’t achieve is needed [26]. So we suggest using the offline and marker-less AR method, which eliminates markers that degrade the reality of landscapes and the stability of AR systems, free users from seeking markers, tracking images or text, and overcome the need for internet or GPS information.

3 Methodology This section describes the process of developing an android AR mobile app named ‘SamAR’, which works based on marker-less AR type. We employ waterfall as a conventional model of the System Development Life Cycle method. The waterfall approach has clear and simplified steps, easy to apply and understand, as well as suitable for simple projects that need to be understood by the user [27]. As a result, future systems that support historical site exploration by AR can readily adopt it. Figure 1 show system waterfall model.

Data Collection Software analysis Software design Implementation Testing

Fig. 1. Research methodology

3.1 Data Collection The data collection process used in this study consisted of collecting data about the Great Mosque of Samarra and its minaret. Looking for pieces of information, images, or other

172

H. B. Althewaynee et al.

materials that might help the application or system fulfill its goals. Mobile augmented reality application was also created using a literature review methodology including books, the internet, journals or conferences, and other sources that are then cited and discussed in the report. 3.2 Software Analysis It is a stage that includes defining the requirements for establishing and operating the system, the software and Hardware materials used and the limitations of the system’s work. It was divided as follows. • Software/Hardware requirements Object Visualization based on the AR app uses Unity version 2020 software. Unity’s multiplatform game engine allows multiple extensions and scripts to be integrated without consuming many resources or memory while creating the final mobile app. Unity Technologies was made and launched in the year 2005 [28]. Three-dimensional, twodimensional, virtual reality, and augmented reality games, as well as simulations and other experiences, can be made with this engine. Unity has a core C# scripting API. The ARcore SDK for the Unity engine was chosen to work well in Ground plane detection [26]. Table 1 lists the requirements. Table 1. Software and hardware requirements Software/hardware

Extensions

Unity 2020.3.30f1

ARfoundation 4.1.9 -Support forEditor-2020.3.30f1 ARcore XR Plugin 4.1.9 Jet Brains Rider Editor 2.0.7 Timeline 1.4.8 Unity UI 1.0.0 Canvas Main menu Script

ARcore

AR session AR Camera Plane Detection Script

Visual Studio 2019

Visual Studio 2019 for Unity

Sketch up Pro 2020

Default

Laptop dell 851DNL3

Windows 10 OS

Mobile ( Galaxy A30)

Android OS 11

Offline Marker-Less Augmented Reality Application

173

• Construction 3d model Generating 3D models with excellent fidelity to the actual old buildings is highly crucial in cultural heritage-based AR applications [29]. The model was created by Sketch up 2020. The construction of the 3d model took place in two stages. The first was the design of the spiral cylindrical shape, consisting of a square base and an external ladder that gradually decreases as it rises towards the top, as shown in Fig. 2. The second is to design the unfinished land of the mosque, which is in the form of a rectangle.

Fig. 2. 3d model parts

Textures must be supplied to the model after the design process to make it more lifelike. Three distinct textures for the main section, marble areas, and walls have been created for this purpose Figs. 3 and 4. After all these processes, the generated model can be seen in Fig. 5.

Fig. 3. Textures

• Software Limitations The suggested system has some limitations, such as: 1. System intended to operate on Android 7 Nougat operating system and up.

174

H. B. Althewaynee et al.

Fig. 4. 3d model after adding textures

Fig. 5. Completely 3D model

2. A system utilizes the ARcore library, a software development kit for augmented reality that Google supports on some devices but not others [30]. Any user can find out if his\her mobile device supports operating in an augmented reality environment through the official Google website. 3.3 Software Design Based on analysis details in the previous phase. Figure 6 is a Flowchart of the System design. The application creates an augmented reality environment by turning on the camera and setting it up to explore a suitable ground. After determining the horizontal surface, the three-dimensional model of the Great Mosque of Samarra will be displayed with other digital information, such as audio and text. Based on Fig. 7, 7 processes occur in the application, including starting the application, starting augmented reality, ground detection, show 3d model, reading the information, listening to audio descriptions and quitting the application. The next step is designing an activity diagram to show system activity and functionality more in-depth. Figure 8 show an activity diagram of the proposed application. After opening the application, the user will see the main menu window through which he will choose to open the camera, which leads to ground detection. After that, digital contents related to the Great Mosque of Samarra will be displayed, such as a three-dimensional model, descriptive sounds and explanatory texts.

Offline Marker-Less Augmented Reality Application

175

Start

Prepare camera for detecting process No

Scanning for ground plane

Are Ground detected Successfully?

Rendering and displaying

Yes

3d model

Reading details on points and Listening on sound description

End

Fig. 6. System design flowchart

Fig. 7. Use case diagram

3.4 User Interface Designing The interface was made with Canvas which included all contents. The virtual object was added to the Unity canvas as well as scripts written in C# that were associated with four buttons in UI, three POI interfaces in the 3d model, zoom in/out buttons, sound

176

H. B. Althewaynee et al.

Fig. 8. Activity diagram

playing/pause button and most important script that responsible for surface detection consisting camera ray estimation.

4 Implementation The proposed system workflow is shown in Fig. 9. The application is directed to work on Android devices, where the features of Unity and ARcore have been harnessed to complete the task. ARcore was chosen because it can increase its awareness about the area if the device’s location changes over time [31]. At the start of the application, the user will deal with a simple interface, as shown in Fig. 10-a. The interface will include four buttons. The SamAR application will ask the user for permission to use the camera. After giving permission and pressing the first button, the camera will open, and the ARcore features for detecting the horizontal surface will start working. The camera starts shooting a ray to determine the surface, and as soon as the surface is identified, a blue circle will appear to inform of the discovery of the surface. Then select and press the object button at the bottom of the interface, and the 3d model will be displayed in Fig. 10-b. The user can deal with the 3d model in terms of zooming in/out by using the buttons on the left of the user interface, and the visitor can also rotate around the model and see it more closely. Three red dots will also appear on the 3d model. When clicking on it, the user will be directed to a simple interface that explains the most prominent information related to that position. The audio explanation button was added to the right of the user interface to add some attraction to the method of the application’s work and the user’s interest. The second button in the main menu navigates the user to a simple text interface with an explanation of essential historical facts concerning the Great Mosque of Samarra and its tower Fig. 10-c. The user navigates to education instructions on how to run the application to make dealing with it more accessible, as well as some information about the developer and the environment in which it was developed, through the last button in the main menu Fig. 10-d.

Offline Marker-Less Augmented Reality Application

177

Fig. 9. System workflow

a. main menu

b. rendering

c. historical facts

d. read me page

Fig. 10. User interface

5 Testing Black Box testing is used for application functionality testing. It is one of the simplest testing techniques for software [32]. Black box testing only requires a lower and upper limit from hoped data. Samsung A30, Samsung A52, Samsung A72, and Xiaomi Redmi

178

H. B. Althewaynee et al.

Note 9 Pro are all used in the testing process. The information in Table 2 is derived from the testing results. Table 2. Black box testing results Function

Result

Status

Installing application

Installation is done

Success

Application running

Application running normally

Success

Open camera button

The button works and opens the camera

Success

Ground detection

Showing a 3D model

Success

Play audio Icon

Icon plays an audio explanation

Success

Pause audio Icon

Icon pauses an audio explanation

Success

Main menu button

Returns to the main menu

Success

Historical facts button

Text information interface appears

Success

Read me button

Text information interface appears

Success

Quit button

Exits the application

Success

6 Conclusion and Future Work Iraq has a long history of diverse civilizations and cultures. According to UNESCO, the Samarra Mosque and its minaret are among the areas under threat and at risk. Quick and new solutions must apply to Preserve, document, and present this cultural heritage and make it easily accessible to tourists and researchers who care. So we created Marker-less Mobile augmented reality application named “SamAR” that displays through a simple user interface a 3D model of the Great Mosque of Samarra in addition to text information and audio explanation. The stages of development as well as the tools employed were examined. The app was also designed to run without an Internet connection or GPS to overcome needing for a user to be onsite. The application was tested based on the Blackbox method. The test results showed that the application works ideally and without any errors. For future work, we seek to include other Iraqi historical places which are at risk and threatened in the UNESCO list.

References 1. Slak Valek, N.: Art tourism: definitions, opportunities, and discussions based on a case study from Abu Dhabi. J. Tour. Cult. Chang. 20(1–2), 100–114 (2022) 2. Khurramovich, K.A.: Main methods of reconstruction of the tourism system of historical city centers. Eur. J. Life Saf. Stabil. (2660–9630) 13, 82–86 3. Wei, L., Ullah, S.: International tourism, digital infrastructure, and CO2 emissions: fresh evidence from panel quantile regression approach. Environ. Sci. Pollut. Res. 29(24), 36273– 36280 (2022)

Offline Marker-Less Augmented Reality Application

179

4. Nawar, A.: Insights into the main difficulties of achieving sustainable development of tourism in Iraq. Ann. Univ. Oradea Geog. Ser./Analele Universitatii din Oradea, Seria Geografie 24(1), 32–43 (2014) 5. Condratov, I.: E-tourism: concept and evolution. Ecoforum J. 2(1), 10 (2013) 6. Xu, F., Buhalis, D., Weber, J.: Serious games and the gamification of tourism. Tour. Manag. 60, 244–256 (2017) 7. Jiang, S., et al.: Augmented reality and the enhancement of memorable tourism experiences at heritage sites. Curr. Iss. Tour., 1–16 (2022) 8. Mekni, M., Lemieux, A.: Augmented reality: applications, challenges and future trends. Appl. Comput. Sci. 20, 205–214 (2014) 9. Arena, F., et al.: An overview of augmented reality. Computers 11(2), 28 (2022) 10. Rashid, R.A., Mohamed, H., Hussin, A.R.C.: Mobile augmented reality tourism application framework. In: Saeed, F., Gazem, N., Patnaik, S., Saed Balaid, A.S., Mohammed, F. (eds.) IRICT 2017. LNDECT, vol. 5, pp. 108–115. Springer, Cham (2018). https://doi.org/10.1007/ 978-3-319-59427-9_12 11. Demir, Ö.F., Karaarslan, E.: Augmented reality application for smart tourism: GökovAR. In: 2018 6th International Istanbul Smart Grids and Cities Congress and Fair (ICSG). Ieee (2018) 12. Kolivand, H., et al.: ReVitAge: realistic virtual heritage taking shadows and sky illumination into account. J. Cult. Herit. 32, 166–175 (2018) 13. amgapn, G.A.: Cultural Resources Of Iraqi Heritage In The Context Of The Development Of Iraqi Tourism And Economy. in Cbopniki konfepenci HIC Cociocfepa. Vedecko vydavatelske centrum Sociosfera-CZ sro (2016) 14. Kettana, L.J.A.-L.: A Case Study in the History of Islamic Towns: Spatial Patterns of Historic Samarra in the Reign of Abbasid Residence at Samarra (221-279 AH/835-892 AD). The University of Manchester (United Kingdom) (1979) 15. Salim, I.H., Khaleel, R.I., Al-Hilali, B.M.I.: Evaluation of subsurface soils for selected sites in Samarra city-Iraq. Appl. Soil. Ecol. 175(5), 1–13 (2022) 16. Birendra, K.: A comprehensive analysis of threats to UNESCO WHSs in danger. Ann. Tourism Res. Empir. Insights 2(1), 100013 (2021) 17. Alsadik, B.: Crowdsource drone imagery–a powerful source for the 3D documentation of cultural heritage at risk. Int. J. Archit. Herit., 1–11 (2021) 18. Al Chalabi, A.S.: The Extent of Application of E-Rourism Marketing Technology in the Website of the Iraqi Tourism Board 19. Rahaman, H., Champion, E., Bekele, M.: From photo to 3D to mixed reality: a complete workflow for cultural heritage visualisation and experience. Dig. Appl. Archaeol. Cult. Herit. 13, e00102 (2019) 20. Sauter, L., Rossetto, L., Schuldt, H.: Exploring cultural heritage in augmented reality with GoFind!. In: 2018 IEEE International Conference on Artificial Intelligence and Virtual Reality (AIVR). IEEE (2018) 21. Younes, G., et al.: Virtual and augmented reality for rich interaction with cultural heritage sites: a case study from the Roman Theater at Byblos. Dig. Appl. Archaeol. Cult. Herit. 5, 1–9 (2017) 22. Xin, S., et al.: Application of 3D tracking and registration in exhibition hall navigation interaction. In 2020 International Conference on Intelligent Computing, Automation and Systems (ICICAS). IEEE (2020) 23. Unal, M., et al.: Geo-location based augmented reality application for cultural heritage using drones. In: 2018 2nd International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT). IEEE (2018) 24. Meriem, K., Makram, M., Farah, I.R.: Virtual and augmented reality in the valuation of the Tunisian cultural heritage: application to thysdrus (ElJem) Amphitheater. In 2018 International Conference on Intelligent Systems (IS). IEEE (2018)

180

H. B. Althewaynee et al.

25. Hincapié, M., et al.: Augmented reality mobile apps for cultural heritage reactivation. Comput. Electr. Eng. 93, 107281 (2021) 26. Syahputra, M., Hardywantara, F., Andayani, U.: Augmented reality virtual house model using ARCore technology based on android. In: Journal of Physics: Conference Series. IOP Publishing (2020) 27. Chandra, V.: Comparison between various software development methodologies. Int. J. Comput. Appl. 131(9), 7–10 (2015) 28. Haas, J.K.: A history of the unity game engine. Diss. Worcest. Polytech. Inst. 483, 484 (2014) 29. El-Hakim, S.F., et al.: Detailed 3D reconstruction of large-scale heritage sites with integrated techniques. IEEE Comput. Gr. Appl. 24(3), 21–29 (2004) 30. Alkandari, A., et al.: Google project tango and ARCore under the view of augmented reality. J. Comput. Theor. Nanosci. 16(3), 1127–1133 (2019) 31. Oufqir, Z., El Abderrahmani, A., Satori, K.: ARKit and ARCore in serve to augmented reality. In: 2020 International Conference on Intelligent Systems and Computer Vision (ISCV). IEEE (2020) 32. Nidhra, S., Dondeti, J.: Black box and white box testing techniques-a literature review. Int. J. Embedd. Syst. Appl. (IJESA) 2(2), 29–50 (2012)

A Robust Tuned K-Nearest Neighbours Classifier for Software Defect Prediction Abdullah B. Nasser1(B) , Waheed Ghanem2 , Antar Shaddad Hamed Abdul-Qawy3 , Mohammed A. H. Ali4 , Abdul-Malik Saad5 , Sanaa A. A. Ghaleb6 , and Nayef Alduais7 1 School of Technology and Innovation, University of Vaasa, 65200 Vaasa, Finland

[email protected] 2 Faculty of Ocean Engineering Technology and Informatics, Universiti Malaysia Terengganu,

Kuala Terengganu, Malaysia [email protected] 3 Faculty of Science, Department of Science and Information Technology, SUMAIT University Zanzibar, Zanzibar, Tanzania [email protected] 4 Faculty of Engineering, Department of Mechanical Engineering, University of Malaya, Kuala Lumpur, Malaysia [email protected] 5 Division of Electronic and Computer Engineering, School of Electrical Engineering, Faculty of Engineering, Universiti Teknologi Malaysia, 81310 Johor Bahru, Malaysia [email protected] 6 Faculty of Informatics and Computing, Universiti Sultan Zainal Abidin, Kuala Terengganu, Malaysia [email protected] 7 Faculty of Computer Science and Information Technology, Universiti Tun Hussein Onn Malaysia, Johor, Malaysia [email protected] Abstract. If the software fails to perform its function, serious consequences may result. Software defect prediction is one of the most useful tasks in the Software Development Life Cycle (SDLC) process where it can determine which modules of the software are prone to defects and need to be tested. Owing to its efficiency, machine learning techniques are growing rapidly in software defect prediction. K-Nearest Neighbors (KNN) classifier, a supervised classification technique, has been widely used for this problem. The number of neighbors, which measure by calculating the distance between the new data and its neighbors, has a significant impact on KNN performance. Therefore, the KNN’s classifier will perform better if the k hyperparameters are properly tuned and the independent inputs are rescaled. In order to improve the performance of KNN, this paper aims to presents a robust tuned machine learning approach based on K-Nearest Neighbors classifier for software defect prediction, called Robust-Tuned-KNN(RT-KNN). The RT-KNN aims to address the two abovementioned problems by (1) tuning KNN and finding the optimal value for k in both the training and testing phases that can lead to good prediction results, and (2) using the Robust scaler to rescale the different independent inputs. The experiment results demonstrate that RT-KNN is able to give sufficiently competitive results compared with original KNN and other existing works.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 181–193, 2023. https://doi.org/10.1007/978-3-031-20429-6_18

182

A. B. Nasser et al. Keywords: Software defect prediction · Machine learning · K-nearest neighbors classifier · Hyper-parameters Tuning

1 Introduction Software development has been growing rapidly, and it’s getting more complex in terms of software size and functionalities. As a result, software reliability is an effect, which plays a critical role in the Software Development Life Cycle (SDLC) process. Catching software defects is one of the most difficult tasks in the software development life cycle. As a result of that, many software testing approaches [1, 2] have been proposed to reduce software defects, such as static testing, which includes reviewing and analysing the software’s documents in order to prevent any software defects, while dynamic testing focuses on examining the software’s behaviour to identify any software defects. Machine learning is a well-known technology that is used in many fields. Machine learning’s popularity stems from its advantages in learning the machine from historical data and making predictions on future data. Machine learning in software defect prediction or software testing, in general, is still quite new, However, similar to other areas of artificial intelligence, it is growing rapidly. The code is a small unit in any application or program in which each portion of code is designed to do a certain task. In most cases, these codes contain some defects, which may result in a consequence disaster such as losing data, losing fortune or even losing lives. Software defect prediction is one of the most useful tasks in the SDLC’s where it can determine which modules are prone to defect and need to be tested. It also allows the test manager to use the resources effectively while still adhering to the limits. The purpose of software defect prediction includes classifying the software components into not defect-prone and defect-prone classes. It is also used for identifying any associations between defects. Another purpose of software defect prediction is to estimate the number of faults remaining in software systems. K-Nearest Neighbors (KNN) classifier, a supervised classification technique, has been widely used for this problem. A set of parameters, including the numbers of neighbors k, in the KNN classifier, must be tuned. The performance of KNN is highly dependent on the number of neighbors and the problem at hand. Therefore, the classifier will perform better if these hyperparameters are properly tuned [3]. Another related issue in KNN is that KNN is a distance-based method and the independent inputs or features of the dataset often come in different scale/units therefore calculating the distance will not functions at its best due to some outliers in the dataset [4]. Therefore, this research presents a robust tuned machine learning approach based on K-Nearest Neighbors classifier for software defect prediction, called Robust-TunedKNN(RT-KNN). The RT-KNN aims to address the two abovementioned problems by (1) tuning KNN to select the optimal value for k in both the training and testing phases that can lead to good prediction results and (2) using the robust scaler to rescale the different independent inputs. First, to find the optimum values of hyperparameters, which are the initial values of the model’s parameters that are set by the data analysts before training the model, we investigate the number of neighbors manually and using Grid Search. In

A Robust Tuned K-Nearest Neighbours Classifier

183

the tuning process, the number of neighbors is varied while the other parameters remain constant based on the recommended values. Then compare their results and find the best number of neighbors. Then, the optimal results of the hyperparameters obtained are used in the second step which uses the robust scaler to scale the model’s features. The rest of the paper is organized as follows: Sect. 2 explains how the K-Nearest Neighbors Algorithm works, Sect. 3 reviews the related works, while Sect. 3 presents the methodology of RT-KNN for software defect prediction. The results and discussion are presented in Sect. 4, and finally, the conclusion is in Sect. 5.

2 The k-Nearest Neighbors Algorithm The K-Nearest Neighbors Algorithm (KNN) is a supervised learning approach proposed by Evelyn Fix and Joseph Hodges [5]. It is used for both the classification and regression of data. In both cases, the nearest k neighbors are used for testing the models. The idea behind KNN is a very simple algorithm. The algorithm defines k nearest neighbors of the new data, then it is classified based on a majority vote of its neighbors. Suppose we have two classes, A with a blue circle and B with a green triangle, and we want to classify the red pentagon (it is called the test vector), as shown in Fig. 1. The algorithm calculates the distance between the test vector and k neighbors. Suppose k = 4, then the algorithm will calculate the distance between the test vector and its four neighbors, then the test vector is classified (also called labelled) as class A, since there are 3 vectors out of the 4 nearest neighbor vectors from class A and only 1 from class B.

Fig. 1. Illustration of KNN’s classification

3 Related Works In general, software defect prediction can be classified into two categories: manual classification methods and automatic classification methods. In manual classification methods, analysts utilize their own experience to classify software defects into distinct groups [6]. The analyst begins by establishing the category of defects. Then, based on

184

A. B. Nasser et al.

their experience, they locate the defect in the matching category. However, this method of classification is difficult and requires a huge number of human resources [7]. In 1992, IBM introduced Orthogonal Defect Classification (ODC). ODC is a manual classification tool that extracts some information from the software to provide insight and diagnostics. ODC defines eight defect attributes in its newest version [8], and it employs several attributes to describe defect features. Software defects are classified into eight categories based on the characteristics of these attributes, such as function, verification, algorithm, assignment, timing, association, interface, and documentation. Similar to ODC, IEEE introduces IEEE standard classification for Anomalies. The classification using IEEE stander is based on user experience. Therefore, engineers must first classify the anomalies as defects or failures and then use the standard. Another manual classification is called Thayer classification where the tester fills out error reports during testing and comments from users to categorise errors according to their nature [6]: calculation errors, logic errors, I/O errors, data processing errors, operating system’s errors, and so on. In literature, there are many works similar to ODC, IEEE standard classification, and Thayer classification such as Roger classification, Putnam classification and Michael classification, to name a few. On the other hand, automatic classification methods use algorithms such as machine learning to classify the defects. In the literature, several machine learning approaches have been proposed to address the software defect prediction problem, such as logistic regression [9], neural network [10], support vector machines [11], and random forests [12]. Based on the information extracted from ODC, LiGuo Huang et al. proposed new classification methods using a support vector machine (SVM) called ODC automatic classification. Before using the SVM classifier, the method divides the defects manually into six classes and then creates vectors to be used in SVM [11]. Close to ODC automatic classification method, Thung et al. proposed another automatic classification based on SVM. However, they believe that the reported bugs and the source code are very important for the classification process. Xia Xin et al. proposed an automatic defect classification model that uses fuzzy set feature selection and a 10-fold cross-validation method. The method starts extracting the information from the bug report, then extracting the subset of features then evaluating the model using cross-validation. Another automatic classification based on trigger mode was proposed by Xia Xin et al. By analyzing the code, the method classifies the defect into Bohrbug, which is a simple defect and easy to isolate, or Mandelbug defect which is more complex than the Bohrbug defect. The method uses a fuzzy set to select the important features that can help to classify the defects [6].

4 Robust Tuned K-Nearest Neighbors Classifier for Software Defect Prediction This section describes the methodology of Robust Tuned K-Nearest Neighbors Classifier for Software Defect (RT-KNN) prediction including describing and analyzing the dataset and describing the steps that use in tuning and rescaling K-Nearest Neighbors Classifier.

A Robust Tuned K-Nearest Neighbours Classifier

185

A. Dataset The JM1 dataset was used as the basis for our research. The dataset is one of NASA Metrics Data Program which is real-time predictive ground system. The dataset consists of 10885 instants, each instant represents a module, and 22 attributes including the target attribute that label the software as DEFECTED class 1 or UNDEFEATED or CLEAN class 0, as Fig. 2 shows. The dataset contains some statistical measurements of the software such as count of lines (LOC), McCabe Cyclomatic Complexity v(G), McCabe Essential Complexity ev(G), McCabe Design Complexity iv(G) and some Halstead’s Complexity Measures such as Halstead total operators (n) Halstead “volume” (v), Halstead “program length” (l), Halstead “difficulty” (d), Halstead “intelligence” (i) so on. More information about the dataset can be found in [13].

Fig. 2. JM1 dataset information

The classification of JM1 is challenging because there are many modules with the same identical attribute values having different defect labels [14]. The correlation between the defect class and the other features of the modules is not strong as shown in Fig. 3. The correlation in the figure is scaled from 0 to 1. The last column and row of the figure, which present the correlation between the defect classes and other features,

186

A. B. Nasser et al.

are low for all features. This issue may spur research on selecting the best combinations of features for better classification.

Fig. 3. Correlation between the features of JM1 dataset

The dataset also has another issue which is unbalancing since it contains 8779 defected modules and 2106 undefeated modules. There is also overlapping between the two classes as the pairwise relations of the attributes of the dataset are shown in Fig. 4. B. Method and implementation A good correlation between the independent variables and the target variables is a good sign for the prediction. The performance of KNN is highly dependent on the number of neighbors. Selecting the optimal value for k in both the training and testing phases can lead to good prediction results. In KNN, The training phase is just sorting the training data thus the main computation of the algorithm is at the time of prediction rather than at the time of fitting. In the prediction phase, the algorithm finds the k neighbors of the new point by calculating the distance between the new point and the neighbors, and uses these values to classify the new point. Therefore, the accuracy of the model is highly dependent on the number of neighbors. Additionally, due to the nature of and being a distance-based method therefore outliers’ values in the training data may affect the performance of the method, consequently, it is necessary to reduce the effect of the outliers by rescaling the training data. Accordingly, this section aims to propose a robust

A Robust Tuned K-Nearest Neighbours Classifier

Fig. 4. The pairwise relations of the attributes of the dataset

187

188

A. B. Nasser et al.

tuned machine learning approach based on K-Nearest Neighbors (RT-TKK) classifier for software defect prediction. The RT-KNN model has been implemented from the screech using python. Following are the main steps of the implementation: 1. Reading and Pre-processing the dataset: The JM1 dataset is used in this work. The data in the dataset is clean and does not require preprocessing. 2. The dataset is split into training (80%) and testing (20%) subsets. 3. In order to find the optimal value of k, the value of k is varied from 1 to 25 based on the recommendation values. Since the goal is to find the optimal value of k, the other parameters have been fixed. 4. The KNN model is trained and tested. The accuracy results of both training and testing are recorded. We can simply select the optimal value of k based on the testing accuracy results, however, we prefer to use another method to select the optimal value based on both training and testing results. Here the Pareto frontier method, which is multi-objective optimization, is used. Pareto frontier takes two equally-sized lists (training and testing accuracy) and returns the line of points (a.k.a Pareto curve) that is fitted to the points, in order. 5. The optimal value of returning k by Pareto frontier which is the efficient points of both the training and testing phase is used for prediction. 6. For comparison purposes, the GridSearchCV is used also for finding the optimal values, and then the results of our method and GridSearchCV are compared. 7. Steps 3 through 5 are repeated, but this time after rescaling the dataset using the Robust Scaler. Robust Scaler is useful for removing outliers. In case some input variables have relatively big values compared to other input variables, these big values, may affect the accuracy of the model, are rescaled using the robust scaler. To calculate the robust scale, the value is subtracted from the median, and then it is divided by the interquartile range (75% -25%) value. As a result, the scaler focuses on the large values while ignoring the variables with lesser values.

5 Results and Discussion As mentioned earlier, the JM1 dataset was used in the experiments. The prediction accuracy is used to assess the performance of the tuning model. The prediction accuracy can be calculated using the formula: Accuracy =

number of correct predicted test modules the total number of test modules

The following experiments have been carried out in order to assess the accuracy of RT-KNN: Experiment 1: Tuning the K Value manually and using GirdSerachCV In this experiment, the number of neighbors is varied from 1 to 25, while the other parameters of KNN are fixed. Figure 5 shows the accuracy scores of both the training and testing phases.

A Robust Tuned K-Nearest Neighbours Classifier

189

The figure shows that when the number of neighbors is less than SIX (6), the accuracy of training data is high while the accuracy of test data is low which is called underfitting, and when the number of neighbors is greater than or equal to SIX (6), the model obtains good and stable results. Even though it is recommended to use an odd number for k, the best point that got the best results for both training and testing is when k = 22.

Fig. 5. Train and test accuracy for KNN manually tuning

In the second part of this experiment, the KNN is tuned using GirdSerachCV. Similar to manual tuning, the value of k is varied from 1 to 25, and the other parameters have been fixed. Figure 6 shows the accuracy scores obtained using GridSearchCV. The results show that overfitting when the number of neighbors is small then as expected the results are getting better with a large number of neighbors.

Fig. 6. Tuning KNN using GridSearchCV

Experiment 2: Comparison of Manual and GridSearchCV tuning The third part of experiment 1 is comparing the manual and GridSearcCV. The results of comparing manual and GridSearchCV tuning are presented in Figs. 7 and 8. The figures

190

A. B. Nasser et al.

show the comparison of accuracy scores between manual and GridSearchCV tuning. The results show that the performance of the two models start to obtain good and stable results when the number of neighbors (k) is greater or equal to 12. The best accuracy in both models is obtained by manual tuning is when k equal to 22.

Fig. 7. Visual presentation of accuracy scores comparison between manual and GridSearchCV

Fig. 8. Accuracy scores comparison between manual and GridSearchCV

A Robust Tuned K-Nearest Neighbours Classifier

191

Experiment 3: Applying Robust Scaler on KNN As stated earlier, outlier values in the dataset may affect the performance of the KNN classifier. In this experiment, the dataset was rescaled using the robust scaler to lessen the influence of outliers. It is clear how much the effect of rescaling the dataset on accuracy; the accuracy scores have been improved significantly. In fact, other two scalers have been tested in this data set, which are MaxMin Scaler and standard Scaler, only the robust scaler achieves better performance than the others (see Fig. 9).

Fig. 9. Comparison of RT-KNN with GridSearchCV after applying robust scaler

Experiment 4: Comparison with Previous Works In this experiment, the RT-KNN results are compared with the best-obtained results from the existing work. The optimal k obtained from the previous experiments that applied the robust scaler is used for predicting the same test data. Only previously published works that use the same dataset and have the same number of features are considered. The results are collected from [7, 12, 15]. The comparison includes different types of classifiers, some are distance-based similar to KNN while the others are based on SVM and decision tree. As Fig. 10 shows, among the existing works, the tuning-KNN and Decision Tree (FR) achieved the highest accuracy.

6 Conclusion This paper presents a machine learning software defect prediction approach based on the K-Nearest Neighbors classifier, called a robust tuned K-Nearest Neighbors (RT-KNN). The method attempts to address two KNN problems, which are finding the optimal value of the number of neighbors and rescaling outliers in the dataset. Manual and using GridSearchCV are applied to find the optimal value of k, and the robust scaler is used to rescale the big values. Several experiments have been carried out in order to assess the accuracy of RT-KNN, including tuning the value of K manually and using GirdSerachCV and comparing their results, then applying the robust scaler and finally comparing the RS-KNN with some published results. The results of the experiment demonstrate that RT-KNN generates outcomes that are competitive with existing published works.

192

A. B. Nasser et al. 90 80 70

ACCURACY

60 50 40 30 20 10 0 KNN

KNN-K

SVM

L-SVM LS-SVM

LP

Decision Decision Decision Decision RT-KNN Tree Tree Tree (RF) Tree C4.5 CART ADT

CLASSIFIER

Fig. 10. Comparison of Tuning-KNN with the existing classifier

References 1. Pachouly, J., et al.: A systematic literature review on software defect prediction using artificial intelligence: Datasets, Data Validation Methods, Approaches, and Tools. Eng. Appl. Artif. Intell. 111, 104773 (2022) 2. Akimova, E.N., et al.: A survey on software defect prediction using deep learning. Mathematics 9(11), 1180 (2021) 3. Mabayoje, M.A., et al.: Parameter tuning in KNN for software defect prediction: an empirical analysis. Jurnal Teknologi dan Sistem Komputer 7(4), 121–126 (2019) 4. Yu, C., et al.: Indexing the distance: an efficient method to knn processing. In: Vldb (2001) 5. Fix, E., Hodges, J.L.: Discriminatory analysis. Nonparametric discrimination: Consistency properties. Int. Stati. Rev./Revue Internationale de Statistique 57(3), 238–247 (1989) 6. Gao, J., et al.: Research on software defect classification. In: 2019 IEEE 3rd Information Technology, Networking, Electronic and Automation Control Conference (ITNEC). IEEE (2019) 7. Lessmann, S., et al.: Benchmarking classification models for software defect prediction: a proposed framework and novel findings. IEEE Trans. Software Eng. 34(4), 485–496 (2008) 8. Chillarege, R.: Orthogonal defect classification. Handbook of Software Reliability Engineering, pp. 359–399 (1996) 9. Olague, H.M., et al.: Empirical validation of three software metrics suites to predict faultproneness of object-oriented classes developed using highly iterative or agile software development processes. IEEE Trans. Softw. Eng. 33(6), 402–419 (2007) 10. Arar, Ö.F., Ayan, K.: Software defect prediction using cost-sensitive neural network. Appl. Soft Comput. 33, 263–277 (2015) 11. Elish, K.O., Elish, M.O.: Predicting defect-prone software modules using support vector machines. J. Syst. Softw. 81(5), 649–660 (2008) 12. Guo, L., et al.: Robust prediction of fault-proneness by random forests. In: 15th International Symposium on Software Reliability Engineering. IEEE (2004)

A Robust Tuned K-Nearest Neighbours Classifier

193

13. Mike, C.: JM1/software defect prediction (2004). [22-June-2022]. https://www.openml.org/ d/1053 14. Zhong, S., Khoshgoftaar, T.M., Seliya, N.: Analyzing software measurement data with clustering techniques. IEEE Intell. Syst. 19(2), 20–27 (2004) 15. Cetiner, M., Sahingoz, O.K.: A comparative analysis for machine learning based software defect prediction systems. In: 2020 11th International Conference on Computing, Communication and Networking Technologies (ICCCNT). IEEE (2020)

Smart Virtual Robot Automation (SVRA)-Improving Supplier Transactional Processes in Enterprise Resource Planning (ERP) System: A Conceptual Framework Tiong Yew Tang(B) , Narishah Mohamed Salleh, and Mikkay Ei Leen Wong Department of Business Analytics, Sunway University, 5, Jalan University, 47500 Petaling Jaya, Bandar Sunway, Selangor, Malaysia {tiongyewt,narishahm,mikkayw}@Sunway.edu.my

Abstract. Robotic Process Automation (RPA) is a technique that assists in automating processes by employing numerous technologies for process automation which essentially helps in cost-saving, error-free, scalable and efficient solutions for real-world business problems. Albeit being highly relevant and beneficial, particularly for accountants who are responsible to make decisions in Enterprise Resource Planning (ERP) systems, the integration of RPA with ERP systems is still in the infant stages. Thus, with that in mind, this research proposes a novel Smart Virtual Robot Automation (SVRA) conceptual framework to explore the research design, development phase, and conceptual phase of the Robotic Process Automation (RPA) framework. The SVRA conceptual framework will be developed by employing the Selenium and Sikuli Framework tools for process automation. To execute automation instructions, the Python language script will be used along with the Selenium and Sikuli tools. Moreover, the SVRA development will be executed using the SAGE X3 ERP system. By utilizing the proposed SVRA conceptual framework design, it is anticipated that real-world ERP application problems in the Account Payable process in an organization can be resolved. Also, it is expected that this framework would contribute to the automation of the supervised deep learning approach. Keywords: Enterprise resource planning · ERP · Robotic process automation · RPA · Intelligent process automation · IPA · Deep learning · Machine learning · Framework

1 Introduction This research proposes a novel Smart Virtual Robot Automation (SVRA) conceptual framework to explore the research design, development phase, and conceptual phase of the Robotic Process Automation (RPA) framework. The development of the SVRA conceptual framework utilizing the Selenium1 and Sikuli2 [1] Framework tools for process automation. The Python language script is used with the Selenium and Sikuli tools 1 Selenium Framework, https://www.selenium.dev. 2 Sikuli Framework, http://www.sikulix.com.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 194–203, 2023. https://doi.org/10.1007/978-3-031-20429-6_19

Smart Virtual Robot Automation (SVRA)-Improving Supplier

195

to execute automation instructions. SVRA is a conceptual framework to describe the proposed deep learning [2] RPA development to automate the account’s human expert decisions on Enterprise Resource Planning (ERP) system. The proposed SVRA conceptual framework design is developed to resolve the realworld ERP application problems in the Account Payable process of an organization. The specific process focused in this work is the “Purchase Invoices Process” known as PIN. The PIN is the process of capturing the invoices from suppliers that are related to Purchase Order (PO) documents. The SVRA development will be done on the SAGE X3 ERP system. The primary aim of the development is to develop a digital co-worker that can run the PIN process in a virtual environment unattended and efficiently.

2 Literature Review In comparison to several decades ago, business organizations today have evolved significantly, and this is contributed by the rapid digital transformation. Most if not all business organizations are constantly improving their business process to ensure that they would be able to keep up with the world’s digital technological advancement. The terms automation and robotics are no longer peculiar terms within digital transformation. In 2012, the term Robotic Process Automation (RPA) was first coined by Blue Prim [3]. Since the inception of the term RPA, noticeably a considerable number of corporations have begun to shift their attention toward automation initiatives–to emphasize large cost savings. According to research conducted by Everest Group [4] and Horses for Sources Research [5], the demand for the international RPA market which constitutes both RPA software and RPA services has increased by a staggering 64% from the year 2016 to the year 2017. Along with that, as reported by HFS Research, there is a jump in the demand for RPA in the global market by 42% from the year 2017 to the year 2018 and as much as 94% improvement from the year 2018 to the year 2021 [5]. It is evident that over the past 10 years, RPA has gained significant traction and business acceptance in the global market and is still demonstrating a growing trend in the market [6–8]. Willcocks [9] defined RPA as a technique for automating processes utilizing various technologies for process automation, each of which is apt for various processes and objectives. RPA functions as a transition element between human work and extensive business process automation which helps organizations to save cost and time. The implementation of RPA, it enables users to reduce the repetitive tasks that require manual data entry and processing [2]. Typically, for an application to interact with another application, it requires to go through the API (Application Programming Interface) or the integration bus. On the other hand, the RPA framework simplifies this by allowing applications to interact with each other via the graphical user interface (GUI). RPA would be able to interact with different programs by emulating repetitive actions which are typically performed by humans. The manual user action simulation is advantageous as it does not require any modifications to the existing IT systems to utilize the RPA infrastructure [2]. Therefore, RPA does not change the IT solution itself; hence, its deployment time is short and cost-saving. If it is necessary to set another workflow, it is flexible enough to pause the virtual robot and continue the processing of the task for the employee. RPA significantly reduces costs and improves operational performance with minimal changes in the system.

196

T. Y. Tang et al.

Anagnoste [4] highlighted that in the industry, RPA is identified as the simplified form of Artificial Intelligence (AI). This form of simplified AI in RPA is designed to focus on ease of use, short implementation time, and cost-efficiency for improving business competitiveness. Irrefutably, the integration of AI in RPA is still in its infancy stage, thus, this study [10] intends to investigate the effectiveness of modelling human expert knowledge by employing business process analysis while extracting the knowledge about Decision-Making Logic (DML) in a form of taxonomy. The rationale for employing AI is due to the higher possibility of obtaining better text mining prediction simulation outcomes. Over the years, there have been various case studies conducted on the application of RPA, for instance, Business Process Outsourcing (BPO) [11]. Furthermore, based on the Capgemini study [12], it was pointed out that through the implementation of RPA, there is a noticeable improvement in process speed and compliance, a reduction in operating cost and of utmost importance, it reduces human error [13]. Despite all these pertinent contributions, it is imperative to note that the past studies focused on the BPO perspective instead of the effectiveness of modelling the human expert’s action. Therefore, this study intends to address this gap–to determine the effectiveness of modelling human experts’ actions by employing a novel deep learning approach. 2.1 Methodology Fit Within the accounting department, many routine processes are done to screen and perceive the company’s financial situation and prepare reports that enable managers to make decisions. The time intervals for reports vary based on the need and demand for those reports. One of the most repetitive routine processes is Account Payable. The data extracting methods vary according to the type of invoice. As for paper-based invoices at Alteams, paper-based invoices are received by postdelivery or via email in PDF format. Those that arrive through the post are scanned and uploaded using a software product called ReadSoft–which utilizes Optical Character Recognition (OCR) technology to extract information from invoices. The OCR technology makes it possible to convert many types of documents such as scanned paper documents, PDF files, or images captured by the camera into editable and searchable machine-coded text that can be used in business processes (OCR technology 2018 [14]). Nevertheless, the ability and accuracy of the OCR technology in extracting data is inconsistent as it relies upon the quality of the document, specifically, the resolution of images or documents, the quality of the document’s materials and the clarity of the handwriting. If received invoices failed to meet the standard or are in an illegible manner, thus, the process of extracting invoice information could only be considered a semi-automated function. The semi-automated function is where humans are required to step in to manually extract data and accomplish functions which OCR fails to perform. The invoices received can be arranged in three distinct types, in terms of a template, information completeness and readability with OCR technology. According to Forrester [15], “Repeatable tasks that search, collate, update, access multiple systems, and make simple decisions are the best RPA targets”. Matching RPA to the right process is challenging, and today, it is an art form. The healthcare, finance,

Smart Virtual Robot Automation (SVRA)-Improving Supplier

197

insurance, and banking industries are awash in overlapping systems that must work together, and few firms have an approach that helps direct RPA project efforts. 2.2 Technology Fit The open-source frameworks such as Selenium1 and Sikuli2 [1] frameworks enable costsaving and rich online community supports for practical automation solution implementation. Selenium framework [1] is built with many popular browser drivers support such as Google Chrome, Mozilla Firefox, Microsoft Edge etc. Hence, automation can be executed in different environments with little or no effort for browser migration. Moreover, technical documents are professionally written for troubleshooting. Next, deep learning [2] supervised learning will be trained with Graphic Processing Unit (GPU) with Python programming language. Sikuli2 framework [1] is an image-based automation framework that performs mouse-clicking actions based on the target GUI picture. In this work, the Sikuli framework is selected due to the difficulty of SAGE X3 ERP web application GUI automation. The Sikuli framework image-based automation is selected is because the HTML tag unique ID is constantly changing and this is a challenge as it would not be possible to capture the unique ID HTML tag for automation referencing.

3 Methodology In this study, the Agile development approach along with business process management will be implemented because it promotes the iteration process in the development life cycle. According to the study by Beck et al. [10], adopting Agile methodologies helps to minimize the risk as the concept is to develop a mini-project where it has a few components such as design, analysis, development, test, and deployment before being moved to the next cycle. 3.1 Requirement Development Requirement development consists of defining the project scope, identifying the current business process, analyzing the gap, and establishing the proposed RPA conceptual framework. The discussion with the Account Payable (AP) finance team from Company A established the project scope which the specific domain of interest will be the Purchase Invoice module (PIN) which tackles only the supplier invoices process related to Purchase Order. This is aligned with the SAGE X3 system capacity. The flow as shown in Fig. 1 is established based on the SAGE X3 functionality and organization business process. There are two main roles in this process which involve the process users from the operation department and AP from the finance department. A Payment Advice (PA) template is a documentation to inform or trigger the AP team about the payment to the vendors. The PA template consists few fields which contain the information required by the SAGE X3 interface such as Purchase Order (PO) number, Supplier name, Supplier Invoice Number, and the items for the invoices that were issued.

198

T. Y. Tang et al.

Fig. 1. Purchase Invoice Process Workflow of a selected case study: Company A

The role of AP during the screening phase is to confirm whether the process users filled up the required information according to the specifications. Once the AP executive is satisfied with the information, then the data will be entered manually into the SAGE X3 interface. The data to be filled up in the PIN module is highly dependent on the information that was entered in Good Receive Note (GRN) module previously. The system will be rejecting the creation of the PIN transaction number when there are discrepancies in the information from GRN–like the concept of three-way matching. Upon the completion of the three-way matching, the user would be able to upload the Supplier Invoice document into the attachment folder. This signifies the completion of the entire PIN process which then can be picked up by the payment team. Requirement gathering using the covert observation technique [16] was done with the end-users to walk through the SAGE X3 interface functions. This is also to confirm the cognitive process between the human process and the system capability. The observation study was done based on all five AP executives and three project secretaries from the studied organization with ten supplier invoices per person. The observation study process was conducted over several days to ensure there are no disruptions caused in their daily activities. This is an imperative step to obtain the requirements which will be used to develop the conceptual framework for RPA development. The observations are captured and summarized in Fig. 2. On top of that, the proposed actions are formulated from the observations to be transformed into RPA logic process flow before the algorithm is decoded.

Smart Virtual Robot Automation (SVRA)-Improving Supplier

199

Fig. 2. Summary of the observation

As depicted in Fig. 2, the transformation of the methodological and technological fit into RPA algorithm flow is one of the critical steps in the process–reflects how the robot should be working digitally. The first part of the action is to re-layout the GRN template in SAGE X3 to minimize the human intervention to fill up the PA template. There are many fields from the PA template that can be auto-generated from the GRN template in SAGE X3 especially the PO No, the Supplier Code, the Site, project code, department code, Delivery No, and Supplier Invoice No. The interface of GRN also required a minor change to ensure the information is meeting the PIN requirement at the AP level. Figure 3 depicts the proposed layout for the PA template where most of the date are generated from the SAGE X3 GRN module. The proposed template is recommended because it requires minimal adjustment from the SAGE X3 crystal report to fit into RPA methodological and technological principles. Another minor enhancement for this requirement is to request the IT facility to create a common shared drive for the operation users to upload the file name into the common shared folder which is labelled as PA Raw Folder, PA Completed Folder, and PA Rejected Folder. The proposed process flow is meeting the Kanban concept in the lean manufacturing management approach where the algorithm provides the tasks to the robots in terms of actions. In addition to the above enhancement, this project also created an additional login ID for the digital worker to ensure there is no disruption in the daily operation process. This is also important to be able to distinguish the transaction carried out by the digital worker and the human worker where, future enhancement of the algorithms can be done appropriately.

200

T. Y. Tang et al.

Fig. 3. Proposed template layout

3.2 Design and Development of RPA The development of RPA is based on the “To-be” process flow as shown in Fig. 4 using the swim lane diagram. The process flow consists of three main tasks and a few subtasks; PA Template folder, PA Template file and Actions, PIN Interface and Actions. There are several template folders which determine the robot’s task to move from one folder to another folder depending on the rules assigned for each task. The RPA tasks will be moving back and forth between the role of PA Template File and PIN Interface until documents from the folder PA Raw are completed. The last interface is the PIN to generate the document number for each transaction. This role is to search the for correct PO number and select all the relevant items generated from the GRN module. The next sub-section briefly describes each task in the PA Template Folder. • PA Template Folder There are three different sub-folders labelled as “PA_Raw”, “PA_Completed” and “PA _Rejected”. All these sub-folders are used as storage after the RPA robot completed the task. Sub-folder “PA_Raw” is the folder where the RPA robots will start to launch the template to copy from the “PA Template” into the PIN Interface. If the RPA task failed for executions, then it will move the “active” files into “PA_Rejected”. If the task meets all the requirements, then it will move the active file into the “PA_Completed” subfolder. The RPA_Logbook subfolder is to store all the file lists that were processed on the day either pass or fail. The RPA robots for this main task will follow the below specific rules: If “Copy-Paste Field Task” passed all the basic rules then Move the file into “PA_Completed” Else move the file into the “PA_Rejected”

Smart Virtual Robot Automation (SVRA)-Improving Supplier

201

• PA Template File and PIN Interface In this section, the task of the RPA is to copy all the relevant fields as shown in the template of Fig. 3 to the relevant SAGE X3 fields; for example, the “Site”, supplier code number, Supplier Invoice Number, and Supplier Date. The data copied from the raw file will be pasted into PIN Header Interface. At any point where the fields from the template are not matched with the PIN Header Interface, the notification of the failures will be captured and notify the respective teams via email. At the same time, it will move the failed excel files into the rejected file folder. Once all the items are created, it will launch the “Receipt Selection Interface” to initiate the matching process with the Purchase Order (PO) module.

Fig. 4. Proposed process generic “To-Be” for RPA algorithm development

• Purchase Receipt Interface The most important task in the last role will be the matching PO selection process. The robot will identify the PO number from the PA template and select all the relevant rows as shown in Fig. 5. The logic rules for this section are: If a PO number existed, then select all the rows And create the PIN number and Move the file into “Completed File”

202

T. Y. Tang et al.

Else the file is rejected and move the file into the folder “PA Rejected” and RPA will generate all the lists into an excel logbook and save the excel logbook into the subfolder “RPA_logbook” and send to the identified process owners The algorithm of coding development is using the Python language with the Pycharm IDE platform. There will be an iteration process at this stage to confirm the location of the fields between the “PA Template Files” and “SAGE X3 PIN interface module”. Once the testing of the coding is done as per the proposed “To-Be” processed, then the validation and confirmation of the RPA will be tested at the SAGE X3 Pi-Lot platform. The next section is to discuss the planning of the validation and confirmation phase.

Fig. 5. PO selection process

3.3 Validation and Confirmation Process and Development of RPA This section focuses more on the validation of the RPA algorithm. There will be two different companies that agreed to participate in this validation process. The first company is a service contractor for oil and gas while the another is from the manufacturing of chemical industries. Both companies are using SAGE X3 as their main ERP system. The validation process will be done in a pilot environment of the SAGE X3 platform to avoid disruption of the companies’ daily operations. With the help of the SAGE X3 implementer, the production data will be copied into another folder which captures all the transactions up to the GRN module. The subsequent PIN process will be done by the RPA robot with the deep learning predicted actions. The data will be compared with the production PIN data. The key variables of comparison will be several PIN transaction documents generated between the RPA robot and the AP executives within the same period. This is to confirm the accuracy of the data being generated from a cognitive point of view. The other significant parameters are the transaction processing time between the human and the robots.

4 Limitations, Conclusion and Future Work In this work, we presented the theoretical contribution of the SVRA framework to solving practical automation problems for accountants’ decision-making tasks in the SAGE

Smart Virtual Robot Automation (SVRA)-Improving Supplier

203

X3 ERP system. The SVRA framework automation will reduce the operation cost and time significantly for the organization. However, the limitation of the scope of work is only focused on specific “Purchase Invoices Process” steps in the accounting module in the SAGE X3 ERP system. However, we are planning to extend to automate different modules in the SAGE X3 ERP system in our future work such as report retrieval for Balance Sheet and Income Statement performance of each organization, Cascading report analysis of procurement and inventory management system. In this paper, we presented the SVRA conceptual framework and the methodology to automate the “Purchase Invoices Process” steps in the SAGE X3 ERP system.

References 1. Lathwal, A.: A literature review on automation testing using Selenium + Sikuli. Int. J. Distrib. Artif. Intell. 11(2), 35–40 (2019) 2. Mittal, V., Gangodkar, D., Pant, B.: Deep graph-long short-term memory: a deep learning based approach for text classification. Wirel. Pers. Commun. (2021) 3. Lacity, M., Willcocks, L., Hindel, J., Khan, S.: Robotic process automation: benchmarking the client experience. Electron. Mark. (2018). No. November 2017 4. Global, E.: Robotic process automation (RPA): technology vendor state of the market report. Retrieved 4, 2020 (2017) 5. Fersht, P.S., Snowdon, J.: The robotic process automation market will reach $443 million this year. horsesforsources (2017) 6. Chakraborti, T., et al.: From robotic process automation to intelligent process automation. In: Asatiani, A., et al. (eds.) BPM 2020. LNBIP, vol. 393, pp. 215–228. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-58779-6_15 7. Doguc, O.: Robot process automation (RPA) and its future. In: Research Anthology on CrossDisciplinary Designs and Applications of Automation, IGI Global (2022) 8. Enriquez, J.G., Jimenez-Ramirez, A., Dominguez-Mayo, F.J., Garcia-Garcia, J.A.: Robotic process automation: a scientific and industrial systematic mapping study. IEEE Access 8, 39113–39129 (2020). https://doi.org/10.1109/ACCESS.2020.2974934 9. Willcocks, L., Lacity, M., Craig, A.: The IT function and robotic process automation. Outsourcing Unit Working Research Paper Series (2015). No. October 2015 10. Rizun, N., Revina, A., Meister, V.: Method of decision-making logic discovery in the business process textual data. In: International Conference on Business Information Systems, pp. 70–84 (2019) 11. Koch, O., Buchkremer, R., Kneisel, E.: Graph Databases and Robotic Process Automation: Achieving Improvement in Project Knowledge Management (2020) 12. Kroll, C., Bujak, A., Darius, V., Enders, W., Esser, M.: Robotic process automation-robots conquer business processes in back offices. Capgemini Consult. (2016) 13. Robotic Process Automation (RPA) in the Financial Sector Technology-ImplementationSuccess for Decision Makers and Users (Mario Smeets, Ralph Erhard, Thomas Kaußler) (2021) 14. Sahu, S., Salwekar, S., Pandit, A., Patil, M.: Invoice processing using robotic process automation. Int. J. Sci. Res. Comput. Sci. Eng. Inf. Technol. (2020) 15. Le Clair, C., Cullen, A., King, M.: Digitization leaders share robotic process automation best practices. Enterprise Architecture Professionals, Forrester Research, Cambridge, MA, USA, Technical Report E-RES134021, pp. 1–12 (2016) 16. Grove, S.J., Fisk, R.P.: Observational data collection methods for services marketing: an overview. J. Acad. Mark. Sci. 20(3) (1992)

Recent Trends in Software Engineering

Design and Implementation of Modified Vedic Multiplier Using Modified Decoder-Based Adder Arti Kumari1 , Saurabh Kharwar1 , Sangeeta Singh1 , Mustafa K. A. Mohammed2 , and Salim M. Zaki3(B) 1 Microelectronics and VLSI Design Lab, National Institute of Technology, Patna, India

[email protected] 2 University of Warith Al-Anbiyaa, Karbala, Iraq 3 Computer Sciences Department, Dijlah University College, Al-Masafi Street, Baghdad, Iraq

[email protected]

Abstract. Low power design has attracted much attention since the energy dissipation is a significant factor in digital integrated circuit design. A multiplier is one of the arithmetic circuits, which plays a major role in many computational systems based on the real time applications. The power consumption in the systems greatly depends on the power consumption of its multiplier. In this digitalization era, it becomes necessary to increase the speed of the digital circuits while reducing on-chip area and memory consumption. Vedic architectures have advantages in partial product generation and additions, which are done concurrently. In this research, slice LUT’s and power of the proposed 2 × 2 and 4 × 4 novel decoder based Vedic multiplier using Urdhva Tiryakbhayam sutra are calculated and compared with conventional multiplier. Therefore, utilizing the advantages of Vedic architectures with the proposed idea to solve the problem of balancing power consumption and speed increase in circuits. The simulations carried out and synthesis of the proposed 2 × 2 bit and 4 × 4 bit multiplier has been implemented using artex-7 on Xilinx Vivado. The results of the proposed Vedic multiplier with existing Vedic multiplier exhibits a significant improvement in term of resource utilization. Keywords: Vedic multiplier · Urdhva Tiryakbhayam Sutra · Decoder · Full adders · Decoder based Vedic multiplier

1 Introduction Nowadays, each scientific computation continuously requires multiplication, which has increased the demand of multiplier in the basic fundamental block of microprocessor [1–3]. The high speed and low power multipliers have wide potential applications in the optimized network of digital signal and wireless networks-based systems [4–8]. Furthermore, the multipliers are also used in various fast Fourier transform (FFT) and discrete Fourier transform (DFT) based algorithm. The calculation of binary floating-point is mostly performed by the standard IEEE754 binary floating point, which is also resolve © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 207–215, 2023. https://doi.org/10.1007/978-3-031-20429-6_20

208

A. Kumari et al.

the various representation issues. The fixed point is obliged to a fixed cut-off, which limits it from tending to amazingly gigantic or little numbers similarly when two colossal numbers are isolated, a fixed cutoff point is introduced for the loss of precision [9–11]. In between 1911 and 1918, the Swami Sri Bharati Krsna Tirthaji rediscovered Vedic mathematics in his research he employed sixteen sutras (formulae) and Upa sutras (sub formulae) [12]. Vedic mathematics is an old arrangement of mathematics. The Urdhva Tiryakbhayam is one of the multiplication sutras which is capable of minimizing the number of steps involved in the multiplication process. In Sanskrit, the words Urdhva and Tiryakbhayam mean vertically and crosswise respectively [13]. Vedic math is not simply limited to its straightforwardness, routineness yet in addition it is sensible. This attribute of Vedic arithmetic makes it more mainstream and consequently it is a profoundly engaged examination theme in India as well as in the remainder of the world. Steps and rationale of Vedic science can be applied straightforwardly to the issues which incorporate geometry, calculation, differential analytics, and different sort of applied math [14]. The advantage of Vedic is attracting the solving problems in all streams of mathematics including calculus, arithmetic, etc. [15]. The Urdhva Tiryakbhayam algorithm is applicable for all types of numerical formats which includes Hexadecimal, Decimal, and Binary, etc. The concept behind this formula is that partial product generation can be done and then the concurrent addition of these partial products is carried out which leads to the reduction in the computational time. The full adder is used as basic building block of combinational circuit. There are various literatures reported for the area and power optimized full adder circuit [16]. Therefore, a novel Look-up table (LUT’s) optimized 2 × 2 and 4 × 4 bit Vedic multipliers using decoder are proposed and compared with conventional existing multiplier. The focus on developing the hardware essentials is a vital in the applications of computing where the applications dependency on hardware increases where there are research articles focus on hardware in different areas for example [17–18]. The organization of the paper begins with the background of the Vedic multiplier in Sect. 2. In Sect. 3 proposed Vedic multiplier are discussed, which explains the implementation of 2 × 2 and 4 × 4 bit Vedic multiplier based on Urdhva Tiryakbhayam sutra using the decoder. Result and Comparison of the implemented design with the existing multiplier are discussed in Sect. 4. Conclusions are given in Sect. 5.

2 Conventional Architecture of Vedic Multiplier 2.1 Vedic Multiplier The methodology of Vedic multiplier depends on the Vedic sutra’s standard of cross and vertically augmentations of two-bit binary framework. A model is offered underneath to show the multiplication of 2 × 2 bit numbers using Vedic multiplier. The perusal of Fig. 1 depicts the Vedic multiplication technique of 2 × 2 bit using Vedic multiplier. The above 2 × 2 bit Vedic multiplier consists of basically three steps: Step 1-Firstly, the vertically multiplication are performed using least significant bit (LSB’s) of 2-bit numbers their product is taken as output.

Design and Implementation of Modified Vedic Multiplier

209

Step 2-From that point forward, the augmentation is done in a tangle route between the LSB of the principal operand with most significant bit (MSB) of the resulting operand and by then the things from both the exercises are then added to give entirety as yield with carry. Step 3-In last step, the vertical multiplication of two operands have been performed as depicted in the Fig. 1, and the final product is obtained with the combined results of step 2.

Fig. 1. Conventional Vedic multiplication method of 2 × 2 bit binary numbers.

2.2 Conventional 2 × 2 Bit, 4 × 4 Bit Vedic Multiplier This architecture of conventional 2 × 2 bit Vedic multiplier is categorized in three stage and design with the help of AND logic and half-adder. In first stage, 4 AND gates are used and the output of LSB AND gate consider as first LSB of Vedic multiplier. The output of two intermediate AND gate act as input of half adder in second stage. Further, the out of MSB at fourth AND gate and the sum of second stage half adder act as input for third stage half adder, while the carry of second stage half adder act as second bit from right side for Vedic multiplier. Finally, the output of third stage half-adders are used as third bit and fourth bit of Vedic multiplier respectively. Similarly, 4 × 4 bit conventional Vedic multiplier is simply designed using combination of 2 × 2 bit Vedic multiplier and 4-bit ripple carry adders (RCAs) in logic realization. The architecture of the reported 4 × 4 conventional Vedic multiplier categorized in three stage. First stage consists of four 2 × 2 bit Vedic multiplier. The product of first 2 × 2 Vedic multiplier act as output first two LSB of 4 × 4 bit Vedic multiplier. Further, the product of the next 2 × 2 bit Vedic multiplier is used as input for 4-bit RCA in stage two. In third stage, two RCAs are used, which used output of previous ripple carry adder and 2 × 2 Vedic multiplier and produces next two bit as output for 4 × 4 Vedic multiplier. This conventional engineering of the Vedic multiplier is in a manner restricted from the exhibition enhancements as far as LUT’s power utilization, and speed and along these lines to accomplish a superior silicon level execution of an advanced Vedic multiplier plan with the improved asset

210

A. Kumari et al.

usage. In this paper, novel architecture is proposed that substitutes the utilization of full-adder circuit in its plan with decoder-based adder which is discussed in Sect. 3.

3 Proposed Vedic Multiplier 3.1 Modified Architecture of Vedic Multiplier The proposed architecture of Vedic multiplier is designed with the help of decoder-based adder. The 4 × 4 bit proposed Vedic multiplier is designed with 4-bit decoder-based RCA. The 4-bit adder is designed by 3:8 decoders and OR gate modules for 4 × 4 bit Vedic multiplier. The gate level representation of 2:4 decoder and block architecture representation of 1-bit full adder using 3:8 decoders is depicted in Fig. 2.

Fig. 2. (a) Gate level representation of 2:4 decoder and (b) block architecture representation of 1-bit full adder using 3:8 decoder

3.1.1 2 × 2 Vedic Multiplier Using 3:8 Decoders This section describes the 2 × 2 Vedic multiplier using 2:4 decoder. The block diagram has been manifested in Fig. 3 (a). This multiplier consists of four AND gate, two 2:4 decoder and two OR gate respectively. The carry from the last stage has been fed as input to the next stage. The adder has decoder-based design, which is relatively higher than normal adder due to the small circuit diagram show in Fig. 3.

Design and Implementation of Modified Vedic Multiplier

211

3.1.2 4 × 4 Vedic Multiplier Using 3:8 Decoder In this section using four 2 × 2 Vedic multiplier and three four-bit RCA based on decoder shown in Fig. 3 (b). The product of all 2 × 2 bit Vedic multiplier is consider as input of 4-bit RCA and the last output of 4-bit RCA is taken as output of 4 × 4 bit Vedic multiplier.

Fig. 3. Block architecture representation of (a) 2 × 2-bit Vedic multiplier using 2:4 decoder and (b) a 4 × 4-bit Vedic multiplier using 3:8 decoder.

4 Result and Analysis The proposed designs of Vedic multiplier using decoder are functionally verified through a logic simulation process, design elaboration and run implementation. To perform simulation, test benches are created for proposed 2 × 2 bit and 4 × 4 bit Vedic multiplier designs. The Verilog HDL is used to code the designs. The simulation is carried out using model sim simulation tool of Xilinx Vivado design suit. The FPGA implementation of proposed multipliers has been performed on artex-7 family device XC7a35tcpg236-1 FPGA. The simulation waveform of the proposed 2 × 2 Vedic multiplier is depicted in Fig. 4. Figure 5 shows the RTL configuration of Proposed 2 × 2 Vedic multiplier, which are according to our proposed architecture. The analysis of resource utilizations of 2 × 2 proposed Vedic multiplier are listed in Table 1. The proposed 2 × 2 bit Vedic multiplier consists 2 slice LUT’s. The Fig. 6 shows the proposed 4 × 4 bit Vedic multiplier. The RTL schematic and synthesized device of proposed Vedic 4 × 4 bit Vedic multipliers has been shown in Fig. 7 (a) and Fig. 7 (b) respectively. The analysis of resources utilization of 4 × 4 proposed Vedic multiplier are listed in Table 1. The proposed 4 × 4 Vedic multiplier consists of 50 slice LUT’s. The total dynamic power consumption of proposed Vedic multiplier for 2 × 2 bit and 4 × 4 bit are observed to be 2 mW and 17 mW and the delay of proposed Vedic multiplier for 2 × 2 bit and 4 × 4 bit are observed to be 4.793 ns and 12.845 ns respectively. The number of slices LUT’s and delay of proposed 4 × 4 Vedic multiplier are less than the previous reported multipliers using booth and array multiplier, respectively [19].

212

A. Kumari et al.

Table 1. Resources utilization and power of proposed 2 × 2 bit and 4 × 4 bit Vedic multipliers. Proposed Vedic multiplier

Slice LUT’s (20800)

Power (mW)

Delay (ns)

2 × 2 Vedic multiplier

2

2

4.793

4 × 4 Vedic multiplier

50

17

12.845

Fig. 4. The simulated output of the proposed 2 × 2 bit Vedic multiplier using the decoder

Fig. 5. The synthesized RTL schematic of proposed 2 × 2 Vedic multiplier using decoder.

Fig. 6. The simulated output of proposed 4 × 4 bit Vedic multiplier using decoder.

Design and Implementation of Modified Vedic Multiplier

213

Fig. 7. The synthesized (a) RTL schematic 4 × 4 proposed Vedic multiplier and (b) Synthesized 4 × 4 proposed device using decoder.

5 Conclusions The goal of this work is to design and implementation of Vedic Multiplier architectures based on Urdhva Tiryakbhyam sutra in Vedic Mathematics. In this paper, the design and simulation of novel 2 × 2 bit and 4 × 4 bit Vedic multiplier using decoder are

214

A. Kumari et al.

presented. This design has been synthesized and simulated on artex-7 using Xilinx Vivado model sim simulator. The utilized resources are reported in term of slice and slice LUT’s respectively. The proposed 2 × 2 bit Vedic multiplier uses 2 slice LUT’s. The proposed 4 × 4 bit Vedic multiplier uses 50 slice LUT’s. The total dynamic power of proposed Vedic is found to be 2 mW/17 mW for 2 × 2 bit/4 × 4 bit. The delay of proposed Vedic multiplier for 2 × 2 bit and 4 × 4 bit are observed to be 4.793 ns and 12.845 ns respectively. The architecture of proposed Vedic multiplier is synthesizable and is flexible in design. The proposed architecture can be extended for higher bits’ multiplier. The design can be implemented on FPGA kit. Reducing the number of slices and LUT’s is an important requirement for FPGA based design. The idea proposed in this paper may set a way for future research in this direction.

References 1. Krishna, A.V., Deepthi, S., Devi, M.N.: Design of 32-bit mac unit using Vedic multiplier and XOR logic. In: Proceedings of International Conference on Recent Trends in Machine Learning, IoT, Smart Cities and Applications, pp. 715–723. Springer (2021) 2. O’Connor, M., Swartzlander, E.E.: Exploiting asymmetry in booth-encoded multipliers for reduced energy multiplication. In: 2015 49th Asilomar Conference on Signals, Systems and Computers, pp. 722–726. IEEE (2015) 3. Saravanan, S., Madheswaran, M.: Design of hybrid encoded booth multiplier with reduced switching activity technique and low power 0.13 µm adder for DSP block in wireless sensor node. In: 2010 International Conference on Wireless Communication and Sensor Computing (ICWCSC), pp. 1–6. IEEE (2010) 4. Saravanan, S., Madheswaran, M.: Design of low power multiplier with reduced spurious transition activity technique for wireless sensor network. In: 2008 Fourth International Conference on Wireless Communication and Sensor Networks, pp. 36–39. IEEE (2008) 5. Ram, C.G., Lakshmanna, Y.R., Rani, D.S., Sindhuri, K.B.: Area efficient modified Vedic multiplier. In: 2016 International Conference on Circuit, Power and Computing Technologies (ICCPCT), pp. 1–5. IEEE (2016) 6. Ram, C.G., Rani, D.S., Balasaikesava, R., Sindhuri, K.B.: VLSI architecture for delay efficient 32-bit multiplier using Vedic mathematic sutras. In: 2016 IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT), pp. 1873–1877. IEEE (2016) 7. Abhilash, R., Dubey, S., Chinnaiah, M.: ASC design of signed and unsigned multipliers using compressors. In: 2016 International Conference on Microelectronics, Computing and Communications (MicroCom), pp. 1–6. IEEE (2016) 8. Morghade, K., Dakhole, P.: Design of fast Vedic multiplier with fault diagnostic capabilities. In: 2016 International Conference on Communication and Signal Processing (ICCSP), pp. 0416–0419. IEEE (2016) 9. Javeed, S., Patil, S.S.: Low power high speed 24-bit floating point Vedic multiplier using cadence (2018) 10. Marchesan, G.C., Weirich, N.R., Culau, E.C., Weber, I.I., Moraes, F.G., Carara, E., de Oliveira, L.L.: Exploring RSA performance up to 4096-bit for fast security processing on a flexible instruction set architecture processor. In: 2018 25th IEEE International Conference on Electronics, Circuits and Systems (ICECS), pp. 757–760. IEEE (2018) 11. Kamaraj, A., Parimalah, A.D., Priyadharshini, V.: Realisation of Vedic sutras for multiplication in Verilog. SSRG Int. J. VLSI & Signal Process. (SSRG-IJVSP) 4(1), 25–29 (2017)

Design and Implementation of Modified Vedic Multiplier

215

12. Gowthami, P., Satyanarayana, R.: Design of an efficient multiplier using Vedic mathematics and reversible logic. In: 2016 IEEE International Conference on Computational Intelligence and Computing Research (ICCIC), pp. 1–4. IEEE (2016) 13. Pohokar, S., Sisal, R., Gaikwad, K., Patil, M., Borse, R.: Design and implementation of 16 × 16 multiplier using Vedic mathematics. In: 2015 International Conference on Industrial Instrumentation and Control (ICIC), pp. 1174–1177. IEEE (2015) 14. Mistri, N.R., Somani, S., Shete, V.: Design and comparison of multiplier using Vedic mathematics. In: 2016 International Conference on Inventive Computation Technologies (ICICT), vol. 2, pp. 1–5. IEEE (2016) 15. Ravali, B., Priyanka, M.M., Ravi, T.: Optimized reversible logic design for Vedic multiplier. In: 2015 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT), pp. 127–133. IEEE (2015) 16. Kumar, P., Singh, S.: Optimization of the area efficiency and robustness of a QCA based reversible full adder. J. Comput. Electron. 18(4), 1478–1489 (2019) 17. Al-khazarji, H.A.H., Abdulla, M.A., Abduljabbara, R.B.: Robust approach of optimal control for DC motor in robotic arm system using matlab environment. Int. J. Adv. Sci. Eng. Inf. Technol. 10(6), 2231 (2020). https://doi.org/10.18517/ijaseit.10.6.8923 18. Novikov, I.I., Shepelev, S.O., Gusev, I.D., Kulemin, R.M.: The research and development of a software and hardware complex for determining the type of the road surface. In: 2021 IEEE Conference of Russian Young Researchers in Electrical and Electronic Engineering (ElConRus), pp. 2730–2733 (2021). https://doi.org/10.1109/ElConRus51938.2021.9396623 19. Kunchigi, V., Kulkarni, L., Kulkarni, S.: High speed and area efficient Vedic multiplier. In: 2012 International Conference on Devices, Circuits and Systems (ICDCS), pp. 360–364. IEEE (2012)

Design and FPGA Implementation of Matrix Multiplier Using DEMUX-RCA-Based Vedic Multiplier Balivada Yashwant Kumar1 , Saurabh Kharwar1 , Sangeeta Singh1 , Mustafa K. A. Mohammed2 , and Mohammed Dauwed3,4(B) 1 Microelectronics and VLSI Design Lab, National Institute of Technolgy, Patna, India 2 University of Warith Al-Anbiyaa, Karbala, Iraq 3 Department of Medical Instrumentation Techniques Engineering, Dijlah University College,

Baghdad 10022, Iraq [email protected] 4 Department of Computer Science, College of Science, University of Baghdad, Baghdad 10070, Iraq

Abstract. Matrix multiplication is a common technique for increasing the computational speed of scientific and engineering tasks. The matrix multiplier is designed in this paper utilizing an optimized Vedic multiplier. Vedic mathematics, which is a collection of sutras for doing mathematical arithmetic simply and more speedily, is utilized to speed up multiplication. These sutras aid in the reduction of several processors performance metrics, such as power and delay. As a result, the current multiplier approaches are replaced with Urdhva Tiriyagbhyam, a Vedic Math multiplication methodology. We used 1:8 demultiplexer-based Full Adders (DFAs) to create the Vedic Multiplier to circumvent the power constraint. As a result, the overall power of matrix multiplication was improved across various bits. The Optimized Matrix multiplier is designed in Verilog HDL, and the Nexys DDR4 of the Artix-7 series is utilized as the target device for synthesis. Keywords: Matrix multiplier · Vedic multiplier · Urdhva Tiryagbhyam Sutra · Demultiplexer · Full adders

1 Introduction Matrix multiplication is a demanding and fundamental block in arithmetic algebra that has a significant impact on the central processing unit (CPU) performance [1]. Much work is still being done on the hardware and software levels to improve the performance of matrix multipliers. This study yielded several ways for speeding up processing, one of which is parallel programming [2–6]. Unfortunately, due to power and memory limits, performance and power improvements have come to an end [7]. Multiplication is also an important operation in ALU (Arithmetic Logic Unit). For better and faster multiplication,

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 216–224, 2023. https://doi.org/10.1007/978-3-031-20429-6_21

Design and FPGA Implementation of Matrix Multiplier Using DEMUX-RCA

217

we need accurate multipliers with reduced power and delay. As most mathematical and logical operations include multiplication, designing an efficient multiplier is critical. Vedic has the advantage of encouraging problem-solving in all branches of mathematics, including calculus, arithmetic, and so on [8]. The Urdhva Tiryagbhyam algorithm works with all numerical formats, including Hexadecimal, Decimal, Binary, and so on. The idea behind this formula is that partial product formation can be done first, followed by concurrent addition of these partial products, resulting in a reduction in computation time [9]. The full adder (FA) is the fundamental component of a combinational circuit. For the area and power-optimized FA circuit, there is a lot of literature [10]. The Vedic multiplier is used to develop the matrix multiplier in this study. After employing demultiplexerbased Full Adders (DFAs), we improved the Vedic multiplier design. The background of the matrix multiplier and Vedic multiplier is presented in Section 2 of the paper. The proposed Vedic multiplier is discussed in Section 3, which discusses how to use the DEMUX to create a 4 × 4 bit Vedic multiplier based on the Urdhva Tiryagbhyam sutra. Section 4 discusses the outcome and compares the implemented design to the pre-existing multiplier. Section 5 covers the conclusions.

2 Conventional Architecture of Vedic Multiplier and Full Adder A. Matrix multiplier: Matrix multiplication is a binary operation in mathematics that produces a matrix from two matrices, particularly in linear algebra. The number of columns in the first matrix must equal the number of rows in the second matrix for matrix multiplication to work. The resulting matrix, known as the matrix product, has the first matrix’s number of rows and the second matrix’s number of columns. AB stands for the product of matrices A and B. If A is ⎡ an m × n matrix and B is an n ⎤ × r matrix,⎡ ⎤ b11 b12 . . . . . . b1r a11 a12 . . . . . . a1n ⎢ ⎥ ⎥ ⎢ ⎢ b21 . . . . . . . . . . . . ⎥ ⎢ a21 . . . . . . . . . . . . ⎥ ⎢ ⎥ ⎥ ⎢ A = ⎢ . . . . . . . . . . . . . . . ⎥ B = ⎢ . . . . . . . . . . . . . . . ⎥the ⎢ ⎥ ⎥ ⎢ ⎣ ... ... ... ... ... ⎦ ⎣ ... ... ... ... ... ⎦ am1 . . . . . . . . . amn bn1 . . . . . . . . . bnr matrix product C = AB is defined to be the m × r matrix ⎡ ⎤ c11 c12 . . . . . . c1r ⎢ ⎥ ⎢ c21 . . . . . . . . . . . . ⎥ ⎢ ⎥ C = ⎢ ... ... ... ... ... ⎥ ⎢ ⎥ ⎣ ... ... ... ... ... ⎦ cm1 . . . . . . . . . cmr C ij = ai1 b1j + ai1 b1j + · · · + ain bnj =

n 

aik bkj

k=1

for i = 1, ..., m and j = 1, ..., r. B. Vedic multiplier: In logic realization, a 4 × 4 bit traditional Vedic multiplier is simply designed by combining two 2 × 2 bit Vedic multipliers with 4-bit ripple carry

218

B. Y. Kumar et al.

Fig. 1. Block architecture representation of 4-bit Vedic Multiplier.

Fig. 2. Block architecture representation of Full adder.

adders (RCAs). Figure 1 shows the block diagram of the 4-bit Vedic multiplier. Four 2-bit Vedic Multipliers, two 4-bit RCAs, and one OR Gate are required for 4-bit multiplication utilizing Vedic Multipliers. The three-stage architecture of the reported 4 × 4 traditional Vedic multiplier is described. Four 2 × 2 bit Vedic multipliers make up the first step. The output first two LSB of the 4 × 4 bit Vedic multiplier is the product of the first 2 × 2 Vedic multiplier. In stage two, the product of the next 2 × 2 bit Vedic multiplier is utilized as the input for the 4-bit RCA. Two RCAs are utilized in the third stage, with the output of the preceding RCA and 2 × 2 Vedic multiplier being used to produce the following two-bit as output for the 4 × 4 Vedic multiplier [11–16]. Our multiplicand should be

Design and FPGA Implementation of Matrix Multiplier Using DEMUX-RCA

219

a3a2a1a0, and our multiplier should be b3b2b1b0. Let’s use the output “res [7:0]” for this. To begin, we’ll split the multiplicand into two parts: “a3 a2,” “a1 a0,” and “b3 b2,” “b1b0.” We’ll take 2 bits at a time, utilizing the 2 × 2 Vedic multiplier to create a 2-bit Vedic multiplier, using the foundations of Vedic Multiplications. As indicated in Fig. 1, we will obtain a 4-bit Vedic Multiplier. The four 2-bit Vedic multipliers’ inputs are “a1 a0” and “b1 b0,” “a1 a0” and “b3 b2,” “a3 a2” and “b1 b0,” “a3 a2” and “b1 b0,” “a3 a2” and “b3 b2,” respectively. Finally, “res [7:0]” is the result of our multiplication. Figure 2 depicts the Full Adder’s block diagram. Full adders, in general, use three bits as input and output sum (S) and carry (Cout ). We connect them to make an N-bit adder and cascade the carry bit from one adder to the next after designing the whole adder (e.g.:–if we are required to design a 5-bit adder, we can string five full adders together and cascade the carry). The normal output is labeled S, whereas the carry output is labeled Cout. This conventional Vedic multiplier design is restricted from exhibition enhancements in terms of power consumption and speed, and therefore from achieving a higher silicon level execution of an advanced Vedic multiplier plan with improved asset utilization [8, 17]. In this work, a novel architecture is presented that replaces full-adder circuits with DFAs in its plan, which is addressed in Sect. 3.

Fig. 3. Block architecture representation of proposed 4-bit Vedic Multiplier using DEMUXRCAs.

3 Proposed Vedic Multiplier The proposed architecture of the Vedic multiplier is designed with the help of DFAs. The 4-bit proposed Vedic multiplier is designed with 4-bit DEMUX-based RCA (DEMUXRCA) as shown in Fig. 3. A Demultiplexer, known as “DEMUX” is a digital decoder

220

B. Y. Kumar et al.

Fig. 4. Block architecture representation of DEMUX-based Full Adder.

that takes a single input and then switches it to any number of individual output lines. A DEMUX converts a serial data signal at the input to parallel data at its output line. The Block diagram of DFA is shown in Fig. 4. The DFA consists of one DEMUX and two OR gates. The input of DFA is Din and output sum (S) and carry (Cout ). After designing the DFA, we string them together to form an N-bit DEMUX-RCA.

4 Result and Analysis Through a logic simulation method, design elaboration, and run implementation, the proposed designs of matrix multipliers employing DEMUX-based Vedic multipliers are functionally verified. To simulate the suggested 2 × 2 bit and 4 × 4 bit Vedic multiplier designs, test benches are built. The designs are written in Verilog HDL. The simulation is run using the Xilinx Vivado design suite’s model sim simulation tool. The FPGA implementation of Proposed multipliers has been performed on Artex7 family device XC7a35tcpg236–1 FPGA. We have designed and implemented three different sizes of matrix multipliers i.e. 2 × 2 matrix, 3 × 3 matrix, and 4 × 4 matrix respectively using the proposed DEMUX-based Vedic multiplier. The modified matrix multipliers are compared with convention RCA-based Vedic multiplier by logic RTL simulation, synthesis, and implementation. The simulation waveform of the DEMUX-RCA-based matrix multiplier of 4 × 4 size is shown in Fig. 5. The implemented RTL schematic and synthesized device design of DEMUX-RCA-based matrix multiplier of 4 × 4 size are shown in Figs. 6 and 7 respectively. The calculated power and delay of 2 × 2, 3 × 3, and 4 × 4 matrix multiplier using both conventional and DEMUX-RCA-based Vedic multiplier are shown in Table 1.

Design and FPGA Implementation of Matrix Multiplier Using DEMUX-RCA

221

Table 1. The calculated delay and power of RCA-Vedic multiplier-based matrix multiplier and DEMUX -Vedic multiplier-based matrix multiplier. Multiplier

Matrix

Delay (ns)

Power (mW)

Matrix Multiplier

2×2

16.425

29

3×3

13.675

81

4×4

18.399

169

2×2

11.763

13

3×3

11.646

44

4×4

15.737

103

DEMUX-based Matrix Multiplier

Fig. 5. The output waveform of proposed DEMUX-RCA-based 4 × 4 matrix multiplier.

The total dynamic power consumption of the proposed matrix multiplier for the 2 × 2, 3 × 3, and 4 × 4 matrix are observed to be 13, 44, and 103 mW. The delay of the proposed matrix multiplier for 2 × 2, 3 × 3, and 4 × 4 matrix are observed to be 11.763 ns, 11.646 ns, and 15.737 ns respectively. The power and delay of the proposed 4x4 bit matrix multipliers of 2 × 2, 3 × 3, and 4 × 4 matrix are less than the conventional Vedic multipliers based matrix multipliers.

5 Conclusion The RTL design of matrix multiplier architectures based on Urdhva Tiryagbhyam sutra in Vedic Mathematics is implemented in this work. The architecture of the 4-bit Vedic multiplier is designed and simulated using demultiplexer (DEMUX) based RCAs. These designs are synthesized and simulated on Artex-7 using the Xilinx Vivado model sim simulator. The proposed 4-bit matrix multiplier of size 2 × 2, 3 × 3, and 4 × 4 matrix consumes the total dynamic power of 13, 44, and 103 mW respectively. The delay of the proposed 4-bit matrix multiplier of size 2 × 2, 3 × 3, and 4 × 4 matrix are observed to be 11.763 ns, 11.646 ns, and 15.737 ns respectively. The architecture of the proposed matrix multiplier is synthesizable and is flexible in design. The proposed architecture can be extended for higher bits’ multiplier. Power and delay reduction are critical criteria for FPGA-based architecture. The concept described in this work could pave the way for more research in this area in the future.

222

B. Y. Kumar et al.

Fig. 6. The synthesized RTL schematic 4 × 4 proposed matrix multiplier using DEMUX-RCA.

Design and FPGA Implementation of Matrix Multiplier Using DEMUX-RCA

223

Fig. 7. The synthesized device of proposed matrix multiplier using DEMUX-RCA.

References 1. Qasim, S.M., Telba, A.A., AlMazroo, A.Y.: FPGA design and implementation of matrix multiplier architectures for image and signal processing applications. Int. J. Comp. Sci. Netw. Sec. 10(2), 168–176 (2010) 2. Memik, S.O., Katsaggelos, A.K., Sarrafzadeh, M.: Analysis and FPGA implementation of image restoration under resource constraints. IEEE Trans. Comput. 52(3), 390–399 (2003) 3. Ebeling, C., Fisher, C., Xing, G., Shen, M., Liu, H.: Implementing an OFDM receiver on the RAPiD reconfigurable architecture. IEEE Trans. Comput. 53(11), 1436–1448 (2004) 4. Goslin, G.R.: Guide to using field programmable gate arrays (FPGAs) for application-specific digital signal processing performance. In: High-Speed Computing Digital Signal Processing, and Filtering Using Reconfigurable Logic, International Society for Optics and Photonics, vol. 2914, pp. 321–331 5. Isoaho, J., Pasanen, J., Vainio, O., Tenhunen, H.: DSP system integration and prototyping with FPGAs. J. VLSI Sign. Proc. Syst. Sign. Image Video Technol. 6(2), 155–172 (1993) 6. Ye, A.G., Lewis, D.M.: Procedural texture mapping on FPGAs. In: Proceedings of the 1999 ACM/SIGDA Seventh International Symposium on Field Programmable Gate Arrays, pp. 112–120. (1999) 7. Tan, Y., Imamura, T.: An energy-efficient FPGA-based matrix multiplier. In: 24th IEEE International Conference on Electronics, Circuits and Systems (ICECS), pp. 514–517. IEEE (2017) 8. Ravali, B., Priyanka, M.M., Ravi, T.: Optimized reversible logic design for Vedic multiplier. In: International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT), pp. 127–133. IEEE (2015)

224

B. Y. Kumar et al.

9. Kumar, G.G., Charishma, V.: Design of high speed vedic multiplier using Vedic mathematics techniques. Int. J. Sci. Res. Publ. 2(3), 1 (2012) 10. Kumar, P., Singh, S.: Optimization of the area efficiency and robustness of a QCA-based reversible full adder. J. Comput. Electron. 18(4), 1478–1489 (2019). https://doi.org/10.1007/ s10825-019-01369-5 11. Ram, C.G., Lakshmanna, Y.R., Rani, D. S., Sindhuri, K.B.: Area efficient modified vedic multiplier. In: International Conference on Circuit, Power and Computing Technologies (ICCPCT), pp. 1–5. IEEE (2016) 12. Ram, C.G., Rani, D.S., Balasaikesava, R., Sindhuri, K.B.: VLSI architecture for delay efficient 32-bit multiplier using vedic mathematic sutras. In: IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT), pp. 1873–1877. IEEE (2016) 13. Abhilash, R., Dubey, S., Chinnaiah, M.: ASC design of signed and unsigned multipliers using compressors. In: International Conference on Microelectronics, Computing and Communications (MicroCom), pp. 1–6. IEEE (2016) 14. Morghade, K., Dakhole, P.: Design of fast vedic multiplier with fault diagnostic capabilities. In: 2016 International Conference on Communication and Signal Processing (ICCSP), pp. 0416–0419. IEEE (2016) 15. Javeed, S., Patil, S.S.: Low power high speed 24-bit floating point vedic multiplier using cadence. Int. Res. J. Eng. Tech. (IRJET) 5, 1771–1775 (2018) 16. Marchesan, G.C., Weirich, N.R., Culau, E.C., Weber, I.I., Moraes, F.G., Carara, E., de Oliveira, L.L.: Exploring rsa performance up to 4096-bit for fast security processing on a flexible instruction set architecture processor. In: 2018 25th IEEE International Conference on Electronics, Circuits and Systems (ICECS), pp. 757–760. IEEE (2018) 17. Mistri, N.R., Somani, S., Shete, V.: Design and comparison of multiplier using vedic mathematics. In: 2016 International Conference on Inventive Computation Technologies (ICICT), vol. 2, pp. 1–5. IEEE (2016)

Analysis and Modeling of Brushless DC Motor PWM Control Technique Using PSIM Software Ola Hussein Abd Ali Alzuabidi, Mohammed Abdulla Abdulsada(B) , and Mohammed Wajeeh Hussein Department of Computer Techniques Engineering, Dijlah University College, Baghdad, Iraq {ola.hussien,mohammed.abdulla,mohammed.alameen}@duc.edu.iq

Abstract. The Permanent Magnet Brushless DC (PMBLDC) motors are becoming increasingly popular because of their high torque density, high efficiency and small size. Variety industry applications which have been used these motors required a control on it’s peed. One of the popular techniques that applied to adjust the speed of BLDC motor is Pulse Width Modulation (PWM). In this study, the analysis and model of a PWM speed control for BLDC motor is presented. The proposed model comprised of voltage source inverter, hall sensor, Proportional plus Integral (PI) controller and commutation circuit. To reduce the current harmonic contents of the inverter, a second order low pass filter is applied with the controller. The proposed model is simulated by using Power Simulation (PSIM) software. The simulation results show that the proposed model gives the desired motor speed of 5000 rpm and electromagnetic torque of 0.6 n.m. Keywords: Permanent Magnet Brushless DC Motor · PWM · Hall Position Sensors · PI Controller

1 Introduction Brushless DC motors are popular because of their low maintenance and great efficiency, as well as the fact that they are tiny in size and easy to install [1, 2]. There are, however, a number of issues with these motors for variable speed operation, which have been overcome over the last few decades by the advancements in microprocessors, power semiconductors; adaptable speed drivers; permanent-magnet brushless electric motor production and control patterns [3–7]. Over the next five years, home appliances are anticipated to be one of the fastest-growing end-product markets for electronic motor drivers. Refrigerators, laundry washers, vacuum cleaners, and freezers are just a few of the key home appliances. Single-phase AC induction, including splitphase, capacitor-start and capacitor-run kinds, and universal motors, have traditionally been used in household appliances. Without regard to efficiency, these old-school motors are frequently powered by the main AC supply. Energy efficiency, improved performance, decreased acoustic noise, and more convenience features are becoming the norm for consumers, who want to save money. Those old-fashioned technologies won’t help [8]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 225–234, 2023. https://doi.org/10.1007/978-3-031-20429-6_22

226

O. H. A. A. Alzuabidi et al.

There are several uses for BLDC motors in every business area. Appliances, industrial control, automation, aircraft, and so on are only a few examples [9, 10]. The motors and the necessary integrated power electronic circuits are usually simulated using modern simulation software. One of the powerful simulation software is PSIM. The simulation results will give precise predictions of the BLDC motor performance in real-time. In this paper the analysis, model and simulation of the PWM technique applied to the BLDC motor based on voltage source inverter is presented. The simulation program used to validate the results is PSIM software.

2 Control Techniques of PMBLDC Motor There are two ways to control a PMBLDC motor using control techniques. These are called sensor-based and sensor-free. The future commutation interval can be calculated from the machine’s current rotor position using sensors. The DC bus rail voltage or PWM approach can also be used to control the motor. Using both can result in a big torque at huge loads and a high efficiency at low loads in some systems [11, 12]. The harmonic current can likewise be controlled with such hybrid architecture. On a DC motor, as brushes from one coil touch the commentator of a second coil, the motor continues rotating. BLDC motors, on the other hand, need electronic switches to commutate, and they require knowledge of the rotor position. Stator windings must be activated at the precise moment the rotor poles align with them. Predetermined commutation intervals can also be used to drive the BLDC motor. Brushless commutation must, however, be performed with knowledge of rotor location in order to obtain exact speed control and maximum generated torque [13, 14]. The rotor position has been determined by means of sensors, mechanical position sensors (like hall sensors), shaft encoders, and resolvers.

3 Proposed Model The PMBLDC motor drive organized through proportional plus integral (PI) is the subject of this inquiry and evaluation. To run the PMBLDC motor driving simulation, the controllers are used. In addition, a BLDC motor setup simulation was used to create the PI controller. Software such as the PSIM application makes it easy to model power electronics and motor control systems. As a result, PSIM’s graphical user interface is simple and intuitive. Creating and modifying a circuit is simple. The waveform display program has a variety of post-processing functions that make it easy to assess the simulation results. The PSIM features include: 1. 2. 3. 4. 5.

Ease of usage. Fast simulation. Controls can be shown in a variety of ways. Modules that are built in. Add-on modules are a type of add-on.

Analysis and Modeling of Brushless DC Motor PWM Control Technique

227

6. Sweep the parameters. 7. Display of the real-time waveform at runtime. It is possible to interact with the PSIM environment. In the middle of a simulation, users can modify parameters and examine the waveform [15].

4 Structure and Building of the Proposed Model The blocks and parameters of the proposed system model are listed as follows: 1. 2. 3. 4. 5.

Inverter motor drive system based on a voltage supply. The speedometer. The BLDC motor has a PI controller for speed. Fifty hertz is the fundamental frequency. DC voltage is 300 V.

Permanent Magnet Synchronous Machine (PMSM) sensors, power converter, and control algorithm make up the four basic components of the brushless dc motor. For example, the PMSM takes electrical energy from the source and uses it to generate mechanical energy. When it comes to brush-less DC motors, one of the most important aspects is the rotor position sensors that are used for determining a gate signal for every semiconductor in the power electronic converter based on information from the rotor position and command signals. 4.1 Mathematical Model of BLDC Motor The voltage equations for BLDC motor of balanced system are: ⎤ ⎡ ⎤⎡ ⎤ ⎤⎡ ⎤ ⎡ ⎤ ⎡ Rs 0 0 ia ia ea Laa Lab Lac vas ⎣ vbs ⎦ = ⎣ 0 Rs 0 ⎦⎣ ib ⎦ + d ⎣ Lba Lbb Lbc ⎦⎣ ib ⎦ + ⎣ eb ⎦ dt vcs ic Lca Lcb Lcc ic ec 0 0 Rs ⎡

(1)

where vas , vbs , andvcs stand for the stator phase voltages; Rs stands for a stator resistance for each phase ia , ib , andic stand for the stator phase currents, Laa , Lbb , Lcc stand for the self-inductance of phases a, b and c; Lca , Lcb , Lcc stand for the mutual inductances among phases a, b and c; ea , eb , aswellasec stand for the phase back electromotive forces. Winding resistance has been considered to be equal. Also it is assumed that if there are no prominent rotating parts, then there is no change in rotational resistance as the angle changes. Laa = Lbb = Lcc = L

(2)

Lab = Lba = Lac = Lca = Lbc = Lcb = M

(3)

228

O. H. A. A. Alzuabidi et al.

From Eq. (1) assuming constant self and mutual inductance, the voltage equation becomes: ⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ Va R00 ia L−M 0 0 e i d ⎣ a⎦ ⎣ a⎦ ⎣ Vb ⎦ = ⎣ 0 R 0 ⎦⎣ ib ⎦ + ⎣ 0 L − M ⎦ (4) 0 ib + eb dt 00R 0 0 L−M Vc ic ic ec In state space form, the equation is arranged as: ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ i i e V d ⎣ a ⎦ −R ⎣ a ⎦ 1 ⎣ a ⎦ 1 ⎣ a ⎦ ib = ib − eb + Vb dt L L L ic ic ec Vb

(5)

The electromagnetic torque (Te) is specified by: Te = (ea ia + eb ib + ec ic )/Wr

(6)

The motor equation is specified by: dwr = (Te − TL − Bwr )/J dt

(7)

where wr angular rotor velocity in radians per sec, J moment of inertia, B is damping coefficient that has been usually insignificant and often ignored in this the system. 4.2 Speed Control of BLDC Motor Position feedback has employed in servo applications. Position data can be used to calculate velocity feedback. For the speed control loop, this eliminates the need for a separate velocity transducer. Rotor position is related to voltage strokes in a BLDC motor. Using Hall sensors, the rotor position may be determined [16]. The speed of the motor can be regulated by altering the voltage across the motor. It is possible to alter the motor voltage by changing the duty cycle for PWM signal while controlling six switches in a three-phase bridge. When a motor’s electrified windings are activated, the strength of the magnetic field they generate affects the motor’s speed and torque. As a result, changing a rotor voltage and current will alter the motor’s output speed. Commutation assures only proper rotation of the rotor. To get the duty cycle, the PI controller takes this discrepancy into account. When precise control of both speed and current is required to achieve the desired torque, the PMBLDC motor is frequently used. Closed loop control of the BLDC motor drive is depicted in Fig. 1. The BLDC motor drive with speed feedback (PWM control) is shown in Fig. 2. The waveform return emf in Fig. 2, nodes a, b, c stand for the stator winding terminals for phase A,B,C of a 3-phase BLDC motor. Motor hall effect sensors have been employed as inverter gating signals, resultant in a six-pulse functioning for the inverter. The sensor is used to model the speed. A dc current feedback loop generates a high-frequency pulse.

Analysis and Modeling of Brushless DC Motor PWM Control Technique Speed/V

PI Controller

Commutation Logic

Inverter

229

BLDC Motor

Pulse Width

Duty Cycle

Hall Signal

Space Feedback

Fig. 1. Block diagram for closed loop speed control

Fig. 2. Structure of proposed system of BLDC motor with speed feedback

5 Simulation Results and Discussion The proposed drive system includes reference current generator, a PI speed controller, BLDC motor and an IGBT inverter. All of these components have been modeled and merged in order to replicate real-time events. PSIM is a power electronics and motor control simulation software with a userfriendly interface that is fast and easy to use. Studies of power electronics, magnetic and motor drive systems, as well as digital and analog control, are all made easier with this powerful simulation tool. Custom DLL blocks allow PSIM to link to third-party software such as MATLAB. The PSIM simulation package comprises of PSIM, the simulator engine, and SIMVIEW the waveform processing application. Figure 3 shows the simulation procedure. System components and BLDC motor parameters are shown in Table 1. The ratio between M and the stator self-inductance L is typically between −1/3 and −1/2, depending on the winding configuration. Assuming M is unknown, −0.4*L can be used as a

230

O. H. A. A. Alzuabidi et al.

reasonable default number. The desired speed of BLDC motor is 5000 rpm and the electromagnetic torque is 0.6 n.m. The result of output waveforms were obtained by applying PSIM program running. Simulation result obtained from start-up mechanical speed (rpm), input current (3-phase) and electromagnetic torque.

Fig. 3. PSIM simulation process

Table 1. System components and BLDC motor parameters. Parameter

Value

R(Stator resistance)

11.9 

L(Stator self inductance) M(Stator mutual inductance)

0.00207 H −0.00069 H

Vpk/KRPM (mechanical speed)

32.3 V/KRMP

Vrms/Krpm(mechanical speed)

22.9 V/Krpm

Figures 4 and 5 demonstrate the output 3-phase currents and voltages for PWM control respectively. From Fig. 4, it can be noted that the output 3-phase current are symmetric about time axis and distributed equally with ripples. The maximum value of the 3-phase current is about 2 A. The output 3-phase voltage has a maximum value of 215 v as shown in Fig. 5. Figure 6 shows the waveforms of the back EMF voltages in three phases. Phase voltage is presented by rotating the display 120°, the mechanical torque is shown in Fig. 7 while the speed of BLDC motor is displayed in Fig. 8. From Fig. 7, it can be noted that the generated ripple in the electromagnetic torque of the BLDC motor is coincides with the commutation instant and has an average value of about 0.6 n.m. It can be noted that the steady state final value of the motor speed is 5000 rpm which reached at 20 ms as shown in Fig. 8. The speed response has zero value of overshoot and zero steady-state error.

Analysis and Modeling of Brushless DC Motor PWM Control Technique

Fig. 4. Output 3-phase current (Ia, Ib and Ic) for PWM control

Fig. 5. Output 3-phase voltage (Va , Vb , and Vc ) for PWM control

231

232

O. H. A. A. Alzuabidi et al.

Fig. 6. Back EMF voltage in three phases

Fig. 7. Electromagnetic torque of BLDC motor

6 Conclusions Performance evaluation and simulation of BLDC motor based on voltage source inverter with PI controller is investigated in this study. For high-performance industrial applications, the PI controller-based BLDC motor drive is best suited for speed applications. The simulation results are performed by using PSIM software. The effectiveness of the simulation in this study is established by prediction of performance of the BLDC motor in real-time. A second order low pass filter is added to the controller to reduce the inverter

Analysis and Modeling of Brushless DC Motor PWM Control Technique

233

Fig. 8. Speed of BLDC motor

current harmonics. The simulation results show that the proposed model provides the required motor speed of 5000 rpm with zero overshoot and zero steady-state error.

References 1. Bimal, K.: Bose: power electronics and motor drives: advances and trends, 2nd edn. Academic Press, London, UK (2021) 2. Gamazo-Real, J.-C., Martínez-Martínez, V., Gomez-Gil, J.: ANN-based position and speed sensorless estimation for BLDC motors. Measurement 188, 110602 (2022) 3. Zhang, H., Li, H.: Fast commutation error compensation method of sensorless control for MSCMG BLDC motor with nonideal back EMF. IEEE Trans. Power Electron. 36, 8044–8054 (2021) 4. Lakshmiprabha, K.E., Govindaraju, C.: An integrated isolated inverter fed BLDC motor for photovoltaic agric pumping systems. Microproc. Microsyst. 103276 (2020) 5. Vanchinathana, K., Selvaganesan, N.: Adaptive fractional order PID controller tuning for brushless DC motor using artificial bee colony algorithm. Res. Control Optimizat. 4, 100032 (2021) 6. López, M.G., et al.: A novel fuzzy-PSO controller for increasing the lifetime in power electronics stage for brushless DC drives. IEEE Access 7, 47841–47855 (2019) 7. Bhuvaneswari, S., et al.: Optimized design of permanent magnet brushless DC motor for ceiling fan applications. Mater. Today Proc. 45(Pt 2), 1081–1086 (2021) 8. Naseri, F., et al.: Predictive control of low-cost three-phase four-switch inverter-fed drives for brushless DC motor applications. IEEE Trans. Circ. Syst. I 68, 1308–1318 (2021) 9. Kumar, J., Jena, P.: Recent advances in power electronics and drives. Springer Nature, Singapore (2021) 10. de Almeida, P.M., et al.: Robust control of a variable-speed BLDC motor drive. IEEE J. Emerg. Select. Top. Indus. Electr. 2, 32–41 (2021) 11. Mishra, P., et al.: Implementation and validation of quadral-duty digital PWM to develop a cost-optimized ASIC for BLDC motor drive. Control Eng. Pract. 109, 104752 (2021)

234

O. H. A. A. Alzuabidi et al.

12. Patel, H., Chandwani, H.: Simulation and experimental verification of modified sinusoidal pulse width modulation technique for torque ripple attenuation in brushless DC motor drive. Eng. Sci. Technol. Int. J. 24, 671–681 (2021) 13. Wang, L., et al.: A commutation optimization strategy for high speed brushless DC drives with inaccurate rotor position signals. In: 2021 Sixteenth International Conference on Ecological Vehicles and Renewable Energies (EVER). Monte-Carlo, Monaco (2021) 14. Trifa, V., et al.: Block commutation vs sinusoidal commutation for BLDC motors. In: International Semiconductor Conference (CAS). Sinaia, Romania (2020) 15. Nayana, J., et al.: Comparison of DC to DC converters for PV application using PSIM model. In: 2020 International Conference on Power Electronics and Renewable Energy Applications (PEREA). Kannur, India (2020) 16. S. Al-Adsani, A., et al.: BLDC motor drives: a single hall sensor method and a 160° commutation strategy. IEEE Trans. Energy Convers. 36, 2025–2035 (2021)

Single-Bit Architecture for Low Power IoT Applications Reeya Agrawal1,2 , Sangeeta Singh2 , Mustafa K. A. Mohammed3 , and Mohammed Dauwed4,5(B) 1 GLA University, Mathura, India 2 Microelectronics and VLSI Design Lab, National Institute of Technology, Patna, India 3 Radiology Techniques Department, Dijlah University College, Al-Masafi Street,

Baghdad 00964, Iraq 4 Department of Medical Instrumentation Techniques Engineering, Dijlah University College,

Baghdad 10022, Iraq [email protected] 5 Department of Computer Science, College of Science, University of Baghdad, Baghdad 10070, Iraq

Abstract. This paper discusses how six transistor static random access memory cells work, how a write driver circuit works, and how different sense amplifiers work, like a current differential sense amplifier and a current latch sense amplifier. This paper also discusses the proposed architecture for low-power Internet of Things applications and how it works and looks. They include a six-transistor static random access memory cell, a write driver, and a variety of sense amplifiers, such as current differential and current latch sense amplifiers, that can help people Schematic out what is going on in their devices. Low-power forced stacks, lowpower sleep stacks, and low-power dual sleep techniques. To make the proposed architectures more powerful, low-power designs make six transistor static random access memory cells less powerful. Using a six-transistor static random access memory cell with less power is the best way to make it work. In this case, the current latch sense amplifier in architecture uses the least amount of power, 8.34µW with 35 transistors, compared to other architectures. This is because of the area of the amplifier increases. Keywords: Low Power Reduction Techniques (LPRT) · Circuit of Write Driver (CoWD) · Six Transistor Static Random-Access Memory Cell (6TSRAMC ) · Current Differential Sense Amplifier (CDSA ) · Current Latch Sense Amplifier (CLSA )

1 Introduction In contrast to the traditional binary method, ternary logic systems can save much energy and money because of more information and complex machines in modern computers. Another benefit is that it will be faster to do math in serial and parallel ways when ternary logic is used. Battery life has become a big problem as more people use mobile devices © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 235–245, 2023. https://doi.org/10.1007/978-3-031-20429-6_23

236

R. Agrawal et al.

[1]. Intelligent clocks, small sensor nodes, wireless networking devices, and other new technologies are working hard to make them better and more useful for people. Random Access Memory (RAM) chips are used in almost every digital system to make the whole thing run more smoothly and save energy. When you give the cell power, six transistor static random-access memory cells (6TSRAMC ) don’t need to be changed often to keep their contents safe. Because of this, 6TSRAMC is a better choice than dynamic random-access memory (DRAM). DRAM is a significant benefit. 6TSRAMC and standard complementary metal oxide semiconductor (CMOS ) technology can be used to make the complete consumer-on-chip system [2]. Much electricity comes from leak capacity when technology decreases to a minimal level. Furthermore, the 6TSRAMC unit has most of the transistors on one chip. Because cache storage is done in an array, cutting power in a single 6TSRAMC will save a lot of energy on the whole device. The basic structure of the 6TSRAMC is used the most [3]. The 6TSRAMC has a lot of power and a lot of power efficiency. The 6TSRAMC architecture could still use much less power in the future, but that doesn’t mean it can’t be improved [4]. 1.1 Background of 6TSRAMC Robert Norman, who worked for Fairchild Semiconductor, developed the bipolar semiconductor 6TSRAMC in 1963. John Schmidt, who worked for Fairchild Semiconductor, developed the metal oxide semiconductor (MOS) 6TSRAMC in 1964 [5]. It was a 6TSRAMC with a 64-bit MOS p-channel, and it had a 6TSRAMC . Since then, the 6TSRAMC has been the driving force behind any new CMOS -based technology that has been made. In 1965, IBM’s Arnold Farber and Eugene Schlig used a transistor gate and a tunnel diode latch to create a hard-wired memory cell. 1.2 Production Charges As soon as FinFET transistors were used to make 6TSRAMC , they were less efficient. The transistor size kept getting smaller. This made it harder to fit more 6TSRAMC into a smaller space. Size isn’t the only problem with the newer 6TSRAMC [6]. A six-transistor memory cell with a power-reduction sleep transistor technique and a current latch sense amplifier in the architecture make the cell run more efficiently. This consumes less power, which means it takes up less space. When it comes to power consumption and space, there is always a trade-off. It’s essential for silicone, efficiency, and critical stability to have 6TSRAMC . The rate at which the sensory enhancer ages varies. 6TSRAMC is the fastest storage unit, but it also costs the most. Most researchers work on a low-power on-chip memory array to cut power consumption and cost. The paper is divided into five sections where Sect. 1 describes the introduction related to cache memory, static random-access memory, and the internet of things; Sect. 2 describes the related work done by different authors, and Sect. 3 describes low power reduction techniques which are used in architecture to save power consumption; Sect. 3 describes proposed architecture with different sense amplifiers, Sect. 4 describes the results and Sect. 5 describe the summary of the paper in the form of conclusion.

Single-Bit Architecture for Low Power IoT Applications

237

2 Literature Review Habeeb et al. [7] describe many very large-scale integrated circuits (VLSI) chips now have 6TSRAMC , which are fast and low-power. Simultaneously, increasing integration and operation speeds have made power dissipation essential. As a result, much work has gone into employing circuit and architectural approaches to reduce the power consumption of CMOS RAM chips. Panda et al. [8] describe that 6TSRAMC has a few benefits over DRAM. The significant advantage is that 6TSRAMC does not require regular periodic refreshing, meaning that whatever is stored in a 6TSRAMC will last until the power is turned on. Rath et al. [9] describe that 6TSRAMC is a critical component of the embedded cache memory in portable digital devices. The 6TSRAMC has become a popular data storage device due to its high storage density and quick access time. Shubham et al. [10] describe that memory is a critical component in the current VLSI system design. Before manufacturing, it must be extensively examined in space, power, and performance. Due to the ever-increasing need for data processing, 6TSRAMC is a critical contender for in-memory design and is receiving much attention. Sphurti et al. [11] explain that researchers have been shrinking CMOS circuits for the last five decades to achieve successful execution in terms of speed, power blow-out, and unchanging quality. Lakshmi et al. [12] describe that power consumption and latency are two more characteristics that play a vital part in determining the device’s performance. Memory is an essential component of many widgets; its size shrinks as the device’s size shrinks. Janniekode et al. [13] describe the numerous short channel effects under nanoscale CMOS technology at lower supply voltages and technology.

3 Low Power Reduction Techniques In this section, researchers discuss ways to reduce how much power designers use and how circuits work. 3.1 Low Power Sleep Transistor Technique (LPSTT) The sleep NMOS transistor is connected to the pull-down network and GND. There are sleep transistors that turn the power rails off [14]. When the circuit is not in use, the sleep transistors turn off. 3.2 Low Power Forced Stack Technique (LPFST) Low power forced stack method; the transistor stack effect is used to cut down leakage. The forced stack strategy is not very good at saving energy [15]. 3.3 Low Power Sleep Stack Technique (LPSST) Sleep and stacking strategies are used in the “sleepy stack” method. One of the split transistors is connected to a sleep transistor in the same way as the other [16].

238

R. Agrawal et al.

3.4 Low Power Dual Sleep Technique (LPDST) There are two more pull-up transistors in sleep mode and two more pull-down transistors in the OFF and ON states. There are fewer transistors in a logic circuit because all logic circuits have the same two-step sleep component to be built with fewer of them [17].

4 Proposed Single Bit Architecture Circuit of write driver (CoWD’s) input pins are word enable (WE ) and bit , and its output pins are BTL and BTLBAR (bit lines), whereas 6TSRAMC’s input pins are word line (WL), BTL, and BTLBAR , with two output pins, V1 and V2 . The input pins of the sense amplifier (SA) are YSEL , SAen , BTL, and BTLBAR , and it has two output pins, V3 and V4 .

Schematic 1. 6TSRAMC CDSA architecture

Schematics 1 and 2 depict single-bit architecture with its schematics. CoWD, 6TSRAMC , and SA are three cache memory designs for single-bit architecture.

Schematic 2. 6TSRAMC CLSA architecture

Single-Bit Architecture for Low Power IoT Applications

239

The volatile nature of 6TSRAMC can further be protected by integrating tamperreactive sensors, which, if interfered with, destroy encryption keys. 6TSRAMC employs a primary bistable circuit to store a file’s bit. 4.1 Circuit of Write Driver (CoWD) Working and Schematic The value is written in the 6TSRAMC using CoWD. CoWD reduces the 6TSRAMC write margin from the full precharge by discharging the bit line tension. As shown in Schematic 3, CoWD would load the data onto bit lines and write it to them. Bit pin value will be stored on the bit lines [18] if WE are high.

Schematic 3. CoWD. Reproduced from Geethumol et al. [19] under the terms of the Creative Attribution Commons License 4.0 (CC-By 4.0). http://creativecommons.org/licenses/by/4.0/.

4.2 6TSRAMC Working and Schematic 6TSRAMC stands for six transistors static random-access memory. It has a feature to store data before receiving power for an extended period [20].

Schematic 4. 6TSRAMC . Reproduced from Shukla, Neeraj Kr et al. [21], under the Creative Attribution Commons License 4.0 (CC-By 4.0). http://creativecommons.org/licenses/by/4.0/.

The 6TSRAMC consists of two back-to-back CMOS transistors with a two-stable transistor, each based on the data in the two transistors [22], as shown in Schematic 4. Bit-lines are used to access data in the 6TSRAMC . It is the most common memory cell due to its low stability and static dissipation. If WL is enabled, access transistors can read and write operations bound to the cell’s bit lines.

240

R. Agrawal et al.

4.3 Sense Amplifier (SA) Preloading and equalizing the SA in the high-gain metastable region is the first step in the sensing process. Because the inputs and outputs are not separated in a SA, the insulation transistors must isolate the SA of the bit-lines to prevent complete unloading of a 0 bit-line, which wastes power and time [23–25]. 4.3.1 CDSA Working and Schematic Schematic 5 shows the circuit architecture for a current detecting amplifier. CDSA works by measuring bit-cell current directly. It does not rely on a different voltage developed across the bit-line [26].

Schematic 5. CDSA . Reproduced from Reeya Agrawal et al. [27], with permission from Elsevier. Copyright (2021)

As a result, bit-line power precharge can be reduced to a minimum. A transmitting circuit with unit transmission characteristics and a sensing circuit monitor the differential current and make up the current differential sense amplifier [28]. 4.3.2 CLSA Working and Schematic The sense amplifier is a critical circuit in single-bit architecture. The sluggish discharge is reduced because of the expanded bit line, and the bit cells have access to the transistor [29]. This way, a slight change in the SA digital level boosts the bit line voltage values. The following is how the circuit works. Schematic 6 shows how the differential voltage is transferred to the V3 and V4 CLSA inputs on bit lines [31, 32].

5 Result Analysis The circuit output waveforms are detailed in this section and their corresponding diagrams. The number of transistors and power consumption of various cache memory designs for single-bit architecture have been compared.

Single-Bit Architecture for Low Power IoT Applications

241

Schematic 6. CLSA . Reproduced from Reeya Agrawal et al. [30], with permission from Springer Nature. Copyright (2021)

Schematic 7. CoWD output waveform

Schematic 7 shows the CoWD output waveform. It can be described in four cases: Case (a): WE are low, and bit is low BTL = VDD and BTLBAR = VDD , Case (b): WE are high and Bit is low BTL = 0V and BTLBAR = high, Case (c): Bit is high and WE are low, BTL = BTLBAR = VDD /2 and. Case (d): Bit is high, and WE are high BTL = VDD and BTLBAR = 0V. Schematic 8 shows the 6TSRAMC output waveform, holding, and writing operation. The CDSA and CLSA reading operation is depicted in Schematics 9 and 10, respectively. SAen = VDD and WL = VDD amplifiers are active during the reading operation. Tables 1 and 2 show how LPRT is used in 6TSRAMC architecture, compare PC and area (in terms of NoT) and conclude that 6TSRAMC with LPSTT CLSA architecture

242

R. Agrawal et al.

Schematic 8. 6TSRAMC output waveform

Schematic 9. CDSA output waveform

Schematic 10. CLSA output waveform

consumes 8.34 µW of power and has 35 transistors, which is the least of any other architecture. PC stands for power consumption and NoT stands for the number of transistors in tables.

Single-Bit Architecture for Low Power IoT Applications

243

Table 1. Parameter analysis of architecture on applying LPRT in single-bit 6TSRAMC CDSA Architecture Techniques over 6TSRAMC

Single-bit 6TSRAMC CDSA architecture PC

NoT

LPSTT

10.62 µW

33

LPFST

10.20 µW

33

LPSST

12.23 µW

34

LPDST

11.16 µW

35

Table 2. Parameter analysis of architecture on applying LPRT in single-bit 6TSRAMC CLSA Architecture Techniques over 6TSRAMC

Single-bit 6TSRAMC CLSA architecture PC NoT

LPSTT

8.3412 µW

35

LPFST

12.56 µW

35

LPSST

9.23 µW

36

LPDST

9.76 µW

37

In comparison, 6TSRAMC with LPFST CDSA architecture consumes 10.2 µW of power and has 33 transistors, implying that area and power are always trade-offs.

6 Conclusion The analysis and implementation of 6TSRAMC CDSA and 6TSRAMC CLSA architectures are described in this paper. Several design aspects, such as power consumption and transistors, were also examined. Aside from that, low power reduction techniques have been used in the architecture of the 6TSRAMC . Compared to other architectures, a single-bit 6TSRAMC with LPSTT CLSA architecture consumes 8.34 µW of power and has the smallest number of transistors, 35. Researchers may be able to do this work in the form of an array in the future.

References 1. Dounavi, H.-M., Sfikas, Y., Tsiatouhas, Y.: Aging prediction and tolerance for the SRAM memory cell and sense amplifier. J. Electron. Test. 37(1), 65–82 (2021). https://doi.org/10. 1007/s10836-021-05932-6 2. Fritsch, A., et al.: 24.1 A 6.2 GHz Single-Ended Current Sense Amplifier (CSA) Based Compileable 8T SRAM in 7nm FinFET Technology. In: 2021 IEEE International Solid-State Circuits Conference (ISSCC), vol. 64. IEEE (2021)

244

R. Agrawal et al.

3. Chen, J., et al.: Analysis and optimization strategies toward reliable and high-speed 6T compute SRAM. IEEE Trans. Circ. Syst. I Reg. Papers 68.4, 1520–1531 (2021) 4. Agrawal, R., Tomar, V.K.: Analysis of cache (SRAM) memory for core I™ 7 processor. In: 2018 9th International Conference on Computing, Communication and Networking Technologies (ICCCNT). IEEE (2018) 5. Krishnaraj, R., et al.: Design of a memory array using tail transistor and sleep transistor based 7T SRAM with low short circuit and standby power. In: IOP Conference Series: Materials Science and Engineering, vol. 1084(1). IOP Publishing (2021) 6. Kumar, R.M., Sridevi, P.V.: Design of 1KB SRAM array using enhanced stability 10t SRAM cell for FPGA based applications 7. Habeeb, M.S., Md, S.: Design of low power SRAM using hierarchical divided bit-line approach in 180-nm technology 8. Zhang, J., Wang, Z., Verma, N.: In-memory computation of a machine-learning classifier in a standard 6T SRAM array. IEEE J. Solid-State Circ. 52(4), 915–924 (2017) 9. Lokesh, S.B., MeghaChandana, K., Niharika, V., Prathyusha, A., Rohitha, G.: Design of reading and write operations for 6T SRAM cell. IOSR J. VLSI Signal Proc (IOSR-JVSP) 8(1), 43–46 (2018) 10. Tripathi, T., Chauhan, D.S., Singh, S.K., Singh, S.V.: Implementation of low-power 6T SRAM cell using MTCMOS technique. In: Bhatia, S.K., Mishra, K.K., Tiwari, S., Singh, V.K. (eds.) Advances in Computer and Computational Sciences. AISC, vol. 553, pp. 475–482. Springer, Singapore (2017). https://doi.org/10.1007/978-981-10-3770-2_44 11. Sanjana, S.R., Banu, R., Shubham, P.: Design and performance analysis of 6T SRAM cell in 22nm CMOS and FinFET technology nodes. In: 2017 International Conference on Recent Advances in Electronics and Communication Technology (ICRAECT), pp. 38–42. IEEE (2017) 12. Singh, V., Singh, S.K., Kapoor, R.: Static noise margin analysis of 6T SRAM. In: 2020 International Conference for Innovation in Technology (INOCON), pp. 1–4. IEEE (2020) 13. Shukla, S., Singh, S., Bansal, K., Tyagi, P., Singh, S.K.: Design of 6T SRAM cells on different technology nodes. In: Smart Computing, pp. 599–605. CRC Press (2021) 14. Banu, S., Gupta, S.: The sub-threshold leakage reduction techniques in CMOS circuits. In: 2020 International Conference on Smart Technologies in Computing, Electrical, and Electronics (ICSTCEE). IEEE (2020) 15. Munaf, K.A., Ramashri, T.: Survey on power optimization techniques for low power VLSI circuit in active & standby mode of operation 16. Deepak, N., Bharani Kumar, R.: Certain investigations in achieving low power dissipation for SRAM cell. Microprocess. Microsyst. 77, 103166 (2020) 17. Agrawal, R., Goyal, V.: Analysis of MTCMOS cache memory architecture for processor. In: Proceedings of International Conference on Communication and Artificial Intelligence. Springer, Singapore (2021) 18. Agrawal, R.: Comparative study of latch type and differential type sense amplifier circuits using power reduction techniques. In: International Conference on Microelectronic Devices, Circuits and Systems. Springer, Singapore (2021) 19. Geethumol, T., Sreekala, K., Dhanusha, P.J.I.: Power and area efficient 10T SRAM with improved read stability. J. Microelectron 3(1) (2017) 20. Mishra, J.K., Upadhyay, B.B., Misra, P.K., Goswami, M.: Design and analysis of SRAM cell using body bias controller for low power applications. Circ. Syst. Signal Proc. 40(5), 2135–2158 (2020). https://doi.org/10.1007/s00034-020-01578-5 21. Aparna, R.C.S.C.J.I.J.: A study of different SRAM cell designs. 9(3) 2021 22. Harshey, V., Bansal, S.K.: Designing of variations tolerant sensing amplifier circuit for deep sub-micron memories

Single-Bit Architecture for Low Power IoT Applications

245

23. Tiwari, N., et al.: Modelling and design of 5T, 6T and 7T SRAM cell using deep submicron CMOS technology. In: Proceedings of Second International Conference on Smart Energy and Communication. Springer, Singapore (2021) 24. Agrawal, R.: Low-power SRAM memory architecture for IoT systems. In: Natarajan, S.K., Prakash, R., Sankaranarayanasamy, K. (eds.) Recent Advances in Manufacturing, Automation, Design and Energy Technologies. LNME, pp. 505–512. Springer, Singapore (2022). https://doi.org/10.1007/978-981-16-4222-7_57 25. Di Nardo, M.J.:Investigating input offset reduction with timing manipulation in low voltage sense amplifiers. MS thesis. University of Waterloo (2021) 26. Shrivastava, Y., Gupta, T.K.: Design of high-speed low variation static noise margin ternary S-RAM cells. IEEE Trans. Dev. Mater. Reliabil. 21(1), 102–110 (2021) 27. Agrawal, R., Kumar, M.J.M.T.P.: Low power single bit cache memory architecture. (2021) 28. Lin, Z., et al.: Two-direction in-memory computing based on 10T SRAM with horizontal and vertical decoupled read ports. IEEE J. Solid-State Circ. (2021) 29. Song, B., et al.: Environmental-variation-tolerant magnetic tunnel junction-based physical unclonable function cell with auto write-back technique. In: IEEE Transactions on Information Forensics and Security (2021) 30. Agrawal, R.: Cache memory architecture for core processor. In: Proceedings of International Conference on Advanced Computing Applications. Springer, Singapore (2022) 31. Agrawal, R.: Analysis of cache memory architecture design using low-power reduction techniques for microprocessors. In: Natarajan, S.K., Prakash, R., Sankaranarayanasamy, K. (eds.) Recent Advances in Manufacturing, Automation, Design and Energy Technologies. LNME, pp. 495–503. Springer, Singapore (2022). https://doi.org/10.1007/978-981-16-4222-7_56 32. Mishra, A.K., Chopra, U., Dhandapani, V.: Comparative analysis in terms of power and delay of the different sense amplifier topologies, vol. 47, p. 57

Hybrid Fuzzy Logic Active Force Control for Trajectory Tracking of a Quadrotor System Sherif I. Abdelmaksoud1(B)

, Musa Mailah2

, and Tang H. Hing2

1 Aerospace Engineering Department, King Fahd University of Petroleum & Minerals,

Dhahran 31261, Saudi Arabia [email protected] 2 School of Mechanical Engineering, Universiti Teknologi Malaysia, 81310 Johor Bahru, Malaysia

Abstract. The quadrotor unmanned aerial vehicle (UAV) system is considered the most widespread machine these days. There are numerous uses for it in both civilian and military sectors. However, it faces challenges and impediments that degrade its performance and affect its path following such as external wind gusts. Besides it is an underactuated, nonlinear, and coupled dynamic model. This paper presents a hybrid control system based on an innovative approach called active force control (AFC). The proposed control structure merges the proportionalintegral-derivative (PID) controller with the AFC approach adjusted intelligently by employing the fuzzy logic (FL) method. To test the feasibility of the suggested hybrid scheme, various types of external disturbances including the harmonic, pulsating, and Dryden wind gust model were applied. Results show the effectiveness of the proposed hybrid control scheme in expelling the applied disturbances while maintaining system stability and trajectory tracking. Findings indicate that the proposed strategy enhanced the performance of around 67, 91, and 50% for forward, sideward, and upward motions, respectively. Results show how effectively the suggested approach may implement for real-time applications. Keywords: Active force control · Intelligent control systems · PID controller · Quadrotor · UAVs · Fuzzy logic · Trajectory tracking · Dryden wind gust · Disturbance rejection

1 Introduction Quadrotor UAV models have gained the attention of scientists and researchers from different disciplines nowadays. Undoubtedly, this stems from the extent of the features that they have allowing them to be utilized in the civilian and military sectors. They are distinguished by vertical take-off and landing, ability to hover at a particular point or limited zones, light weight, and maneuverability [1]. All these features make them suitable for a wide variety of applications for instance surveillance, transportation, inspection, weather detection, disaster management, and smart agriculture [2].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 246–256, 2023. https://doi.org/10.1007/978-3-031-20429-6_24

Hybrid Fuzzy Logic Active Force Control for Trajectory Tracking

247

Despite their many and varied uses, they face many impediments that degrade their performance and obstruct desired trajectory tracking. For example, battery life, limited flight range and endurance, fixed and dynamic obstacles, and limitations of electric motors [3]. Among these, external disturbances and wind gusts represent one of the current bottlenecks in UAV development and are considered to be a major and basic requirement for the flight industry [4]. The need for stable, safe, and efficient flight operations is a vital matter. Besides, the ability to reject different forms of known and unknown perturbations in normal or complex environments is also a necessary aspect. This Approach that tackles these challenges remains open and requires dedicated research to propose a sound control strategy to effectively reject the disturbances and uncertainties [5].

2 Literature Review/Related Work Numerous studies have been performed to illuminate the effect of disturbances on the quadrotor system and different control approaches proposed to reject them. A proportional-integral-derivative (PID) controller is one of the popular control systems used in various sectors. It features simplicity, minimal control efforts, and satisfactory performance. To make the PID controller in the best state and performance, its control parameters, namely, KP , KI , and KD have to be selected appropriately. They can be tuned through a trial-and-error method (TEM) [6], look-up tables for example the Ziegler-Nichols (ZN) method [7], or any optimization/intelligent methods for instance the GA, PSO, etc. [8–10]. A variety of research works related to nonlinear controllers have been performed to handle a broad spectrum of operating conditions and loads. Various research work was conducted using sliding mode control (SMC) [11], backstepping control (BC) [12], and model predictive control (MPC) [13]. However, the complexity and some adverse impacts, such as the chattering effect are major negative predicaments. Despite the fact that numerous control strategies were presented to protect quadrotor systems from various perturbations, some control techniques, such as the active disturbance rejection control (ADRC), are mathematically challenging and need the modification of a significant number of control parameters [14, 15]. One promising control approach is Active force control (AFC). It was initially illustrated by Hewit and Burdess (1981). The approach can be merged with conventional, contemporary, or intelligence control strategies [6, 16]. It depends on two facets, namely, the proper determination of the inertia/mass value of the dynamic model and measurements of torque/force and acceleration signals [17]. Some research works on the rotorcraft UAV systems were reported by [6, 18] that analytically and experimentally combined a PID with the AFC approach. However, the main concern about the AFC strategy is the accurate and appropriate determination of the inertia value. While the sensitivity analysis of the proposed AFC-based control scheme applied to an aerial system was studied by [19].

248

S. I. Abdelmaksoud et al.

The main research contributions of this research study are listed as follows: (a) The PID controller has been integrated with the AFC technique to propose an innovative intelligent control system for a quadrotor system to track the desired trajectories while maintaining system stability and improving the capability to eliminate external perturbations and wind gusts. (b) The establishment of an optimization method for fine-tuning the nominal control parameters. Here, the fuzzy rules are employed as key ideas to determine the optimal value of the AFC parameter. The research paper is structured as follows: the mathematical model of a quadrotor system and the suggested control schemes including a PID controller, and AFC strategy employing the fuzzy logic method are described in Sect. 3. Section 4 describes simulated findings and their discussion in the presence of the Dryden wind gust model. Section 5 shows the paper’s conclusion.

3 Methodology A nonlinear mathematical model is initially described for a quadrotor, based on the Newton-Euler formulation, considering certain assumptions and conditions. To attain the required behaviour, a PID controller is designed as its gains are tuned heuristically. Due to the unsuccessful compensation of the PID controller for disturbances, it was proposed to be merged with an innovative control method known as the AFC technique to create a hybrid control scheme (PID-AFC). Meanwhile, an artificial intelligence (AI)-based method is introduced employing the FL method and embedding it into the AFC loop to automatically tune the control parameters. Simulation methodologies to implement the derived dynamic model and proposed hybrid control scheme were described in the ensuing sections. 3.1 Quadrotor Modeling This part discusses the derivation of the equations of motion of a quadcopter system considering some impacts including aerodynamic, gyroscopic, and drag effects. To obtain equations of motion of the quadrotor model, some assumptions should be considered: • Body-fixed frames’ axes and origin are aligned with the axes and origin of the quadcopter system. • Quadcopters have an inflexible and symmetrical design. • The propellers are inflexible. In a quadrotor, there are two subsystems: the translational subsystem determining the quadcopter’s position (x, y, z) and the rotational subsystem that determines its orientation (φ, θ, ψ). As a result, it is a 6-degrees of freedom (DOF) model. With six DOF and four control actions, the quadcopter is considered an underactuated system. The nonlinear translation motion of the quadrotor model is expressed as:  (1) F B = m˙vB + ωB × (mv B )

Hybrid Fuzzy Logic Active Force Control for Trajectory Tracking

249

where ω is the angular velocity vector. The angular momentum of a body with inertia matrix J in the rotational subsystem can be formulated as:  M B = J ω˙ B + ωB × (J ωB ) (2) Thus, the equations of motion for translational and rotational subsystems taking into account the gyroscopic, aerodynamic, and drag impacts can be given as [2, 20]: u1 x¨ = (cφsθ cψ + sφsψ) − k1 x˙ (3) m u1 y¨ = (cφsθ sψ − sφcψ) − k2 y˙ (4) m u1 z¨ = (cφcθ ) − g − k3 z˙ (5) m   Iyy − Izz θ˙ Jr ωr lu2 φ¨ = − k4 φ˙ (6) − + ψ˙ θ˙ Ixx Ixx Ixx   ˙ r ωr lu3 φJ Izz − Ixx ¨ ˙ ˙ θ= − k5 θ˙ (7) + + ψφ Iyy Iyy Iyy   Ixx − Iyy u4 ψ¨ = − k6 ψ˙ (8) + φ˙ θ˙ Izz Izz Note that in Eq (3) and other related equations that follow, c  cos (·) and s  sin (·) represent the sine and cosine functions, respectively. u1 , u2 , u3 , and u4 are the control input for the total thrust force, thrust variation between the left rotor and the right rotor, thrust change between the back rotor and the front rotor, and torque difference between clockwise rotating rotors and the counter-clockwise rotating rotors, respectively. g is the gravitational acceleration. m is the quadcopter mass. While l is the distance from each motor to the center of mass. k1 , k2 , and k3 are aerodynamic translational coefficients however k4 , k5 , and k6 are aerodynamic rotational coefficients. Jr is the rotors’ inertia whereas wr is the rotor relative speed. Ixx , Iyy , and Izz are the moment of inertia through the principal axes in the body frame. It can be noted that the rotational subsystem is fully actuated while the translational subsystem is underactuated. 3.2 Proposed Fuzzy Logic Active Force Control with (FLAFC) A FLAFC scheme was designed by incorporating the fuzzy logic into the AFC scheme for obtaining the estimated inertia value intelligently, as shown in Fig. 1. Subsequently, it was combined with the designed PID controller generating the PID-FLAFC method. Through designing the FL, the Mamdani type rule was used. There are two inputs, namely, the error e(t) and the actual response y(t), while the estimated inertia I’ value is only output. Table 1 lists the rules that were applied in the FL block. The linguistic variables employed for the e(t), y(t), and I’ were all described as [Small (Sm), Medium (Md), Large (Lg)]. In this study, the triangular membership function was used for all variables. For the defuzzification in the Mamdani engine, the centroid technique was applied.

250

S. I. Abdelmaksoud et al. Table 1. Rule-based inferences applied in the FLAFC. y(t)

e(t)

Sm

Md

Lg

Sm

Lg

Lg

Md

Md

Lg

Md

Md

Lg

Md

Md

Sm

Fig. 1. A schematic diagram of the hybrid PID–FLAFC scheme.

3.3 Simulation The MATLAB/Simulink software was utilized to implement the derived quadrotor system and to simulate the control strategies including the PID and PID-FLAFC schemes. And to apply the FLC facets, the Fuzzy Logic Toolbox was used. The parameters of the PID controller were adjusted heuristically. While the PID-FLAFC was designed, implemented, simulated, and compared with a PID controller to examine its effectiveness. The quadrotor parameters are tabulated in Table 2 whereas the designed PID gains were listed in Table 3. To assess the control strategies’ effectiveness, integral square error (ISE) was selected as the index performance to compare the control schemes. The simulations were solved using ode45 with the step solver was a variable and the relative tolerance was 0.001. Output results were analysed in the time domain. Various forms of disturbances, i.e., harmonic and pulsating waves were introduced to the quadrotor model to give the effect of external disturbances, as follows: Sinusoidal wave disturbance

T0 = 0.5sin(1.6t) Nm

Pulsating disturbance

Amplitude 0.7 Nm for a duration of 1 s

Besides, a Dryden wind model perturbation was fed into the quadrotor to simulate the wind effects. For the Dryden wind model, wind perturbation depends on wind speed [21] and can be described as a summation of sinusoidal excitations [22]. Without loss of generality, it can be presumed that the components of wind disturbances are equal [23].

Hybrid Fuzzy Logic Active Force Control for Trajectory Tracking

251

Based on the following equation, the Dryden wind gust model can be obtained where the values of k,i were taken between 0.05π rad/s and 2.5π rad/s, as shown in Fig. 2. di (t) = di,0 +

n 

ai,K , sin(i,k t + qi,k ), i = 1, 2, 3

(9)

k=1

where di (t) is the time-dependent description of the wind disturbance in a given time t, n is the number of sinusoidal sinusoids, and di,0 is the static wind disturbance. The ai,K , i,k , and qi,k are the amplitude, frequency, and phase shift of the corresponding sinusoid, respectively.

Fig. 2. Dryden wind gust model.

Table 2. Quadrotor system parameters [24] Description

Unit

Value

Quadrotor mass, m

kg

0.65

Distance from the center of mass to each motor, 

m

0.23

Moment of inertia about the x-axis, I xx

kgm2

7.5E-3

Moment of inertia about the y-axis, I yy

kgm2

7.5E-3

Moment of inertia about the z-axis, I zz

kgm2

1.3E-2

Gravity, g

m/s2

9.8

Thrust force coefficient, K F

Ns2

3.13E-5

Moment coefficient, K M

mms2

7.5E-7

Moment of Inertia of the rotor, J r

kgm2

6E-5

Translational drag force coefficients, K 1 , K 2 , K 3

Ns/m

0.01

Aerodynamic friction coefficients, K 4 , K 5 , K 6

Ns/rad

0.012

Attitude’s saturation

20◦ ≥ φ, θ, ψ ≥ −20◦

252

S. I. Abdelmaksoud et al.

Table 3. PID controller gains tuned heuristically for various types of motion for the quadrotor. Motion PID Gains

x

y

z

φ

θ

ψ

KP

5

5

10

0.5

0.5

0.1

KI

0

0

4

0

0

0

KD

1

1

5

0.5

0.5

0.1

4 Results and Discussion A combined summary of system responses in the time domain for all control strategies is shown in Figs. 3 and 4. The flight path of the quadrotor when employing the hybrid PID-FLAFC strategy and the PID controller in the presence of the Dryden wind gust model is shown in Fig. 5. The ISE index performance comparison is given in Table 4. 0.3

0.2

0.2

x (m)

y (m)

0.3

0.1

0.1

Desired

Desired

PID

PID

PID-FLAFC

PID-FLAFC

0

0 0

20

10

0

30

20

10

Time (s)

30

Time (s)

(a)

(b) 0.3

1 0.2

(rad)

z (m)

0.8 0.6 0.4

0.1

Desired

Desired

PID

0.2

PID PID-FLAFC

PID-FLAFC

0

0 0

20

10

Time (s) (c)

30

0

20

10

30

Time (s) (d)

Fig. 3. Quadrotor system responses in the existence of harmonic disturbance.

Heuristically tuned PID controller, demonstrates through the results its ability in settling the rotorcraft model and following the desired trajectory in disturbances-free conditions. It is, however, ineffective in rejecting a variety of disturbances, and its operation is adversely influenced by the coupling impact. This is because it is a linear control system that produces a suitable performance but in limited operating conditions. The findings reveal the effectiveness of the PID-FLAFC system in fending off the introduced perturbations efficiently while maintaining system stability compared to the PID control system. It should be noted that the FL is a very user-friendly approach, as it

Hybrid Fuzzy Logic Active Force Control for Trajectory Tracking

253

Fig. 4. Quadrotor system responses in the existence of pulsating disturbance.

Fig. 5. The flight path of the quadrotor in the presence of the Dryden wind gust.

does not need to rely on any accurate mathematical formulation or model; the user must merely rely on his judgment and experience. It can be seen the remarkable superiority of the AFC-based hybrid strategy in terms of the settling time, peak time, and steady-state error in comparison to the traditional PID controller. It can be observed that, on average, the settling time decreased by more than 55% for the cases with different disturbances. Figure 5 shows the effectiveness of the AFC-based method in eliminating the Dryden wind gust during the flight path of the quadrotor. According to Table 4, the PID-FLAFC

254

S. I. Abdelmaksoud et al.

Table 4. Integral square error (ISE) comparison for the quadrotor in the presence of the Dryden wind model. x

y

z

PID

0.569

0.626

0.849

PID-FLAFC

0.184

0.055

0.411

method has enhanced the performance by around 67, 91, and 50% for forward, sideward, and upward motions, respectively.

5 Conclusion Wind gust disturbances are impediments facing rotorcraft UAV systems in general and twin-rotor helicopters in particular during operation. The twin-rotor helicopter model is also a strongly coupled and highly nonlinear dynamic model. To defeat these limitations, intelligent control schemes based on the AFC approach utilizing the IL algorithm and FL methods have been proposed. In addition to their outstanding performance, these IAFC-based control schemes also contribute significantly toward ensuring system stability by rejecting different kinds of external disturbances. In the existence of various perturbations, the PID-FLAFC scheme improves the performance by almost 70–75% for pitching motions and almost 30–35% for yawing motions. An important restriction on the application of the IAFC approach is the actuator’s limited bandwidth, which constrains the AFC signal. In regards to future research work, It would be useful to address the application of the suggested PID-AFC controller with nonlinear controllers like sliding mode control to verify system performance.

References 1. Emran, B.J., Najjaran, H.: A review of quadrotor: an underactuated mechanical system. Annu. Rev. Control. 46, 165–180 (2018). https://doi.org/10.1016/j.arcontrol.2018.10.009 2. Hasseni, S.-E.-I., Abdou, L.: Adaptive nonlinear robust control of an underactuated micro UAV. Int. J. Dyn. Control 9(3), 1144–1166 (2021). https://doi.org/10.1007/s40435-020-007 22-y 3. Abdelmaksoud, S.I., Mailah, M., Abdallah, A.M.: Control strategies and novel techniques for autonomous rotorcraft unmanned aerial vehicles: a review. IEEE Access 8, 195142–195169 (2020). https://doi.org/10.1109/ACCESS.2020.3031326 4. Lungu, M.: Backstepping and dynamic inversion combined controller for auto-landing of fixed wing UAVs. Aerosp. Sci. Technol. 96, 105526 (2020). https://doi.org/10.1016/j.ast. 2019.105526 5. Hua, C., Chen, J., Guan, X.: Fractional-order sliding mode control of uncertain QUAVs with time-varying state constraints. Nonlinear Dyn. 95(2), 1347–1360 (2018). https://doi.org/10. 1007/s11071-018-4632-0 6. Omar, M., Mailah, M., Abdelmaksoud, S.I.: Robust active force control of a quadcopter. J. Mekanikal 40(2):2 (2017)

Hybrid Fuzzy Logic Active Force Control for Trajectory Tracking

255

7. Praveen, V., Pillai, A.S.: Modeling and simulation of quadcopter using PID controller. Int. J. Control Theory Appl. 9(15), 7151–7158 (2016) 8. Abdelmaksoud, S.I., Mailah, M., Abdallah, A.M.: Improving disturbance rejection capability for a quadcopter UAV system using self-regulating fuzzy PID controller. In: 2020 International Conference on Computer, Control, Electrical, and Electronics Engineering (ICCCEEE), pp. 1–6. (2021) https://doi.org/10.1109/ICCCEEE49695.2021.9429661 9. Khuwaja, K., Tarca, I.C., Lighari, N.-Z., Tarca, R.C.: PID controller tuning optimization with genetic algorithms for a quadcopter. Rec. Innov. Mechatron. DUPress 5(1), 7 (2018) 10. Noordin, A., Basri, M.A.M., Mohamed, Z., Abidin, A.F.Z.: Modelling and PSO fine-tuned PID control of quadrotor UAV. Int. J. Adv. Sci. Eng. Inform. Technol. 7(4), 1367–1373 (2017). https://doi.org/10.18517/ijaseit.7.4.3141 11. Maqsood, H., Qu, Y.: Nonlinear disturbance observer based sliding mode control of quadrotor helicopter. J. Electr. Eng. Technol. 15(3), 1453–1461 (2020). https://doi.org/10.1007/s42835020-00421-w 12. Wang, R., Liu, J.: Trajectory tracking control of a 6-DOF quadrotor UAV with input saturation via backstepping. J. Franklin Inst. 355(7), 3288–3309 (2018). https://doi.org/10.1016/j.jfrank lin.2018.01.039 13. Eskandarpour, A., Sharf, I.: A constrained error-based MPC for path following of quadrotor with stability analysis. Nonlinear Dyn. 99(2), 899–918 (2019). https://doi.org/10.1007/s11 071-019-04859-0 14. Najm, A.A., Ibraheem, I.K.: Altitude and attitude stabilization of UAV quadrotor system using improved active disturbance rejection control. Arab. J. Sci. Eng. 45(3), 1985–1999 (2020). https://doi.org/10.1007/s13369-020-04355-3 15. Ma, D., Xia, Y., Li, T., Chang, K.: Active disturbance rejection and predictive control strategy for a quadrotor helicopter. IET Control Theory Appl. 10(17), 2213–2222 (2016). https://doi. org/10.1049/iet-cta.2016.0125 16. Abdelmaksoud, S.I., Mailah, M., Abdallah, A.M.: Disturbance rejection for a quadrotor using robust active force control with genetic algorithm. Int. J. Model. Ident. Control 36(3), 200–210 (2020). https://doi.org/10.1504/IJMIC.2020.116915 17. Sabzehmeidani, Y., Mailah, M., Hing, T.H., Abdelmaksoud, S.I.: A novel voice-coil actuated mini crawler for in-pipe application employing active force control with iterative learning algorithm. IEEE Access 9, 28156–28166 (2021). https://doi.org/10.1109/ACCESS.2021.305 8312 18. Abdelmaksoud, S.I., Mailah, M., Abdallah, A.M.: Practical real-time implementation of a disturbance rejection control scheme for a twin-rotor helicopter system using intelligent active force control. IEEE Access 9, 4886–4901 (2021). https://doi.org/10.1109/ACCESS.2020.304 6728 19. Abdelmaksoud, S.I., Mailah, M., Abdallah, A.M.: Sensitivity analysis of intelligent active force control applied to a quadrotor system. In: Al-Emran, M., Al-Sharafi, M.A., Al-Kabi, M.N., Shaalan, K. (eds.) ICETIS 2021. LNNS, vol. 322, pp. 153–163. Springer, Cham (2022). https://doi.org/10.1007/978-3-030-85990-9_14 20. Abdelmaksoud, S.I., Mailah, M., Abdallah, A.M.: Robust intelligent self-tuning active force control of a quadrotor with improved body jerk performance. IEEE Access 8, 150037–150050 (2020) 21. Wang, C., Song, B., Huang, P., Tang, C.: Trajectory tracking control for quadrotor robot subject to payload variation and wind gust disturbance. J. Intell. Rob. Syst. 83(2), 315–333 (2016). https://doi.org/10.1007/s10846-016-0333-4

256

S. I. Abdelmaksoud et al.

22. Han, Q., Liu, X.: Robust I&I adaptive control for a class of quadrotors with disturbances. IEEE Access 8, 216519–216528 (2020). https://doi.org/10.1109/ACCESS.2020.3041030 23. Shi, D., Wu, Z., Chou, W.: Generalized extended state observer based high precision attitude control of quadrotor vehicles subject to wind disturbance. IEEE Access 6, 32349–32359 (2018). https://doi.org/10.1109/ACCESS.2018.2842198 24. Zhou, Z., Wang, H., Hu, Z., Wang, Y., Wang, H.: A multi-time-scale finite time controller for the quadrotor UAVs with uncertainties. J. Intell. Rob. Syst. 94(2), 521–533 (2018). https:// doi.org/10.1007/s10846-018-0837-1

Semantic Analysis of Moving Objects in Video Sequences Emad Mahmood Ibrahim1(B) , Mahmoud Mejdoub2 , and Nizar Zaghden3 1 National School of Electronics and Telecommunication of Sfax, University Sfax, Sfax, Tunisia

[email protected]

2 Faculty of Sciences of Sfax, University Sfax, Sfax, Tunisia 3 Higher School of Business, University Sfax, Sfax, Tunisia

Abstract. On the subject of using moving object analysis to find a specific item or replace a lost object in video sequences, numerous studies and papers have been written. To accurately describe each meaning and track down the behaviors of moving objects is a difficult task for semantic analysis investigations. In order to describe a clear text, some machine learning algorithms have looked at the proper interpretation of scenes or video clips. In order to translate visual patterns and characteristics into visual words, the paper makes use of dense and sparse optical flow methods. This study’s goal is to more easily evaluate moving objects utilizing the Gaussian Mixture Model (GMM) and baseline methods (Lagrange et al. 2017). A lot of moving pieces in the video sequence are synced together for tracking and mapping poses. Additionally, this study’s goal is to use two datasets to evaluate the proposed model on people and other moving objects (videodataset.org and HumanEva). The findings should be displayed along with the diagnosis of the moving object and its synchronization with video sequences after looking at the map or tracking an object. Personalized Depth Tracker and GMM Procedures are supplied for the paper. In order to convey the emotional states of the moving objects, the paper first identifies, diagnoses, and then describes them. The emotional states are reflected in people’s cheerful or sad faces, or in objects like cars and bicycles that move quickly or slowly. The proposed model’s goals were developed using a machine learning technique that integrates the validity of obtaining the necessary moving components with the realism of the delivered results. Because it is an easy-to-use and very effective platform, the project was carried out on the MATLAB and Python versions 3.7 platforms. Keywords: Semantic · Moving Objects · Gaussian Algorithm

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 257–269, 2023. https://doi.org/10.1007/978-3-031-20429-6_25

258

E. M. Ibrahim et al.

1 Introduction Moving objects (like cars, or human or animals) in certain video sequences include titles and descriptions for that video. These objects are moved either extemporaneously or not, as well as these objects move straightly or in cursives. Also, these objects may not clearly visible because when we downloading them may have some noises (Varghese 2021). The internet has a hug stream of pictures and videos which they need to handle them by scheme for analysis and understanding automatically. In the figures below depict one meaning that the players will jump but from left to right the first picture the play diving to water and the second-long jump and the last is tumbling (Fig. 1).

Fig. 1. Explains the moving objects that are mean the three types of jumps (Li 2010)

Semantic Analysis is a way to explore a meaning or concept of any picture or video that is saved inside it but how a computer will figure out and understand all aspects of certain video to convert them into objects for completion its story. (Moradi 2020). There are many challenges that face Semantic Analysis with video Sequences for automation and establishing an architecture of certain video that will have to analyze accurately. These challenges will be focused to find the optimal solution because the automation needs algorithms and techniques in ontology that produce results in high performance of applications in search engines, video recommenders, and video summarizers. So, the semantic analysis deals with features of video and the related of these features that contain colors, edges, arcs… Etc. (Greco 2021). The work of paper deals with the challenges of semantic video analysis and processing by using computer devices to retrieve an accurate information through searching process. The paper will utilize Semantic analysis and build a model in order to diagnose the moving object and indicate its state of the object like speed, feelings of human beings and some others states, like human beings’ behaviors when see a certain scene in live. The paper’s scope of work covers the following areas: dependence on movement objects or segments in video sequences, rely on the coordinates of each object and their features, ignore the fixed and repeated objects like a picture’s background, the paper will focus to a speed of movement objects and evaluate the proposed model via analysis of videos. The paper will take forward to the concept of machine and deep learning systems.

Semantic Analysis of Moving Objects in Video Sequences

259

The paper is expected to enhance the operation of analysis for movement objects in video sequence in order to create a new model to analyze these objects and diagnose them as human behaviors. The paper includes three levels, first level is features of moving objects, second level is a classification level by ontology techniques, and third level is labeling or tag generation.

2 Related Works There are many studies in video processing that were worked for analysis moving objects by semantic algorithms to recognize these objects to utilize them in retrieval operations. These studies are mentioned as the following as: 1. Zhu et al. (2021): They presented the abilities of methods for deep video anomaly detection models in a lot of scenarios and technical issues which handled by deep learning for saving human resources in a large degree, reducing financial burden on the government, and identifying the anomalous behaviors timely and accurately. The paper’s model and the Jing’s future aspects compared to integrate a video detection system. He utilized these aspects to implement and enhance the detection technique to be suitable with his proposed model. 2. Moradi (2020): They used visual patterns and features of vehicles behaviors in crowded traffic to transform them to visual words by using dense and sparse optical flow and learning traffic motion patterns with group sparse topical coding (GSTC) algorithm that is divided a video into scenes and then find the categories of each scene for extracting by using dual TV-L1 as intensive optical flow and Lucas–Kanade as a sparse optical flow and changed to flow words. These words are classified to be patterns for learning traffic of motion patterns. The experimental results show that the proposed algorithm and dual TV-L1 extracts more traffic motion patterns than the GSTC + Lucas–Kanade algorithm. The paper is made the same methodology but it is used a Gaussian method to detect the visual moving objects and understanding it behavior by using track system. The machine learning system supports model to present the diagnosis of moving object status. 3. Greco (2021): the researchers presented the description of semantic web technologies abilities for enhancing the efficiency of available algorithms and their results for enabling the functions of video analysis more advanced. The researcher presents also a classification of SW technology for video analysis to explore the increasing in the use of Semantic Web technologies to detect and activate many advance techniques that allows to deduce the complex and ambiguous videos.

260

E. M. Ibrahim et al.

The paper is utilized the paper to explore that his model is walking in a right path or not and it is compared the results with the results of paper. 4. Zhang et al. (2017): They presented the Martial Arts, Dancing and Sports dataset (MADS). They utilized MADS dataset which consists of five depth data that had been enhanced the quality of human pose estimation. They tested and corrected baseline algorithm on their MADS dataset, including APF-based Bayesian tracker, twin Gaussian processes, PDT, and a GMM-based tracker. The paper is used the same dataset and tested the proposed model to improve it and integrated with mapping method to recognize and named the similar tasks such as dance, sport and treatment. 5. Saad (2012): they proposed a system for semantic annotation movements in video. The system depended on the interim segmentation method that parts the movement objects in unchanged scenes. It also includes three levels of movement concepts to reduce the semantic gaps between each level. They proposed knowledge-based Model for exploring the movement features and their relations among them. The proposed system includes rules which are diagnosed any movement objects to classify into one of the three levels. The system enhanced the quality of annotation of movements in the videos and can discover the hidden information by reasoning video knowledge and movement’s features. The paper is proposed a model to find the synchronization of moving object and draw the map of it to know it behavior and diagnose it to indicate its status by machine learning system. 6. Ammar et.al. (2019): Combining Deep Sphere, an unsupervised anomaly detection framework, with Generative Adversarial Networks yielded a model for segmenting and categorizing moving objects in video sequences (GANs). The re-sults demonstrate that the suggested method outperforms state-of-the-art meth-ods in segmenting and categorizing moving objects from video surveillance. Using the Deep Sphere architecture, the proposed Deep DC technique locates, separates, and categorizes moving objects in video sequences. First, moving objects are seg-mented using the Deep Sphere framework. The network outputs are then thresholded to provide binary segmentation labels and morphological filtering. It has been demonstrated that Deep Sphere-based object segmentation is outper-formed by all BGS Library techniques. The proposed method collects deep fea-tures, then uses the GAN discriminator’s ability to effectively classify data to cate-gorize extracted images. The paper is made the same methodology but it is used a Gaussian method to detect the visual moving objects and understanding it behavior by using track system. The machine learning system supports model to present the diagnosis of moving object status.

Semantic Analysis of Moving Objects in Video Sequences

261

3 Background of the Proposal 3.1 Semantic Analysis Methods Any video in the world includes a sequence of images that have many objects. These objects are move and the rest are not. So, the semantic technology will summarize these objects to refer to the story of video. The challenge is extracting these objects automatically as the figure below and then will analyze each case individually to give them with certain labels or titles in order to retrieve these objects easily. (Otani 2016). The main techniques of Semantic analysis include to two methods are mentioned on the table below (Abirami and Gayathri 2016, Table 1):3.2 Features of Video All digital videos contain a chain of images therefore the features of image have the same features of video. These images have information such as lines, curves, and colors that are features of images. The features of videos are divided into many objects in global feature but some of these objects are moving and the rest are fixed such as (background). The global features consist of colors, shapes, and textures. An image is using colors for representing itself and these colors are also partitioning to present a segmentation of images. And the local features consist of the edge pixels, corners, or small image (object) that refer to pattern or special structure found in image of video. The models of video segmentation concerning with the moving objects and left its background in a vision time so that the local and global features are obtained from parts of video in the emotion that global features are combined different of parts of a video when extracted and while local feature are descriptions of a part of video. 3.3 Extraction of Video Features Extraction of video features is a process for collecting or indexing the number of segments that describe the important objects and refer to the subject of video. There are many types of representations to content features. One illustration is the behavior of moving items in a film, which is a smooth distribution of the moving objects. Accordingly, the model will describe the movements that happened what does it means? The features should be cleared enough for present a high quality of description in order to save in database. (Khalid et al. 2021). The database of extraction is collected and indexing them to be useful for handling other stages like diagnosing and machine learning methodologies. (Ammar 2020). 3.4 Machine Learning at Video Processing Machine learning is a process of recording a text or action so that computer retrieves them when user wants to write it again. Machine learning consists of two categories: Supervised learning and unsupervised learning. Supervised or Semi-Supervised learning uses a training set with correct targets and the method is tuned based on the training set to perform more accurate. Unsupervised learning does not have any correct targets and

262

E. M. Ibrahim et al. Table 1. The main techniques of Semantic analysis

Semantic analysis models

Techniques

Classification model Topic classification

Sentiment analysis

Advantage

Limitation

Text mining techniques, Not accuracy enough such as entity to retrieval process recognition which aims to generalize to previously unseen terms and expressions, (ii) text-level recognition of severity indicators, such as international travel or the contamination of blood products, (iii) ontology-based inferencing to fill in the gaps, e.g., between a mentioned pathogen and the unmentioned disease that caused it or between symptoms and diseases and (iv) direct knowledge of term equivalence within and across languages Is a one of machine learning method to find out negative or positive of people reactions

Complex approach and it needs a hug of dataset

Classification model Intent classification

It is high level of Difficult to classify classification. The paper accurately is needed to reach to this level

Extractor model

Video extraction

Reduce dataset for all similar video of related features is retrieved

Similarity measurement and video retrieval perform many times so it increases calculation

Entity extraction

Reduce dataset for all similar video of related features is retrieved

Similarity measurement and video retrieval perform many times so it increases calculation (continued)

Semantic Analysis of Moving Objects in Video Sequences

263

Table 1. (continued) Semantic analysis models

Techniques

Advantage

Limitation

Automatically classifying tickets

The method allows tickets to focus on tasks that provide more value while improving the whole customer experience

It’s very slow to execute all tickets because it should handle them one by one

Insights derived from data

This method supportive to detect areas of enhancement and make decisions

It is complicated to implements

the method instead tries to identify similarities to be able to categorize similar input together. (Pollyanna Gonçalves 2014). Unsupervised machine learning model recognize irregular human behaviors by learning normal behaviors. It is also learned the state in moving objects by modelling the dynamics and interactions of its features. The advantage of this model is that it can present the description of its reasoning of visualization of identical factors (Morais 2019). 3.5 Diagnosing Moving Object Status The paper deals with 2D video sequence that provides baseline results using representative tracking algorithms. For single view and multi view tracking. The paper deals with robust Bayesian tracker and twin Gaussian processes as the baselines for generative and discriminative method respectively. The results of above, the proposed model will implement Gaussian Mixture Model (GMM) estimation algorithm and will present the various baseline methods. (Zhang et al. 2017). The model executes many algorithms to find out the correct moving object status such as, Luxand, BioID.. etc. These algorithms are implemented to find human’s face and recognize it status but not the all body. The model will implement a mapping algorithm that will indicate the features during training database as keys of descriptions to indicate the status of moving objects. (Ha et al. 2020). The paper is comparing with an article or paper about diagnostic moving object in any video sequence but we didn’t find a paper closely. The paper is attempted to deduce some methods close to the model to find the solutions.

4 Practical Part The paper is covered many functions that are mentioned above, and also the most important function that is automated the detection and diagnosing stages to explore a status

264

E. M. Ibrahim et al.

of moving object in the video sequences. Some of smart systems detect moving objects but did not diagnose the status of these objects. The model consists of four main phases that represent the functions and processes to handle the input video. In this section, the plan of proposed model will describe all its parts in details and present all techniques that will implement in the model to get a high quality of results. The parts are depicting in Fig. 2 with its description as following as:

Input Video

Extraction features

Object Detection

Object Diagnosing

Machine learning Fig. 2. Phases of model Architecture

4.1 Input Video There are two public datasets, will be used as a source for measured (raw) data, the first dataset (HumanEva) (Sigal 2010) includes human action such as sport and dance movements, while the second dataset (videodatasets.org) include additional features such as smile, sad faces. The selected input videos have been captured at 25 FPS with resolution (360 × 288) in different places and times. The two datasets are randomly taken for the proposed model to find out more cases and different patterns of moving objects. Figure 3a is an example of the paper’s hypotheses. The paper has been attempted about 20 clips which are included around 15K frames per clips as case studies for extracting best results by the proposed model as shown in Fig. 3b. Based on the available dataset, additional calculated features will be extracted and added to an extended dataset for the experimental evaluation.

Semantic Analysis of Moving Objects in Video Sequences

a.

Input Image

b.

265

Output Image

Fig. 3. Depicts two images the first is an input image that used to extract and identify some required features of moving objects.

4.2 Extraction Feature The purpose of the paper is to uncover potential quantifiable qualities that are accessible on computers and to compare its findings to those of previous studies. The datasets contain two different kinds of characteristics. The first step in the segmentation process is to prepare and process the raw data features. Second, calculated features, such as total time, total distance, average time between video sequences, and average distance of moving objects, are features that are derived from raw data features. The challenge of extraction process has been dealt with a lot of moving object which is mentioned previously. Many papers dealt with a specific object such as human movements (Sport, Dance, fighting..etc.) or traffic of vehicles and how can analyze their movement to avoid a certain problem. The strong of proposed model is handling different moving objects and finds out their features. The proposed model used GMM equations [1, 2] p(x) =

K 

∅N (x|μi , σi )

(1)

i=1

  1 (x − μi )2 N (x|μi , σi ) = √ exp − 2σ 2i σi 2π K 

(2)

∅i = 1

i=1

A Gaussian Mixture Model with K components, μk is the mean of the kth component. Furthermore, a univariate case will have a variance of σk whereas a multivariate case will have a covariance matrix of k. k is the definition of the mixture component weights  which is for every component k. This has a constraint that Ki = 1φi = 1 such that the total probability gets normalized to 1. There are two variables (μ = 2, σ = 4) that had been focused to learn the proposed model. Many videos (type MP4) performed the source code (Gaussian Distributed) of detecting multi-moving objects, each object has many features that are categorized to be separated object. The Fig. 3 is shown working environment and presented the estimated

266

E. M. Ibrahim et al.

parameters. The figure is compared with other techniques that deal with same datasets in order to adjust the comparison. Table 2. Primary result of detection techniques Datasets

CCP [17]

LPB [18]

GMM

Notes

Video1

100.00

100.00

100.00

No noise

Video2

82.50

80.95

99.51

20% noise

Video3

62.00

61.76

62.63

40% noise

Video4

40.50

40.78

41.23

60% noise

Video5

21.50

21.21

32.22

80% noise

Table 2 presents the primary results that depends on GMM iterations for correction or converge to the best boundaries and to reach to high level of segmentation. The videos include ratio of noises in order to explore the behavior of each algorithm. GMM presented an accepted result because it is trying to find a best Gaussians moving objects. The results of proposed model compared with Conventional Neural Network (CNN) and LBP algorithms because they have different techniques (Li et al. 2015). The purpose of comparison to find the best results. The noise of video1 is very low so that the results are very close. The over videos have high noises so that the results are different because Gaussian algorithm explores to the best detection. 4.3 Object Detection Figure 4 describes an Object detection model that is a process to find out required object by using one of clustering techniques after framing process to find (background and foreground). Gaussian mixture is a process to indicate the related pixel’s location for making an object classification. Moving objects depend on tracking of the serial images to make a map of the object movement. The results of classification process are presenting the moving object proprieties that are including the physical such as the speed of objects and logical proprieties such as the object relationship with other frames. 4.4 Object Diagnosing The paper is built an algorithm that depends on how many Gaussians existed inside the video to diagnose by using comparison method so that the model will depend on database include almost all the status of moving objects. The model includes analysis and estimation functions that will present the ratio of errors. The ratio of error will be convergent to the correct object or divergent of the object itself. The model measures the speed of frame per second and the synchronization of moving object and its relativity to find suitable equations in order to calculate these categories for presenting a high level of diagnosis.

Object classification

Background model

Fig. 4. Describes the model of object detection

Gaussian mixture

Feature Extraction

Objects detection

Frames separations

Input Video

Semantic Analysis of Moving Objects in Video Sequences 267

268

E. M. Ibrahim et al.

4.5 Machine Learning According to the description in the subject 4.4, The paper will implement unsupervised machine learning model that is described in the Fig. 5. The model will execute by Python platform because the last includes powerful libraries and easy to construct the parts of model.

Unsupervised Machining Learning

Clustering Analysis

Gaussian Mixture Model Fig. 5. The proposed model—unsupervised machine learning model

For example, moving objects have some data points, such as those represented by the objects in the picture below, the GMM algorithm may assign labels to those clusters, which is comparable to what human being see with eyes. If human beings have basic objects of data, for example, the GMM algorithm can swiftly label such clusters in a style that is similar to what they might accomplish by eye.

5 Conclusions The predictions of the Gaussian Mixture Model for automated driving are assessed based on feature extraction. The paper is investigated the temporal consistency of semantic moving objects in this research. The paper is put up and defined the problems with temporal consistency using one sequence of unlabeled images. In order to comprehend the semantic and labeling process, we first developed a GMM. The paper is assessed an unsupervised technique to ascertain the precision of semantic prediction based on video sequences. In order to identify and categorize certain moving objects, we performed pixel analysis using a Python tool. The module recognizes moving objects using the videodatasets.org dataset. By using the machine learning GMM algorithm, we attempted to improve the identification scenario in order to produce detection and decisions for the right semantic of moving objects. The paper should increase the pace of its procedures in future.

Semantic Analysis of Moving Objects in Video Sequences

269

References Abirami, A.M., Gayathri, V.: A survey on sentiment analysis methods and approach. IEEE 72–76 (2016) Ammar, S., Bouwmans, T., Zaghden, N., Neji, M.: Moving objects segmentation based on deepsphere in video surveillance. In: International Symposium on Visual Computing, pp. 307–319. Springer, Cham (2019) Ammar, S., Bouwmans, T., Zaghden, N., Neji, M.: Deep detector classifier (DeepDC) for moving objects segmentation and classification in video surveillance. Surveillance 1490–1501 (2020) Greco, L.: On the use of semantic technologies for video. Semantic-web-journal. Retrieved from http://www.semantic-web-journal.net/system/files/swj1789.pdf (2021) Ha, S.V.U., Chung, N.M., Phan, H.N., Nguyen, C.T.: TensorMoG: a tensor-driven gaussian mixture model with dynamic scene adaptation for background modelling. MDPI Sensors 1–29 (2020) Hamouda, M., Bouhlel, M. S.: Modified convolutional neural networks architecture for hyperspectral image classification (Extra-convolutional neural networks). IET Image Processing. (2021). https://doi.org/10.1049/ipr2.12367 Khalid, N., Ghadi, Y.Y., Gochoo, M., Jalal, A., Kim, K.: Semantic recognition of human-object interactions via gaussian-based elliptical modeling and pixel-level labeling. IEEE 111249– 111266 (2021) Lagrange, A., Fauvel, M., Grizonnet, M.: Large-scale feature selection with Gaussian mixture models for the classification of high dimensional remote sensing images. In: The French National Research Agency (ANR) under Project Grant ANR-13-JS02–0005–01 (Asterix project) (2017) Li, H.: Automatic detection and analysis of player action in moving background sports video sequences. IEEE 351–364 (2010) Li, S., Liu, Z.-Q., Chan, A.B.: Heterogeneous multi-task learning for human pose estimation with deep convolutional neural network. Int. J. Comp. Vis. (IJCV) 113(1), 19–36 (2015) Moradi, A., Shahbahrami, A.: An unsupervised approach for traffic motion patterns extraction. In: IET Image Processing, pp. 428–442. (2020) Morais, R., Le, V., Tran, T., Saha, B., Mansour, M., Venkatesh, S.: Learning regularity in skeleton trajectories for anomaly detection in videos. In: IEEE/CVF Conference on Computer Vision and Pattern Recognition, p. 11 996–12 004. Long Beach, CA, USA, IEEE, 2019 Otani, M.: Video summarization using deep semantic features. In: Asian Conference on Computer Vision, pp. 361–377. Oulu, Spring (2016) Pollyanna Gonçalves, M.A.: Comparing and combining sentiment analysis methods. ACM 1–11 (2014) Saad, S.: Semantic analysis of human movements in Videos. ACM 141–148 (2012) Sigal, L., Balan, A.O., Black, M.J.: Synchronized video and motion capture dataset and baseline algorithm for evaluation of articulated human motion. Int. J. Comp. 4–27 (2010) Varghese, S.: An unsupervised temporal consistency (TC) loss to improve the performance of semantic segmentation networks. In: CPVR, pp. 1–9. Germany, IEEE (2021) Zaghden, N., Khelifi, B., Alimi, A. M., Mullot, R.: Text recognition in both ancient and cartographic documents (2013). arXiv preprint arXiv:1308.6309 Zhang, W., Liu, Z., Zhou, L., Leung, H., Chan, A.B.: Martial arts, dancing and sports dataset: a challenging stereo and multi-view dataset for 3d human pose estimation. Image Vis. Comp. 61, 22–39 (2017) Zhu, Y., Zhang, L., Chen, Q., Xiao, W.: Opportunities and challenges. In: Song, H., Yao, J. (eds.) In Vitro Diagnostic Industry in China, pp. 11–16. Springer, Singapore (2021). https://doi.org/ 10.1007/978-981-16-2316-5_2

Improved Automatic License Plate Recognition System in Iraq for Surveillance System Using OCR Yasir Dawood Salman1(B) , Hussam S. Alhadawi1 , Ahmed Salih Mahdi2 , and Fahad Taha AL-Dhief3 1 Computers Technologies Engineering Department, Dijlah University College, Baghdad, Iraq

[email protected] 2 Computer Engineering Technology, Al-Nisour University College, Baghdad, Iraq 3 School of Electrical Engineering, Department of Communication Engineering, Universiti

Teknologi Malaysia, UTM, Johor Bahru, Johor, Malaysia

Abstract. Video surveillance is a system for continuously monitoring the surrounding environment and processing the gathered foreign data. Some common implementation areas are traffic surveillance, vehicle monitoring, license plate recognition, and motion detection. This research suggests a technique for identifying and detecting automobile license plates in Iraq. Every aspect of the system’s logic is based on morphological operations and the OCR edge detection approach, with the ultimate goal of constructing and developing efficient image processing methods and strategies to place a license plate. The technology may benefit various sectors, such as highway speed detection, automated charging systems, security, manuscript papers, and lighting infractions. A combination of hardware, software, and auto plate recognition can read a vehicle’s license plate and determine its owner’s identity. Because of its numerous impacts, such as light and speed, automatic license plate recognition (ALPR) has a lot of negative implications. To compare, the accuracy of our findings is higher than that of existing systems that use plate detection and character segmentation as part of their character recognition process. Keywords: Character segmentation · Image processing · License plate · Character recognition · Optical character recognition (OCR)

1 Introduction Many industries, including remote sensing, robotics, industrial automation, and virtual reality, employ the technique of extracting usable information from images and videos [1]. Numerous algorithms and methods have been utilized to extract data from images. The Automatic License Plate Recognition (ALPR) is well-known for identifying license plate numbers from moving vehicles [2]. Additionally, this technology is becoming more common in traffic and security systems. Security has recently been a top priority due to the rise in terrorist operations worldwide. More than ever, there is a necessity for security-related services, and there is a © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 270–277, 2023. https://doi.org/10.1007/978-3-031-20429-6_26

Improved Automatic License Plate Recognition System in Iraq for Surveillance

271

pressing need to use information technology to develop new means of protecting people or enhancing existing ones [3]. Computer-controlled automated surveillance systems that might operate autonomously with little to no human involvement have garnered attention [4]. A computerized system that might detect suspicious vehicles passing by may send out notifications or promptly report such incidents to the appropriate authorities. This technology will accelerate response times and may even save people. This study aimed to create a real-time program to identify license plates of passing vehicles at gates, such as parking lots and border crossings [5]. The video system is built on a typical PC that has a camera. It records and analyses video frames with a clearly visible automobile license plate. Once a license plate is located, its digits are identified and either presented on the user interface or crossreferenced with a database. The main topic of discussion is the design of the algorithms that are used to separate the characters from the license plate and identify each individual character. The following categories may be used to categorize techniques for locating the license plate area in pictures or films from earlier works: Binary Images Processing, Gray-Level Processing, Colour Processing, and Classifier [6]. Prior to character recognition, character segmentation is an equally significant stage. Character segmentation techniques may be divided into three categories: Binary Image Processing, Gray-Level Processing, and Classifiers. Several algorithms that match patterns or templates or classify data based on learning have been developed to identify the segmented characters [7, 8]. This study focuses on the issue of real-time license plate recognition using images and videos. The approach will aid in the identification and registration of vehicles as well as serve as a point of reference for activity and vehicle monitoring in the future. There are two primary stages to our paper license plate identification method. Firstly, to extract specific features from videos that encode the frames or images. Secondly, to create a classifier or detector to identify whether a specific area in the frames or images is the license plate.

2 Related Works Number plate recognition, also known as “license plate realization” or “recognition” using image processing techniques, is one of the main areas of image processing study. This identification is a possible study topic in smart cities and the Internet of Things [9]. Most current solutions for the ALPR task have reportedly a multi-stage strategy, which comprises three primary phases. The identification or extraction of license plate data is the initial phase. Existing systems combine object identification with deep learning and conventional computer vision approaches to find the license plate in an image [10]. The primary characteristics of the license plate, like shape, texture, color, and symmetry, are the basis for traditional computer vision approaches [11]. The characters are retrieved from the license plate in the second phase utilizing specific widely used methods, including vertical and horizontal projection, mathematical morphology [11], linked components, and relaxation labelling [12]. Unfortunately, not all multi-stage ALPR systems complete the character segmentation phase since specific segmentation-free algorithms ignore this phase.

272

Y. D. Salman et al.

In the pre-processing phase, Ganta and Svsrk [13] apply many image processing methods, including Gaussian smoothing, Gaussian thresholding, and morphological transformation. For number plate segmentation, contours are applied using border following, and contours are filtered based on character dimensions and spatial localization of the number plate characters. Good results were obtained when extracting license plate areas from background images using algorithms based on a mix of morphology and edge statistics. After smoothing and normalizing a grayscale image, Hongliang and Changping [14] used edge operators to obtain horizontal and vertical edge maps. Subsequently, edges were statistically analyzed to identify the license plate’s rectangle. The technique was carried out hierarchically at several scales. Following the rule-based fusion, many license plate areas remained. The eventual decision relied on the CCA (connected components analysis). They asserted that their method could obtain a 99.6% identification rate over 9825 images. In license plate detection, image transformation techniques based on the Hough transform, wavelet transform, and Gabor filters have been used. Hough transform is a standard detecting straight lines since lines may specify a license plate’s shape. Zheng et al. [15] detected the border of a license plate using the Hough transform. The format of license plates is tightly enforced in many countries and regions. Since the color of the text and backdrop are unchanging, several algorithms utilize color information to recognize license plates [16]. Lastly, following the interested region filtering and de-skewing, the K-nearest neighbor method is utilized for character recognition. In another research by Keraf et al. [17], they put the system at the main entrance to guarantee that only authorized cars may access the university area. The input sensors recognize the car by collecting the image of the vehicle plate number. Optical character recognition is used to extract a character segment from an image. The approach to detecting a license plate is the pre-processing image and utilizing a mix of Sobel Edge Detection and Laplacian Edge Detection Techniques. The NP may be found using the Bounding Box method, and characters can be recognized.

3 Iraqi License Plate The Iraqi license plate requirements are discussed in this section. Iraqi license plates were first issued in their most current iteration in 2022. According to Fig. 1, the design and dimension of these plates are based on European norms. The shape, size, and position of the digits on the plate that identifies the vehicle are the standards, as seen in the figure. The plates are composed of aluminum and coated with reflective phosphorous tape. These plates were manufactured utilizing an extreme pressure method to print the characters. The governorates of Iraq will be symbolized by a code rather than their complete names on the new license plates, which will be written in the Latin alphabet.

4 Methodology ALPR is a computer vision technique that utilizes license plates to detect vehicles. For security or traffic applications, including parking lot access control, traffic surveillance,

Improved Automatic License Plate Recognition System in Iraq for Surveillance

273

Fig. 1. An example of an Iraqi private vehicle license plate.

and information management, ALPR has become a frequently utilized fundamental technique in latest years [5, 18]. This study’s technique and application of ALPR detection were based on OCR. It was implemented using MATLAB 2020a. Figure 2 displays a logical block representation of the whole procedure.

Input Image

License Plate Extracon

Image Binarizaon

Character Recognion

Plate Number Save

Fig. 2. A logical block diagram of the entire process

The research algorithm was developed to identify the car license plates. Figure 3 illustrates a vehicle’s license plate number as the system input. The segmentation part receives its output from this image’s optical character recognition (OCR) processing. The recovered plate region is processed further during the segmentation step, the image characters are separated, and the data is saved as a collection of numbers and letters in an Excel spreadsheet. According to the suggested system, a camera takes a license plate image, which is then processed as follows: a. The acquired image is first uploaded into MATLAB, as illustrated in Fig. 3, and then passed to the image processing part. b. Using the “rgb2gray()” function tool, the image was converted to grayscale, as illustrated in Fig. 4. c. The third step was a noise-removing filter. The transformed photos were noise-free using 2-D median filtering. d. The function “imbinarize()” was run to transform the image into a binary image before segmentation could begin. As illustrated in Fig. 5, this function eliminates the lowest 50% of pixel values and then uses Otsu’s approach to calculate a threshold on the normalized image. e. OCR Character Segmentation is used to segment and detect license plate information. The number plate region has just been divided up in this phase by employing the bounding box technique. The identified characters are then turned into a string of characters using template matching. f. The characters are taken from the identified number plate and recorded in an Excel spreadsheet at the final stage. As a result, it would be readable to identify the license plate.

274

Y. D. Salman et al.

Fig. 3. Read Image.

Fig. 4. Convert the images to grayscale.

Fig. 5. Convert the images to binary image

A. The Input Image A high-resolution camera captures the vehicle’s license plate. According to the chosen image, the plate recognition system’s resolution changes. The RGB image must thus be converted into a grayscale image.

Improved Automatic License Plate Recognition System in Iraq for Surveillance

275

B. Pre-Processing Before processing, the image is improved using various methods to convert it from a grey image to a binary one. Before being transformed into a binary image, the image is processed to reduce noise. Pre-processing may be performed by employing the 2-D filter. C. Localization of Number Plate A form inspection or a color sample selection is performed to extract the licensing platform. The algorithms are rectangular because the General License Panel is rectangular [19]. Iraq’s new main licensing plate is white; therefore, using black and white images is the most appropriate direction. Before locating the rectangle in the image, the method might either have a binary image size or recognize the image’s borders. The subsequent step is identifying and connecting to the relevant rectangle corners [20]. The last step joins all regions connected to the rectangle and eliminates all rectangular areas.

5 Results The following equation is used to determine the total plate recognition efficiency for both the old and improved ALPR systems: Accuracy =

Number of correctly recognized plates × 100% Total number of plates

It should be emphasized that the accuracy of each phase is strongly influenced by the outcomes of the stage before. As a result, an integrative or chain linkage is developed across the whole system. The accuracy of the system was 96.3% of the total inputted images. The system may successfully or unsuccessfully identify the correct characters on the license plate depending on the outcomes of the last step of recognition; the failure might happen when any character is wrongly read. Table 1 displays examples of input photos together with the related detections and recognition outputs. The ALPR algorithm shows a high accuracy rate of detecting the vehicle’s plate numbers, especially for the new Iraqi plate numbers since they were recently changed in Iraq at the time this research was conducted. The system will store the extracted plate number in an excel sheet that can be converted to a database in future work to compare the results to use it for security purposes or any other fractionally.

6 Conclusions In this research, application software is built for Iraqi license plate recognition. It offers an understanding of the technical features of important detection methods. To sum it all up, the captured image is first captured, and template matching is used to extract the image characters. In the future, it is suggested that methods for generating intrinsic images (such as illumination, reflectance, and depth images) from captured images or a collection of input images be explored. Nevertheless, the overall performance of the established

276

Y. D. Salman et al. Table 1. Examples of input images with associated detection and recognition outcomes.

Vehicle image

Detection output

Vehicle plate number 100067EBL

229354SUL

685329EBL

ALPR system has potential for improvement via more sophisticated machine learning algorithms that could learn and make intelligent decisions on their own instead of making informed decisions depending on what they have achieved. The authors’ team is presently investigating these and other potential improvements.

References 1. Salman, Y.D., Ku-Mahamud, K.R., Kamioka, E.: Distance measurement for self-driving cars using stereo camera. In: International Conference on Computing and Informatics. (2017) 2. Mufti, N., Shah, S.A.A.: Automatic number plate recognition: a detailed survey of relevant algorithms. Sensors 21(9), 3028 (2021) 3. Reddy, G.S., et al.: Real time security surveillance using machine learning. (2019) 4. Sung, C.-S., Park, J.Y.: Design of an intelligent video surveillance system for crime prevention: applying deep learning technology. Multimedia Tools Appl. 80(26), 34297–34309 (2021) 5. Anagnostopoulos, C.-N.E., et al.: License plate recognition from still images and video sequences: a survey. IEEE Trans. Intell. Transp. Syst. 9(3), 377–391 (2008) 6. Yohimori, S., et al.: License plate detection system by using threshold function and improved template matching method. In: IEEE Annual Meeting of the Fuzzy Information, 2004. Processing NAFIPS’04. IEEE, (2004)

Improved Automatic License Plate Recognition System in Iraq for Surveillance

277

7. Quan, S., Shi, Y., Xue, Z.: A fast license plate segmentation and recognition method based on the modified template matching. In: 2009 2nd International Congress on Image and Signal Processing. IEEE, (2009) 8. Li, X., et al.: Research on the recognition algorithm of the license plate character based on the multi-resolution template matching. In: 4th International Conference on New Trends in Information Science and Service Science. (2010) 9. Kumar, J.R., Sujatha, B., Leelavathi, N.: Automatic vehicle number plate recognition system using machine learning. In: IOP Conference Series: Materials Science and Engineering. IOP Publishing, (2021) 10. Shashirangana, J., et al.: Automated license plate recognition: a survey on methods and techniques. IEEE Access 9, 11203–11225 (2020) 11. Zou, L., et al.: License plate detection with shallow and deep CNNs in complex environments. Complexity 2018 12. Budianto, A.: Automatic license plate recognition: state of the art and the opportunities for implementation in Indonesia (Literature Review). J. Telemat. Informat. 6(4) (2018) 13. Ganta, S., Svsrk, P.: A novel method for Indian vehicle registration number plate detection and recognition using image processing techniques. Proc. Comp. Sci. 167, 2623–2633 (2020) 14. Hongliang, B., Changping, L.: A hybrid license plate extraction method based on edge statistics and morphology. In: Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR 2004. IEEE, (2004) 15. Zheng, D., Zhao, Y., Wang, J.: An efficient method of license plate location. Pattern Recogn. Lett. 26(15), 2431–2438 (2005) 16. Suryanarayana, P., et al.: A morphology based approach for car license plate extraction. In: 2005 Annual IEEE India Conference-Indicon. IEEE, (2005) 17. Keraf, N.D., et al.: Automatic vehicle identification system using number plate recognition in POLIMAS. In: IOP Conference Series: Materials Science and Engineering. IOP Publishing, (2020) 18. Anagnostopoulos, C.N.E., et al.: A license plate-recognition algorithm for intelligent transportation system applications. IEEE Trans. Intell. Transp. Syst. 7(3), 377–392 (2006) 19. Yuan, B., Yang, Y.: Research on the algorithm of locating and cutting in license plate character extraction. J. Phys. Conf. Ser. IOP Publishing (2021) 20. Wang, C., et al.: Synthesizing large-scale datasets for license plate detection and recognition in the wild. In: Chinese Conference on Pattern Recognition and Computer Vision (PRCV). Springer, (2020)

Do We Use the Right Elements for Assurance Case Development? Abdul Rehman Gilal1(B) , Abdul Sattar Palli1 , Jafreezal Jaafar1 , Bandeh Ali Talpur2 , Ahmad Waqas3 , and Ruqaya Gilal4 1 Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, 31750

Seri Iskandar, Malaysia [email protected] 2 School of Computer Science and Statistics, Trinity College Dublin, Dublin, Ireland 3 Knight Foundation School of Computing and Information Sciences, Florida International University, Miami, USA 4 School of Computing, Universiti Utara Malaysia, Bukit Kayu Hitam, Kedah, Malaysia

Abstract. Assurance case (AC) is a well-structured document to build the stakeholders’ confidence towards the critical properties of the system. Graphical notations, such as Graphical Structural Notation (GSN), are used to graphically document the ACs. It has been identified that several research papers do not apply the graphical notations properly. Therefore, the objective of this study is to assess the compliance of graphical notations used for ACs development. We assessed the papers based on the GSN standard. We have selected few studies in which the ACs are graphically presented. The results indicate that the elements and the relationship between elements are mostly wrongly presented when the standard tools are not used. For example, in the selected studies, the authors have commonly made mistakes in denoting Assumption and Context elements and SolvedBy and InContextOf relationships. In the conclusion, the study recommends to check the “language” related issues in the ACs. Keywords: Assurance case · Security case · Safety case · GSN

1 Introduction Assurance case (AC) is a well-structured document to build the stakeholders’ confidence towards the critical properties of the system (i.e., safety or security). Based on the Bloomfield et al. definition [1], “An AC is a documented body of evidence that provides a convincing and valid argument that a specified set of critical claims about a system’s properties are adequately justified for a given application in a given environment.” ACs are mainly constructed, either textual or graphical, based on the claims, arguments, and evidence components. ACs have been studied and practiced in the recent past a lot [2–5]. For instance, safety ACs are widely used in the aviation, cyber security [6], health [7] and military domains [8, 5]. On the contrary, security AC (a.k.a. SAC) is emerging trend in the critical system development [9, 8]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 278–286, 2023. https://doi.org/10.1007/978-3-031-20429-6_27

Do We Use the Right Elements for Assurance Case Development?

279

Effective AC just not increases the confidence but also increases the quality of the overall system under consideration [10]. It is suggested in that during AC development, authors should focus the clarity, comprehensibility and veracity of arguments. One of the major problems occur when the authors of AC assume too much knowledge from the reader’s side. It can also happen that readers will have less knowledge about the system and the properties under consideration. These leaps can cause misunderstanding or concepts drifts. Therefore, it is always important to discuss the confusing terminologies if you want to get your ACs public or publish them for the public read. On the other hand, the poorly constructed AC can reduce the confidence and the quality of the system under consideration. It is important to assess the quality of AC from its development and validation point of views. Usually, the AC are evaluated by the authors and external evaluators [10]. Readers also evaluate the ACs at their ends. Readers cannot directly influence the development of the ACs. However, the quality of AC can influence the readers’ perception. Eventually, the reader’s perception would surely influence the adoptability of the AC. For instance, Chowdhury et al., [10] say that “AC must be understandable by all the stakeholders”. We consider researchers and/or reviewers play a critical role in ACs assessments. Therefore, in this study, we categorize them, researchers and reviewers, as the main stakeholders in ACs. Moreover, we have identified several research articles do not apply the graphical notations as per standard demands. For instance, Neši´c el al. [11] have used GSN to develop the ACs. However, the authors have not used the GSN recommended elements and relationships. Therefore, we set the main objective of the study is to assess the compliance of graphical notations (i.e., Goal Structural Notation) used for ACs development. We assess the compliance of the AC’s graphical notations from the published research papers. We would like to mention here that we have focused the shapes or elements of the selected studies. Our study does not claim anything about the quality of the papers. The remainder of the paper presents following sections: related work, methodology results and discussion, and conclusion and future work. Related work covers the commonly used notations used for ACs development. We present the important findings of assessments in the results and discussion section. At the end, conclusion section presents the limitation and the future directions of this study.

2 Related Work Significant research has been performed to guarantee that assurance cases can be presented, debated, and evaluated in a clear and effective manner. Numerous notations have been developed for the purpose of expressing assurance scenarios. For examples, Goal structuring notation (GSN) [12], Claims-Arguments and Evidence (CAE) [13], 2.Structured Assurance Case Meta model (SACM) [14] are the main notations which are frequently used for representing assurance cases graphically.

280

A. R. Gilal et al.

GSN is a graphical representation used to record and demonstrate that safety goals have been met, in a more comprehensible manner than plain text [31–35]. The notation is a diagram that automatically constructs its safety case via the use of logic-based mappings. The core elements of GSN are goal, solution, strategy, context and undeveloped goal, assumption and justification [15]. Within the arguments, a Goal represents a safety assertion. A strategy is a term that refers to the nature of the inference that exists between a primary objective and its supporting objective (s). A Solution is a reference to a piece of evidence or a collection of elements of evidence. A Context object represents a contextual artifact, which might be a statement or a reference to contextual data. An Assumption is a statement that is made throughout the reasoning process. A Justification is a logic statement. Undeveloped elements indicate that a line of reasoning has not been developed yet (meaning it being abstract and needs to be instantiated). The Undeveloped notation may be used to refer to Objectives and Strategies. The GSN’s core pieces are linked through two kinds of connectors: SupportedBy and InContextOf. The SupportedBy property enables the documentation of inferential or evidentiary links. The InContextOf attribute establishes a connection between contextual components (i.e. Context, Assumption, and Justification) and Goals and Strategies. When GSN components are connected in a network, they are often referred to as a goal structure. The objective of a goal structure is to demonstrate how goals are incrementally broken down into sub-goals until they can be supported directly by existing data (Solutions). GSN has been embraced by a rising number of firms within safety-critical sectors such as aircraft, railroads and military for the presenting of safety arguments inside safety cases [36–40]. The GSN is utilized on trusted execution environments (TEEs) that support a certain hardware standard, which restricts developers who use such TEEs [16]. Another study [17] used GSN to present software system safety. Claims-Arguments and Evidence (CAE) [13] is a simple yet effective notation for presenting and communicating the safety argument. It is a straightforward notation for structuring safety cases. A claim is a true/false assertion concerning a certain object’s attribute. A claim is exactly what the name implies; a statement that someone is attempting to persuade some other person is true. An argument is a set of rules that connects what we know or assume (sub-claims, evidence) to the claim under investigation. The argument that is utilized is determined by the kind, reliability, and breadth of the available evidence, as well as the nature of the claim. Evidence is a piece of art that provides verifiable facts that lead directly to a claim. The CAE has successfully been used in different application domains such as autonomous, medical, nuclear, security and transport. Recently CAE is used in block-chain based application to enhance its security and dependability [18]. The study [19] recommended CAE and GSN to be used as a framework to audit the AI based systems because AI development has generated worries about bias amplifying and loss of privacy as well as digital addictions, misinformation, and detrimental changes to the quality and availability of meaningful work. SACM [14] is a publicly available specification published by Object Management Group (OMG). SACM’s goal is to offer a modeling framework that allows users to specify and exchange argument structures. An argument’s representation in SACM does

Do We Use the Right Elements for Assurance Case Development?

281

not indicate that the argument is comprehensive, accurate, or true. Similarly, the SACM does not include the review or acceptance of an argument by a third party. Structured arguments, according to the SACM model, consist of argument components (mainly assertions) claimed by the author of the argument, as well as connections alleged to exist between those elements. The SACM is being used in verity of application domains such as finance, government, healthcare, manufacturing, military and space exploration. For example, the study [20] used SACM to construct the assurance cases to show the acceptable safety and security of materials transportation using Automated Guided Vehicles (AGVs) platooning. Among the discussed notations, the GSN is the most frequently used notation [21]. Therefore, in this paper, we manually assess the compliance of GSN graphical notions.

3 Results and Discussion This study uses the graphical notations of GSN standard to assess the elements of the published ACs. We have selected 11 papers in which the ACs are graphically presented. We have manually assessed the compliance of GSN elements in the selected papers. We have only added 11 studies which have somehow not applied the GSN standard’s elements and relationships properly. We assessed the selected papers based on the following GSN elements (Fig. 1).

Fig. 1. GSN standard-elements and relationships

GSN is a graphical notation which acts as a communication mechanism, where every symbol has its importance. For clear communication, every symbol of GSN must properly adhere the elements or shapes of shapes of elements as per standard. In case the defined standards of GSN are violated while designing the GSN based design, it may cause misinterpretation of the design as a result it may lead to wrong implementation. In the literature, there are studies where GSN standard is not applied properly. For instance, Othmane et al. [22] have developed a Security Assurance Case Plug-in (SACP) as an

282

A. R. Gilal et al.

Eclipse plug-in to design security assurance cases in the Eclipse development environment. The authors have studied a problem of continues requirement or code change and its impact over the security cases. Indeed, a full security reassessment is required if the requirements are majorly changed. This demands the change in the arguments and evidence. While discussing the ACs in a graphical form, the authors have used wrong element for evidence and wrong inContextOf relationship. Figure 2. Below highlights both errors.

Fig. 2. In context of and evidence element errors in [22]

In the figure above, goal element is supposed to have relation the context (i.e., inContextOf). However, the authors have done it other way around by using “solvedBy” relationship. Similarly, the authors were supposed to use a circle to denote the “evidences” instead the oval shape. To summarize the results, we present the element and relationships errors of the selected studies in the following Table 1. The “x” denotes the wrong use of the shapes in the selected studies. Similarly, the study [11] fails to follow GSN standards for Justification, Context, Evidence, and InContextOf elements and relationship respectively. The arrow sign of InContextOf was not instead they used the sold arrow which is the symbol of SolvedBy. In the same vein, study [27] violated the GSN standards for Strategy, Assumption, Context and InContextOf elements. From the Table 1 it can be observed that, most of the selected studies have not properly used the SolvedBy and InContextOf standards. The main aim of using graphical standards is to make communication understandable to every stakeholder of the system. The mistakes made at design level can cause understanding problems; Therefore, it is necessary to follow the defined standards to avoid the implication at implementation level.

x

Justification

x

x

Context x

SolvedBy

Ray and Cleaveland [5]

Yamamoto and Morisaki [30]

Yamamoto [29]

Othmane and Ali [22]

x

x

Austin et al. [27]

Schmid et al. [28]

x

Asa et al. [26]

x

x

x

Assumption

Matsuno and Taguchi [25] x

Strategy

x

Goal

Rich et al. [24]

Kelly [23]

Neši´c et al. [11]

Mumtaz et al. [17]

Study Ref. No

Table 1. Incorrect use of GSN element shapes in the design

x

x

x

x

x

x

x

InContextOf

x

x

x

Solution

Do We Use the Right Elements for Assurance Case Development? 283

284

A. R. Gilal et al.

4 Conclusion This study assesses the compliance of the AC’s graphical notations from the published research papers. This study focuses to assess the compliance of elements of GSN which are used in the published papers. This study does not claim anything about the quality of the selected papers. Based on the results, we found that several studies have not followed the proposed GSN standard shapes. The shapes ambiguities may lead for the wrong representation of the case. We observed that these errors are caused if the GSN based tool is not used. Moreover, this study has done the assessment manually. Automation can help us to assess the visualization effectively. The future study may also look for the “language” used in the elements. For example, the GSN standard recommends having an atomically of goal. In other words, one goal should have only one claim.

References 1. Adelard, A.: The Adelard safety case development manual. Adelard (1998) 2. Johnson, C.: Using assurance cases and Boolean logic driven Markov processes to formalise cyber security concerns for safety-critical interaction with global navigation satellite systems, vol. 45. (2011) 3. Kläs, M., Adler, R., Jöckel, L., Groß, J., Reich, J.: Using complementary risk acceptance criteria to structure assurance cases for safety-critical AI components. AISafety@ IJCAI (2021) 4. Šljivo, I., Uriagereka, G.J., Puri, S., Gallina, B.: Guiding assurance of architectural design patterns for critical applications, vol. 110, p. 101765. (2020) 5. Ray, A., Cleaveland, R.: Security assurance cases for medical cyber–physical systems. 32(5):56–65 (2015) 6. Zhou, Z., Matsubara, Y., Takada, H.: Quantitative security assurance case for invehicle embedded systems. In: 2021 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech), pp. 43–50. IEEE, (2021) 7. Andersson, A., Fransson, M.: Applying and maintaining security assurance cases in the medical domain: a case study at AstraZeneca. (2022) 8. Mohamad, M., Steghöfer, J.-P., Scandariato, R.: Security assurance cases—state of the art of an emerging approach. 26(4), 1–43 (2021) 9. Brezhniev, E., Ivanchenko, O.: NPP-smart grid mutual safety and cyber security assurance. In: Research Anthology on Smart Grid and Microgrid Development: IGI Global, pp. 1047–1077. (2022) 10. Chowdhury, T., Wassyng, A., Paige, R.F., Lawford, M.: Systematic evaluation of (safety) assurance cases. In: International Conference on Computer Safety, Reliability, and Security, pp. 18–33. Springer (2020) 11. Neši´c, D., Nyberg, M., Gallina, B.: A probabilistic model of belief in safety cases. 138, 105187 (2021) 12. Kelly, T., Weaver, R.: The goal structuring notation–a safety argument notation. In: Proceedings of the dependable systems and networks 2004 workshop on assurance cases, p. 6. Citeseer (2004) 13. Bloomfield, R., Bishop, P.: Safety and assurance cases: past, present and possible future–an Adelard perspective. In: Making systems safer, pp. 51–67. Springer (2010)

Do We Use the Right Elements for Assurance Case Development?

285

14. SACM.: Object Management Group: Structured Assurance Case Metamodel (SACM), vol. 2.1. (2018) 15. Wei, R., Kelly, T.P., Dai, X., Zhao, S., Hawkins, R.: Model based system assurance using the structured assurance case metamodel. 154, 211–233 (2019) 16. Cole, N.: Arguing Assurance in Trusted Execution Environments using Goal Structuring Notation. (2021) 17. Mumtaz, M., Anwar, S., Mumtaz, N., Kausar, T., Pakistan, A.S.: Engineering safety case arguments using gsn standards. 1(1), 124–131 (2019) 18. Piriou, P.-Y., Boudeville, O., Deleuze, G., Tucci-Piergiovanni, S., Gürcan, Ö.: Justifying the dependability and security of business-critical blockchain-based applications. In: In: 2021 Third International Conference on Blockchain Computing and Applications (BCCA), pp. 97– 104. IEEE (2021) 19. Brundage, M., et al.: Toward trustworthy AI development: mechanisms for supporting verifiable claims. (2020) 20. Javed, M.A., Muram, F.U., Punnekkat, S., Hansson, H.: Safe and secure platooning of automated guided vehicles in Industry 4.0. J. Syst. Architect. 121, 102309 (2021) 21. Maksimov, M., Kokaly, S., Chechik, M.: A survey of tool-supported assurance case assessment techniques. ACM Comp. Surv. 52(5), 1–34 (2019) 22. Othmane, L.B., Ali, A.: Towards effective security assurance for incremental software development the case of zen cart application. In: 2016 11th International Conference on Availability, Reliability and Security (ARES), pp. 564–571. IEEE (2016) 23. Kelly, T.P.: Arguing safety-a systematic approach to safety case management. DPhil Thesis York University, Department of Computer Science Report YCST (1999) 24. Rich, K., Blanchard, H., McCloskey, J.: The use of goal structuring notation as a method for ensuring that human factors is represented in a safety case. In: 2007 2nd Institution of Engineering and Technology International Conference on System Safety, pp. 217–222. IET (2007) 25. Matsuno, Y., Taguchi, K.: Parameterised argument structure for GSN patterns. In: 2011 11th International Conference on Quality Software, pp. 96–101. IEEE (2011) 26. Aas, A.L., Andersen, H.S., Skramstad, T.: A retrospective safety case for an advanced driller’s cabin using the goal structuring notation (GSN). In: International Petroleum Technology Conference. OnePetro (2009) 27. Austin, R.A., Mahadevan, N., Sierawski, B.D., Karsai, G., Witulski, A.F., Evans, J.: A CubeSat-payload radiation-reliability assurance case using goal structuring notation. In: 2017 Annual Reliability and Maintainability Symposium (RAMS), pp. 1–8. IEEE (2017) 28. Schmid, T., Schraufstetter, S., Wagner, S., Hellhake, D.: A safety argumentation for failoperational automotive systems in compliance with iso 26262. In: 2019 4th International Conference on System Reliability and Safety (ICSRS), pp. 484–493. IEEE (2019) 29. Yamamoto, S.: Assuring security through attribute GSN. In: 2015 5th International Conference on IT Convergence and Security (ICITCS), pp. 1–5. IEEE (2015) 30. Yamamoto, S., Morisaki, S.: IT demand governance using business goal structuring notation. In: 2016 6th International Conference on IT Convergence and Security (ICITCS), pp. 1–5. IEEE (2016) 31. Almomani, M.A., Basri, S., Gilal, A.R.: Empirical study of software process improvement in Malaysian small and medium enterprises: the human aspects. J. Softw. Evol. Proc. 30(10), e1953 (2018) 32. Basri, S., Almomani, M.A., Imam, A.A., Thangiah, M., Gilal, A.R., Balogun, A.O.: The organisational factors of software process improvement in small software industry: comparative study. In: International Conference of Reliable Information and Communication Technology, pp. 1132–1143. Springer, Cham (2019)

286

A. R. Gilal et al.

33. Jameel, S.M., Gilal, A.R., Rizvi, S.S.H., Rehman, M., Hashmani, M.A.: Practical implications and challenges of multispectral image analysis. In: 2020 3rd International Conference on Computing, Mathematics and Engineering Technologies (iCoMET), pp 1–5. IEEE (2020) 34. Gilal, A.R., Jaafar, J., Abro, A., Omar, M., Basri, S., Saleem, M.Q.: Effective personality preferences of software programmer: a systematic review. J. Inf. Sci. Eng. 33(6), 1399–1416 (2017) 35. Katper, S.H., Gilal, A.R., Alshanqiti, A., Waqas, A., Alsughayyir, A., Jaafar, J.: Deep neural networks combined with STN for multi-oriented text detection and recognition. Int. J. Adv. Comp. Sci. Appl. 11(4) (2020) 36. Gilal, A.R., Omar, M., Mohd Sharif, K.I.: Discovering personality types and diversity based on software team roles. Int. Conf. Comp. Inform. (ICOCI) 2013, 259–264 (2013) 37. Gilal, A.R., Jaafar, J., Omar, M., Basri, S., Waqas, A.: A rule-based model for software development team composition: team leader role with personality types and gender classification. Inf. Softw. Technol. 74, 105–113 (2016) 38. Amin, A., Basri, S., Rehman, M., Capretz, L.F., Akbar, R., Gilal, A.R., Shabbir, M.F.: The impact of personality traits and knowledge collection behavior on programmer creativity. Inform. Softw. Technol. 128, 106405 (2020) 39. Gilal, A.R., Jaafar, J., Capretz, L.F., Omar, M., Basri, S., Aziz, I.A.: Finding an effective classification technique to develop a software team composition model. J. Softw. Evol. Proc. 30(1), e1920 (2018) 40. Gilal, A.R., Jaafar, J., Omar, M., Basri, S., Aziz, I.A.: Balancing the personality of programmer: Software development team composition. Malays. J. Comput. Sci. 29(2), 145–155 (2016)

Development of a Mobile Application for Scheduling Electric Vehicle Charging in Wind Energy Powered Facility Misbah Abdelrahim1 , Ammar Ahmed Alkahtani1(B) , Gamal Alkawsi1 , Yahia Baashar2 , and Sieh Kiong Tiong1 1 Institute of Sustainable Energy, Univesrsiti Tenaga Nasional, 43000 Selangor, Malaysia

[email protected] 2 Faculty of Computing and Informatics, Universiti Malaysia Sabah (UMS), Labuan, Malaysia

Abstract. The electric vehicle revolution is expanding and becoming more popular worldwide because it is one of the most effective alternatives for reducing pollution and fuel consumption, as well as it is also one of the most effective alternatives for reducing pollution and fuel consumption and saving money. Wind energy is one of the green energy sources that can be used for powering charging stations. Generally the wind energy can be used in a hybrid mode with another renewable source. This paper aims to look at the possibility of using wind energy as a standalone energy source to support electric vehicle charging stations by developing a mobile application to schedule charging times at lower prices. As the number of electric vehicles grows, the number of charging stations also increases. This sudden power demand of the fast-charging station impacts grid stability. The mobile application has been developed to let customers charge their electric vehicles using green energy at a lesser cost by informing them of periods of high wind speed—these aids in utilizing the benefits of wind speed to charge electric vehicles using DC fast charging. Keywords: Charging station · Electric vehicle · Mobile app · Wind energy

1 Introduction Nowadays, the electric vehicles (EV) revolution is spreading up and becoming more popular around the world, owing to the fact that they are one of the greatest solutions for reducing pollution, as well as reducing fuel consumption which will save more money. Electric vehicles (EVs) are one of the most promising technologies for lowering emissions in global transportation, but the advantages they provide are contingent on the source of the electricity they use. Currently, renewable energy is used to charge too few electric vehicles. On a daily basis, several types of renewable energy equipment are observed entering the worldwide market as global awareness of green energy grows. However, a few improvements stand out above the others. The transition to renewable energy is a significant step toward green energy. This innovative equipment is the prospective weapon for © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 287–297, 2023. https://doi.org/10.1007/978-3-031-20429-6_28

288

M. Abdelrahim et al.

fighting the huge energy crisis the world is facing today. At the same time, the Increased demand for Electric Vehicles has resulted from rising fuel prices and pollution (EV). "According to the International Energy Agency, worldwide demand for electric cars (EVs) will reach 145 million by 2030 [1]. EV owners may charge their vehicles at home, but the process takes a long time. According to that, this project is required to establish fast-charging stations in order to support the growth of electric vehicles. The EV battery may be charged in roughly 15 min. On the other hand, fast charging has the drawback of high-power consumption and its influence on the grid. Based on that, Renewable energy sources and storage technologies can be implemented in these stations to solve this problem [2]. The grid’s stability is harmed by traditional charging stations, which have difficulties including harmonics, fluctuations, power outages, and a high charging cost [3]. In response, researchers have recommended several alternate solutions, including creating charging stations powered by renewable energy sources like wind and solar panels. After the announcement of the rapid growth of the EV at the turn of the millennium, renewable energy-based charging infrastructure (RCI) research began with the effort of wind and solar for EV charging infrastructure. To solve the shortcomings in the current charging infrastructure, it envisioned a charging station that could match EV demand with renewables and direct current (DC). According to prior research, adding energy storage into the electrical grid was the simplest approach to eliminate unpredictability, intermittency, and power fluctuations, and hence ensuring a consistent and continuous supply of renewable energy [4]. An energy hub (EH) that integrates several types of energy sources (e.g., storage and wind energy) need optimum management to decrease wind uncertainty. In this regard, Najafi et al. introduced an information gap decision theory (IGDT) that takes into account demand response (DR) and an energy storage system (ESS) in order to reduce uncertainty and increase wind turbine production [5]. The requirement to install battery banks for energy storage is seen as the most significant disadvantage of standalone charging stations. The high installation and maintenance costs, the relatively short lifespan when often charged and discharged, and the high cost of the accompanying energy conversion power electronics [6, 7] all contribute to this disadvantage. This project proposes an EV charging station utilizing renewable energy as a business model. The proposed EV charging station purchases power from wind energy systems at a low price and uses that power to charge a fixed battery. The electricity is then sold and utilized to charge electric vehicles during periods of high wind speed. At the same time, the charging infrastructure is expected to increase with the increasing number of EVs, resulting in colossal electricity demand [8]. On the other hand, in this project, a mobile application will be designed to assist EV users in booking a slot to charge their electric vehicles. This mobile application will notify the EV users when the wind speed is high that they can charge their EV at a lower price using a renewable energy source. In addition, it shares with the user the location of the wind charging station.

Development of a Mobile Application for Scheduling Electric Vehicle

289

The remaining of the paper contains four sections, Sect. 2 illustrates the literature review and related works, Sect. 3 explains applied research methodology, Sect. 4 discusses the results, followed by the conclusion in Sect. 5.

2 Literature Review and Related Works Many developed and developing countries seek renewable energy resources to substitute traditional methods. Nowadays, large-scale wind turbines can be seen all over the world. There are a lot of honourable wind turbine farm examples, such as Gansu wind farm in China, AWEC in US, Muppandal in India, and London array in UK. In 2019, wind contributed 7.2% of the nation’s electrical supply, making it a major source of inexpensive, renewable energy. Moreover, wind power was the second-largest source of new electric-generating capacity increases in the United States in 2019, after natural gas, with 9,137 megawatts (MW) of new capacity added and $13 billion spent [9]. According to the research conducted by [10], policy transformation and practice shifting can be a complex transformation process in a liberalized market. Not only there will be some contradiction and tension present in the practice settings, but also these contradictions shall be taught about, encouraged, generalized, and normalized to be capable of establishing a standard urban sustainable charging system amongst the public due to the difficulty of implementation. In another study [11], the researchers also referred to another obstacle: securing access to incentives like carbon credit has not yet been applied in many countries. In addition to that, sociotechnical obstacles are off the table. Consumer attitude and reaction towards this new approach might show resistance. That means EVs might take off in the public sector more than in the private sector. From another perspective, the researchers in [12], went in-depth on designing renewable energy-based charging stations while minimizing the drawbacks. The paper began by discussing the charging algorithm’s importance and explained that it is a crucial aspect of the process because it impacts the charging capacity and battery life. This means the station will use a two-step constant current charging approach. The EV gets charged with 100 Amperes in the first step when its state of charge (SOC) is less than 70%. When this level is reached, the second step begins, in which the charging current is adjusted to 37.5 Amperes, resulting in slower but safer charging. What was noticed by following this process, it will increase the life span of the battery. Furthermore, the system can undertake a charging operation by directly exploiting available renewable energy sources (RES), employing Maximum Power Point Tracking (MPPT) algorithms to collect the abundant and sufficient available power from solar arrays and wind turbines. The existing applications for electric vehicle use have been reviewed; for example, ABRP is a multipurpose application that caters to all of a user’s electric vehicle demands. It’s simple to use; the user selects the EV model and enters the destination and the current battery level. This program will then show the user the quickest route to the location and the expected battery usage when the user arrives. While if a charge is needed, it will locate charging stations along the user trip, reducing the number of unnecessary stops and detours. The ABRP application does not provide charging rates nor a way to pay for charging.

290

M. Abdelrahim et al.

Another application is the CHARGEV, Malaysia’s nation’s largest charging network, with over 300 charging stations throughout the country. This mobile application is quite simple; users should search for a location and show users all the regional charging stations. If a user needs instructions to one of the charging stations, the user should hit the hyperlink to launch Google Maps. Lot availability and opening hours are among the other items presented on the app. Similarly, PLUGSHARE is an application that will significantly benefit users in Malaysia and Singapore. The EV charging is similar to Google Maps. When the user enters their destination, it will display all the charging stations in their immediate vicinity. On the other hand, when a user selects a charging station, the number of charging plugs available, the types of chargers accessible, charging fees, membership requirements, and even local facilities will be displayed. Another essential feature is the option to check in to each location, allowing the user to bookmark the place for later use. Users can also share images of charging ports to determine whether the area is well-lit or populated. Lastly, PARKEASY is an application allowing users to reserve parking spots with electric vehicle charging. To utilize, the user must reserve a spot and then navigate to the location within an hour. When the user arrives, they should notice a remote barrier that prohibits others from grabbing the user slot. The user should utilize the programme to unlock the barrier and park the user’s electric vehicle. ParkEasy employs a credit system to pay for parking and fees, which users can examine after using the facilities.

3 Method and Materials The study aims to make the transition to the new electric mobility situation easier for EV customers by improving advanced service support, reducing the feared problem of range anxiety and optimizing EV charging station utilization by establishing a proper reservation system. The study explores an integrated model that includes EV drivers and charging stations to achieve this goal. The proposed application will provide complete assistance to EV drivers through functionalities of wind speed monitoring and EV wind charging station discovery along the way. It also supports the IoE semantics architecture, allowing EV drivers to reserve a charging slot based on their preferences and current EV supply station availability. The research method is the sequence of steps or stages used as workflows in a study [13]. The flow and steps in this study can be seen in Fig. 1. The first step is to select a subject, followed by a literature review to obtain enough information on previous research related to this project. Moreover, component selection has been reviewed in order to achieve the most efficient project design possible. The next step is to do an experiment to test, analyze, and fix the system of the wind charging station and the mobile application if any defects or bad results are uncovered.

Development of a Mobile Application for Scheduling Electric Vehicle

291

Fig. 1. Research workflow diagram

The project prototype will be built, and all data will be collected in the final phase. The positive and negative results of the flow chart are based on prototypes; if they aren’t, the good outcomes will indicate that the system is working. Otherwise, the experiment must be repeated to confirm the system’s failures.

4 Results and Discussion 4.1 Application Architecture This proposed mobile application begins by requesting EV users to sign up to make an account, as shown in the mobile application flow chart in Fig. 2. The user will be able to login in the following stage. The user will then be asked to choose from one of four options: accurate wind speed, expected wind data, make a reservation or find our location. If the user selects accurate wind data, the charging station will display the

292

M. Abdelrahim et al.

accurate wind speed (real-time) and then provide the user with the choice to book at a lower price if the wind speed is high; or the user will be asked to charge using available power at the rated price.

Fig. 2. Flowchart of the proposed App

Development of a Mobile Application for Scheduling Electric Vehicle

293

The second option is expected wind data, which displays the expected wind speed for the coming 48 h; if the user selects this button, the application will check if the wind speed is high for the next 48 h and shows available slots at a low price. However, if the wind speed remains low, the EV drivers still have to pay the standard charges. More details about the functions are discussed below: 4.1.1 Accurate Wind Speed The mobile application linked to the WEATHER&RADAR website informs the user of the local weather status. This mobile application will display the current wind status at the charging station (see Fig. 3). This will allow the user to assess whether the wind speed is sufficient to charge an electric vehicle using wind energy at a lesser cost. Otherwise, if the wind speed is less than four m/s, the user will be given a choice to check another time to charge the EV or use grid power at a higher cost.

Fig. 3. The accurate wind speed function interfaces

4.1.2 Expected Wind Data Wind speed has a random pattern and a few other intermittent properties. This application presents a novel technique for providing users with wind speed forecasts. The primary distinctive aspect of this function in this mobile application is that it forecasts future wind speed using both temporal and geographical variables and WINDFINDER website [14], to inform the users of the expected wind speed for the next 48 h (see Fig. 4). We

294

M. Abdelrahim et al.

have tested the proposed strategy using historical data on wind speed in the charging station region. Furthermore, EV customers can schedule charging times depending on wind forecasts on this screen. If the wind data is high, the user can schedule a charging appointment to assess the EV using a green source. At the same time, the proposed mobile application would display the wind data in a different colour depending on the wind speed status and the charging station availability.

Fig. 4. The expected wind data for the next 48 h

4.1.3 Scheduling Charging Sessions The ability to make a charging reservation is the essential feature of the proposed application. When the user chooses this option, they enter a settings screen where they may select the date, timing, and quantity of energy required (by kWh) the user inserts. After that, a request is made to the charging station service, which analyses the user’s request and searches for possible reservations in the supplied spatiotemporal frame based on the charging capacity in the charging station.

Development of a Mobile Application for Scheduling Electric Vehicle

295

The user can select one of the available options based on the wind availability or initiate a new request. As mentioned before, a reservation is made in the first situation by entering the date, time slot, and the need for energy (by kWh), as shown in Fig. 5. However, if two or more customers are trying to book at the same time, the service will confirm the first reservation and reject the second, notifying the user that the charging station is booked. When a reservation is finished, it is added to a reservation list, which keeps track of the user’s previous actions and shows that the booking has been confirmed on the mobile application. On the other side, an email will be sent to the user to confirm the booking after completing this process. This behaviour allows for the display of the greatest number of possible choices at all times, effectively managing the scenario of several users accessing the same resource simultaneously.

Fig. 5. The reservation screens

4.1.4 Location of the Charging Station This mobile application was created for a proposed wind charging station. The last function on of this mobile application allows the user to share the location of the wind charging station with others via Google Maps (see Fig. 6). This will make it very simple for the user to go to the wind charging station. However, if there are many charging stations, the mobile application will inform the user of the nearest location of the wind charging station.

296

M. Abdelrahim et al.

Fig. 6. The interface of sharing the location of the charging station

5 Conclusion Electric vehicle demand is increasing daily worldwide. The growing use of electric vehicles has started to integrate pressure on the electrical energy supply of the grid. Yet, there are insufficient renewable energy charging stations compared to traditional charging stations, necessitating the creation of a national road map for renewable energy charging station size, regulation, and other factors to help to lower the cost of charging and the percentage of pollution in the air. Designing a mobile application for this wind charging station was critical to reduce the wasted energy from the wind turbine by giving users the wind data for the high wind speed periods. While the implementation of the wind charging station was designed to feed the EV’s batteries directly, reducing the number of storage batteries will lower the energy costs. The mobile application has been developed particularly for the wind charging station to display the wind speed status to the user, informing the user of the most cost-effective time to charge the EV battery using wind energy. After prompting the user for the date/time, it will notify the user whether or not the charging station is full. However, the reserved charging station time is not included in the reservation screen of the existing application. To increase the service quality of the mobile application, the charging station’s vacancy capacity should be given in the wind charging station application.

Development of a Mobile Application for Scheduling Electric Vehicle

297

Acknowledgements. This work is a part of a project submitted to Universiti Tenaga Nasional (UNITEN), and the authors would like to acknowledge the Bold Research Grant provided by UNITEN, Project No. (J510050002/2021094).

References 1. Frangoul, A.: Global electric vehicle numbers set to hit 145 million by end of the decade, IEA says. CNBC. Retrieved from https://www.cnbc.com/2021/04/29/global-electric-vehicle-num bers-set-to-hit-145-million-by-2030-iea-.html, 29 April 2021 2. Domínguez-Navarro, J.A., Dufo-López, R., Yusta-Loyo, J.M., Artal-Sevil, J.S., BernalAgustín, J.L.: Design of an electric vehicle fast-charging station with integration of renewable energy and storage systems. Science direct. https://www-sciencedirect-com.ezproxy.uniten. edu.my/science/article/pii/S0142061518308019 3. Sanchari Deb, S.D., Kari Tammi, K.T., Karuna Kalita, K.K., Pinakeshwar Mahanta, P.M.: Impact of electric vehicle charging station load on distribution network. MDPI. https://www. mdpi.com/1996-1073/11/1/178 4. Sánchez-Sáinz, H., García-Vázquez, C.A., Llorens Iborra, F., Fernández-Ramírez, L.M.: Methodology for the optimal design of a hybrid charging station of electric and fuel cell vehicles supplied by renewable energies and an energy storage system. Sustainability 11, 5743 (2019) 5. Najafi, A., et al.: Uncertainty-based models for optimal management of energy hubs considering demand response. Energies 12, 1413 (2019) 6. Teleke, S., Baran, M.E., Huang, A.Q., Bhattacharya, S., Anderson, L.: Control strategies for battery energy storage for wind farm dispatching. IEEE Trans. Energy Convers. 24, 725–732 (2009) 7. Zhao, X., Yan, Z., Xue, Y., Zhang, X.P.: Wind power smoothing by controlling the inertial energy of turbines with optimized energy yield. IEEE Access 5, 23374–23382 (2017) 8. Noman, F., Alkahtani, A.A., Agelidis, V., Tiong, K.S., Alkawsi, G., Ekanayake, J.: WindEnergy-powered electric vehicle charging stations: resource availability data analysis. Appl. Sci. 10(16), 5654 (2020). https://doi.org/10.3390/app10165654 9. Utility-Scale Wind Energy.: WINDEchange.energy.gov [Online]. https://windexchange.ene rgy.gov/markets/utility-scale 10. Manfren, M., Caputo, P., Costa, G.: Paradigm shift in urban energy systems through distributed generation: methods and models. Science direct. www.sciencedirect.com 11. Andersen, P.H., Mathews, J.A., Rask, M.: Integrating private transport into renewable energy policy: The strategy of creating intelligent recharging grids for electric vehicles. Science direct. www.sciencedirect.com 12. Gucın, T.N., Ince, K., Karaosmano˘glu, F.: Design and power management of a grid-connected Dc charging station for electric vehicles using solar and wind power. In: 2015 3rd International Istanbul Smart Grid Congress and Fair (ICSG). (2015) 13. Indrawan, I.W.A., Bayupati, I.P.A., Singgih Putri, D.P.: Markerless augmented reality utilizing gyroscope to demonstrate the position of Dewata Nawa Sanga. Int. J. Interact. Mobile Technol. (iJIM) 12(1), 19–35 (2018) 14. Wind, waves & weather forecast London Heathrow Airport.: Windfinder.com. https://www. windfinder.com/forecast/london-heathrow

Evaluating Websites Audit Tools: A Case Study of the Amazon Website Mohammed Fahad Alghenaim(B) , Nur Azaliah Abu Bakar, and Fiza Abdul Rahim Advanced Informatics Department Razak Faculty of Technology and Informatics, Universiti Teknologi Malaysia, 54100 Kuala Lumpur, Malaysia [email protected]

Abstract. Modern businesses and organizations cannot operate without an official website. A well-designed website should be informative, visually appealing, and easily navigable to encourage return visits. The website gives many users the first impression and insight into the company’s nature, organizational structure and culture, product or service offerings, and past successes. Usability testing, therefore, enables firms to evaluate how easy their websites are to understand and use. It evaluates user-friendliness, user satisfaction, efficiency, and error detection. A website’s usability is also affected by its suitability for people of all ages and genders. Usability, or how easy it is to use, is essential to a website’s success, especially as websites become more interactive, complex, and full of features. Various testing tools are available offline/online and legacy/general-purpose. This paper focuses on three online tools - Qualidator, SEOptimer, and Website Grader – used to evaluate the usability or ease of use of the Amazon website. Keywords: Website audit · Online tools · Website efficiency · Review

1 Introduction Websites are an indispensable marketing tool for businesses and individuals, helping organizations and private entities increase their credibility and chances of generating leads and showcasing their brand to prospective clients [1]. As such, a successful website must have an attractive design and offer a pleasant user experience that enhances the website’s credibility and increases the conversion rate [2]. People visit websites to meet specific needs such as shopping for products, booking flights, making hotel reservations, or looking for entertaining content. Websites that fail to satisfy these needs lose out on potential customers. In particular, a non-responsive website design is a major turn-off for many website visitors [3]. Other reasons that prompt visitors to leave a website include slow loading, poor navigation, outdated design, poor content structure, and intrusive use of video and audio [4]. In addition, there are at least 1.8 billion websites, making it difficult for a website to stand out [5]. Website owners must therefore conduct usability tests to assess the efficiency and user-friendliness of their sites and detect errors that might have been overlooked. Usability testing evaluates the functionality of a website or an app, usually by observing users as they navigate it [6]. Usability testing of a © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 298–307, 2023. https://doi.org/10.1007/978-3-031-20429-6_29

Evaluating Websites Audit Tools: A Case Study of the Amazon Website

299

website includes web testing to improve usability and navigation and surveys to get users involved. There is a need to evaluate more website evaluation tools to assess their efficiencies and compare the differences. The study aims to uncover problem areas and identify new opportunities for improving user experience. Several online tools, such as focus groups, heat maps, and surveys, can evaluate a website. However, this paper focuses on three online tools—Qualidator, SEOptimer, and Website Grader—that will be used for assessing the Amazon website.

2 Literature Review Web usability describes the ease with which a website can be used. It involves assessing efficiency, memorability, learnability, ease of use, error prevention, and user satisfaction [7]. The goal of conducting such tests is to develop more user-friendly systems or websites and improve user satisfaction. One advantage of usability tests is their relative affordability and ease of use. However, specially designed lab tests where facilitators observe users are costly, arising from expenditures such as hiring or purchasing a lab and recruiting the facilitators and testers. In contrast, remote usability testing is inexpensive since the evaluation occurs in the user/tester’s typical environment [8]. Various studies have examined the application of usability testing in multiple contexts and its benefits. For example, Abascal et al. [9] examined the guidelines for automatically verifying website accessibility. In their designs, the risk-informed study of website designers perpetuates social exclusion, especially among people with disabilities. The Evallris web service assessed web accessibility by breaking down the HTML mark-up of a web page into components (e.g., the header, table, or image) and evaluating them separately. From the findings, Evallris could detect most errors and provide keyboard shortcuts to vital form controls and links. Similarly, Genuis [10] examined the applicability of usability testing to library environments owing to the need for clients to easily access library resources, converting occasional visitors into loyal customers. Certain conclusions merged from the study. First, the test should be done continuously for more consistent website performance. Second, usability testing can be used for essential website evaluation and for assessing specific tasks, levels of access for different groups of users, and unique information needs of users, especially in specialized libraries. Meanwhile, Sari et al. [7] undertook usability testing on the learning proves on a laboratory website. In light of the persistent difficulty in locating needed information, the study’s goal was to improve the ease of use. A sample of 33 students and lab assistants participated in focus group discussions. The findings proposed the addition of recent updates, a menu for fast download, and a working link to the content domain. The study suggested using bigger readable font, good color, and images/icons for the visual display. Some studies have also focused on the suitability of usability testing for comparative analysis. For instance, in [11], the authors proposed a two-step method for evaluating web-based systems. The first step involved usability testing on users, and the data obtained formed part of the second evaluation. The authenticity and precision of the final results were further compared with user opinions obtained through a questionnaire. The method identified mistakes when performing registration, flight search, and ticket

300

M. F. Alghenaim et al.

purchasing tasks. Alyusuf et al. [12] evaluated the quality of undergraduate education websites in pathology using a new evaluation tool. The findings showed high-reliability scores for internal consistency, intra-class correlation, and content-related validity. Of the 278 websites, 29.9% were recommended for use, and 41.0% were recommended with caution. 29.1% were unfitting for use. Usability testing methods differ in their effectiveness. For instance, Baravalle and Lanfranchi [13] used OpenWebSurvey—a new remote web usability testing method—to record users’ behavior while surfing online. The method was effective and easy to use since it did not need program installation or configuration, and it can therefore be used on large numbers of users. Similarly, the authors in [14] reviewed existing literature on user testing to clarify test procedures and define tools for conducting such tests. The benefits of usability tools identified in the study included determining the duration and frequency of user behavior to help identify performance problems or difficulties. Lastly, usability testing can be an effective mechanism for identifying problems within websites. For instance, Hinchliffe and Kerry Mummery [15] conducted usability testing on a health promotion website using direct observation, performance measures, and subjective user preferences. From the findings, usability testing uncovered 68 unique problems related to design, format, navigation, learnability, terminology, instructions, and feedback. After modification, the time taken to complete tasks improved from 21.59 to 10.18 min. In contrast, Hasan [16] performed usability testing on university websites to identify the common usability problems. The study used observation and post-test questionnaires, with the former more capable of identifying minor and major issues. Twelve significant problems (e.g., non-obvious links, misleading links, inconsistent language, irrelevant content, and ineffective internal search) and7 minor problems (e.g., weak navigation support, misleading images, inappropriate font, and missing functions) were identified. Osterbauer et al. [17] conducted usability tests on various sites belonging to Austria’s insurance companies, banks, and daily newspapers. The study used checklist-based and scenario-based tests in the areas of navigation and graphics. From the findings, low usability is caused by solid usage of domain-specific terms that are incomprehensible to average users, too much animation-based advertising that distracts users, excessive amounts of text on a web page, and too bright colors.

3 Methodological Approach Website usability testing is essential for determining a site’s performance and overall usability. This study used the website “https://www.aboutamazon.com/,” which is the site for Amazon. Three website evaluation tools are used in this regard: (1) SEOptimer, (2) Website Grader, and (3) Qualidator. Kaur et al. (2016) used similar tools to evaluate the performance and overall usability of Punjab Educational Universities. Evaluation of the Amazon website “https://www.aboutamazon.com/” The three automated website assessment tools are discussed.

Evaluating Websites Audit Tools: A Case Study of the Amazon Website

301

3.1 Website Grader Tool (WGT) This tool is mainly used to evaluate a website of interest’s authenticity (or lack thereof) [18]. According to Khandare et al. [19], WGT offers straightforward answers about a site’s stats essential to its users. The following steps were undertaken in using the WGT. Step One: Visit the WGT website at “https://website.grader.com/” (See Fig. 1).

Fig. 1. The WGT official home page

Step Two: Enter the Amazon website name “https://www.aboutamazon.com/,” which needs to be evaluated. A personal e-mail address is also required, and the respective e-mail address fields are provided. Step Three: Run the test by clicking on the “Get your score” tab. After running the test, the WGT evaluator tools display various metrics showing the website’s usability (Fig. 2). For “https://www.aboutamazon.com/,” the overall website score is 74, and the verdict is that the website is OK and is not too shabby [20]. Step Four: As shown in Fig. 2, the WGT optimizes the website for various matrixes: (1) performance, (2) mobile, (3) SEO, and (4) security—and displays the outcome of the evaluation. For instance, the performance of the “www.aboutamazon.com” website is 9/30, SEO is 25/30, mobile optimization score is 30/30, and security is 10/10. Optimizing website performance is essential not only for (1) increasing traffic and (2) improving conversion rates but also for (3) generating more leads and (4) increasing revenue. The WGT tool presents website performance in 8 variables, ranging from page size, requests, and speed to browser caching, image size, and minimal page redirects. It also provides result performance regarding minified JavaScript and minified CSS. Figure 3 presents these scores. Step Five: The GWT website evaluator tool also specifies whether (or not) the website is supported by mobile devices alongside its SEO results. For “www.aboutamazon.

302

M. F. Alghenaim et al.

Fig. 2. Amazon WGT scorecard

com,” the mobile optimization score is 30/30, and the SEO score is 25/30. Specifically, it shows that the ’ ’website’s mobile usage in legibility (e.g., legible font size and easiness to the eye), tap targets (i.e., easy clicking), and responsiveness is fantastic. The site also performed superbly in security—with a score of 10/10. A secure site equipped with a Secure Sockets Layer (SSL)—a secure protocol that establishes an encrypted connection or internet linkage between a given server and a browser—is now the norm online [21]. For “www.aboutamazon.com,” the website has secure HTTPS and JavaScript Libraries. Step Six: Fundamentally, the GWT evaluator also gives recommendations for optimizing the website. For “www.aboutamazon.com”, the following needs to be done (See Fig. 4). 3.2 SEOptimer Tool Like the WGT tool, this SEOptimer is a freely available online website evaluator tool. It is generally an SEO Audit Tool used to perform a thorough SEO analysis across multiple website data points and generate clear and workable recommendations that can be undertaken to improve a ’ ’site’s online presence and subsequently rank better and perform favorably in Search Engine Results [22]. It provides a range of free tools that can be used to improve website performance and usability. According to [19], the tool

Evaluating Websites Audit Tools: A Case Study of the Amazon Website

303

Fig. 3. The GWT Performance Score of the website “www.aboutamazon.com”

calculates several website specifications like Performance and social and Page Analysis. The following are the practical steps for using the SEOptimer tool. Step One: Visit the SEOptimer website evaluation tool at “www.seoptimer.com”. Step Two: Enter the website name “www.aboutamazon.com”, which is to be analyzed. Step Three: Click the audit button to run the test. After carrying out the test, the SEOptimer tool displays the site’s overall usability audit result of the website “www.abo utamazon.com” in the form of grades or levels [19]. The overall audit results for “www. aboutamazon.com” is B- (See Fig. 5).

304

M. F. Alghenaim et al.

Fig. 4. The evaluation reports

Fig. 5. Audit Results for “www.aboutamazon.com”

Step Four: The tool further displays the ’ ’website’s usability audit score. This result is presented in the form of a grade. For “www.aboutamazon.com”, the usability score is C (See Fig. 6).

Evaluating Websites Audit Tools: A Case Study of the Amazon Website

305

Fig. 6. Audit Performance for “www.aboutamazon.com”

Apart from usability, the tool also gives results with respect to (1) On-page SEO, (2) Links, (3) Performance, and (4) Social. This result is presented graphically in Fig. 7 below:

Usability 61 Links

92

0

On-Page SEO78

60

Social

Performance

Fig. 7. The Performance Score of “www.aboutamazon.com”

Along with the grading, the tool also provides practical recommendations for improving the website performance in each of the above aspects. For instance, regarding SEO, it is found that although the website page has a meta description, it should be about 70–320 characters, including the spaces. For “www.aboutamazon.com,” length was 61, below the recommended 70 characters. In addition, the ’ ’site’s primary keywords are not distributed optimally across important HTML tags. Thus, for optimal SEO performance, it is recommended that the ’ ’page’s content be focused on specific keywords that the company would like to be ranked for—ideally, these words should be well-distributed across tags like meta, title, and header tags.

306

M. F. Alghenaim et al.

3.3 Qualidator Tool This is also a website analyzer tool, and it is used to authenticate a ’ ’website’s overall quality. According to [19, 23], the tool evaluates the website’s pages and assesses it concerning such variables as SEO, Accessibility, and the ’ ’site’s Overall Grade. The evaluation outcome is given percentages. This is useful in helping web designers develop user-friendly and optimally functioning websites. Like the other two web validator tools, the Qualidator tool’s procedure is as follows. Step One: Visit the website “www.qualidator.com.” Step Two: Enter the website name “www.aboutamazon.com” to be evaluated and click the test button. Step Three: After running the test, the tool displays the overall result in terms of the website’s (1) Accessibility, (2) Usability, and (3) SEO. The outcome is presented in the form of percentages. For “www.aboutamazon.com,” the Qualidator tool analysis outcome could not be created. However, the tool has significant features essential for evaluating a website’s usability, including the provision of meaningful reports and the option of switching the validation tests on or off. Some of the benefits of the qualidator tool include: • Easy identification of weak spots in website codes • Time savings • Improvement of work efficiency.

4 Conclusion This paper sought to evaluate the various elements needed for optimizing the usability and performance of a website. In the analysis, the focused methodology has been to ascertain and discuss the various paraments essential in improving a website’s performance. The result of this evaluation shows website developers how important it is to use automated tools to test how easy it is to use a website they’ve made. The amazon website is used in this analysis. The evaluation has been done using three automated tools—Qualidator, SEOptimer, and Website Grader—on the Amazon website. Acknowledgment. This work is supported by UTM SPACE with cost center number R.K130000.7756.4J574.

References 1. Fedewa, D.: Why business must heed customer reviews. McKinsey & Company (2022). Retrieved from https://www.mckinsey.com/business-functions/operations/our-insights/whybusiness-must-heed-customer-reviews, 28 May 2022 2. Garett, R., Chiu, J., Zhang, L., Young, S.D.: A literature review: website design and user engagement. Online J. Commun. media Technol. 6(3), 1 (2016) 3. Hussain, A., Mkpojiogu, E.O.C.: The effect of responsive web design on the user experience with laptop and smartphone devices. J. Teknol. 77(4), 41–47 (2015)

Evaluating Websites Audit Tools: A Case Study of the Amazon Website

307

4. Sebastian, N.: Website design stats and trends for small businesses. GoodFirms (2021). Retrieved from https://www.goodfirms.co/resources/web-design-research-small-business, 08 May 2022 5. Fitzgerald, A.: 25+ web design statistics that are essential to know in 2022. HubSpot Blog (2021). Retrieved from https://blog.hubspot.com/marketing/web-design-stats-for-2020, 21 May 2022 6. Limited, H.: What is usability testing? (and what it isn’t). Hotjar (2022). Retrieved from https://www.hotjar.com/usability-testing/, 21 May 2022 7. Sari, A.D., Suryoputro, M.R., Rochman, Y.A., Ulandari, S., Puspawardhani, E.H.: Usability analysis of laboratory website design to improve learning process. Proc. Manuf. 3, 5504–5511 (2015) 8. Gardner, J.: Remote web site usability testing-Benefits over traditional methods. Int. J. Public Inf. Syst. 3(2) (2007) 9. Abascal, J., Arrue, M., Fajardo, I., Garay, N., Tomás, J.: The use of guidelines to automatically verify Web accessibility. Univers. Access Inf. Soc. 3(1), 71–79 (2004) 10. Genuis, S.: Web site usability testing: a critical tool for libraries. Feliciter 161 (2004) 11. El-Firjani, N., Elberkawi, E., Maatuk, A.: A method for website usability evaluation : a comparative analysis. Int. J. Web Semant. Technol. 8, 1 (2017) 12. Alyusuf, R.H., Prasad, K., Satir, A.M.A., Abalkhail, A.A., Arora, R.K.: Development and validation of a tool to evaluate the quality of medical education websites in pathology. J. Pathol. Inform. 4(1), 29 (2013) 13. Baravalle, A., Lanfranchi, V.: Remote Web usability testing. Behav. Res. Methods Instru. Comput. 35(3), 364–368 (2003) 14. Bastien, J.M.C., Metz, U.P.: Usability testing: some current practices and research questions. Int. J. Med. Inform. (2010) 15. Hinchliffe, A., Mummery, W.K.: Applying usability testing techniques to improve a health promotion website. Heal. Promot. J. Aust. 19(1), 29–35 (2008) 16. Hasan, L.: The usefulness of user testing methods in identifying problems on university websites. JISTEM J. Inf. Syst. Technol. Manag. 11, 229–256 (2014) 17. Osterbauer, C., Köhle, M., Tscheligi, M., Grechenig, T.: Web usability testing—a case study of usability testing of chosen sites (banks, daily newspapers, insurances). (2000) 18. Upqode.: Best website Grader tools best website Grader tools. (2022). Retrieved from https:// upqode.com/best-website-grader-tools/, 28 May 2028 19. Khandare, S.S., Gawade, S., Turkar, V.: Survey on website evaluation tools. In: 2017 international conference on recent innovations in signal processing and embedded systems (RISE), pp. 608–615. (2017) 20. Grader, W.: About Amazon.com. HubSpot (2022). Retrieved from https://website.grader.com/ tests/www.aboutamazon.com, 21 May 2022 21. Lab, K.: What is an SSL certificate—definition and explanation. (2021). Retrieved from https://www.kaspersky.com/resource-center/definitions/what-is-a-ssl-certificate, 21 May 2022 22. SEOptimer.: SEOptimer tools. (2022). Retrieved from https://www.seoptimer.com/free-tools, Accessed 21 May 2022 23. Kwangsawad, A., Jattamart, A., Nusawat, P.: The performance evaluation of a website using automated evaluation tools. In: 2019 4th Technology Innovation Management and Engineering Science International Conference (TIMES-iCON), pp. 1–5. (2019)

Blockchain and Internet of Things (IoT): A Disruptive Integration Nazanin Moosavi1 and Hamed Taherdoost2(B) 1 Hamta Group, Hamta Business Corporation, Vancouver, BC, Canada

[email protected]

2 University Canada West, Vancouver, BC, Canada

[email protected]

Abstract. The combination of internet of things (IoT) and blockchain can propose innovative applications in different smart networks. Due to the various applications in IoT ecosystem providing a secure network with high quality of privacy is so important. Blockchain, which is a decentralized and distributed ledger technology, can be a proper solution. In addition, various terms and conditions of different services in IoT need to be guaranteed, where a smart contract feature in blockchain can be used. In this paper, we studied the integration of blockchain in IoT networks applications. This integration is discussed in different use cases such as smart city, smart energy, smart home, smart industry, healthcare, agriculture, unmanned aerial vehicles (UAVs). The advantages of blockchain application in each of these areas is proposed in the paper. The combination and integration of blockchain and IoT can provide significant development in smart networks. Keywords: Blockchain · Internet of Things (IoT) · Smart Contract · Security · Decentralized

1 Introduction In recent years, there are growing demands for smart things and the explosive proliferation of smart projects, leads the way toward an enormous ecosystem of the Internet of Things (IoT). IoT’s most important features such as scalability, interoperability, integration, and convenience make it an incredible prospect for improvement. In addition, it has a wide range of applications in different fields such as agriculture, smart city, smart energy, etc. [1]. Along with IoT usage in different sectors, security vulnerabilities have increased and there is an urge for secure solutions. Many protocols and algorithms have been used to prevent security risks, but they are not enough. IoT needs the aid of proper protocols and technologies to eliminate or decrease its potential security risks [2]. Recently blockchain technology is presented to be a transformative technology in various markets and industries. Blockchain technology is providing a distributed ledger to store transactions with transparency, immutability, and irreversibility features [3]. The © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 308–315, 2023. https://doi.org/10.1007/978-3-031-20429-6_30

Blockchain and Internet of Things (IoT): A Disruptive Integration

309

most famous blockchain network that is used nowadays is Bitcoin, which was first introduced by Satoshi Nakamoto to provide a trusted peer to peer network for transactions[4]. Bitcoin cryptocurrency eliminates the need for third parties to verify and validate the transaction, hence becoming so popular. However, blockchain can offer solutions to various verticals such as energy, healthcare, agriculture, etc. [5]. Hence, the integration of blockchain and IoT can be regarded as a promising solution to offer decentralization, poor interoperability, privacy, and security vulnerabilities for IoT networks [6]. Besides, blockchain technology can be regarded as a potential solution for efficient storing, processing, data sharing and data market framework which can be incorporated into IoT ecosystem [7, 8]. The main contribution of this article is to deal with preliminary concepts of blockchain technology and its application in IoT technology. At first, we investigate various features of blockchain technology, such as blockchain description, consensus mechanism, smart contract, blockchain features and blockchain classification are described briefly. Blockchain application in IoT can provide security, reliability and distributed network, which are important topics in IoT projects. Then, we propose the application of blockchain in smart cities, smart energy, smart home, smart industry, healthcare, agriculture, and Unmanned Aerial Vehicles (UAVs) with some details. Eventually, the goal of this article is to provide an insight into the technical concepts and research developments in blockchain technology usage in IoT ecosystem. The rest of the paper is organized as follows: Section II outlines blockchain structure and features, Sections III introduces the blockchain solutions for IoT smart projects and finally, section IV concludes the paper.

2 Blockchain Structure and Features 2.1 Blockchain Description Blockchain technology is famous as one of the critical emerging technologies in recent years, which gathers attention from both industry and academia. Blockchain is a distributed database where all transactional records (for tangible or intangible assets) are digitally stored. The blockchain contains a sequence of blocks with an increasing number of records. Each block consists of some sections as follow [9]: • • • • • • •

Block version-4 bytes: The ordinal number of a block Merkle tree hash-32 bytes: Hashing algorithm to create a hash for block data Previous block hash-32 bytes: The hash value of all the data in the previous block Timestamp-4 bytes: The time when the data is created nBits-4 bytes: Current hashing target in a compact form Nonce-4 bytes: A random number that is varied to create a unique hash for each block Data-32 bytes: Transaction data

The following figure shows the pictorial representation of block header and block structure (Fig. 1).

310

N. Moosavi and H. Taherdoost

Fig. 1. Block structure

2.2 Consensus Mechanism The process of adding a new block to the block sequence happens through the consensus mechanism that replaces the mutual trust construct for the centralized third parties. The most important and famous consensus algorithm that is used in the blockchain ecosystem is proof of work (PoW) which is used in Bitcoin cryptocurrency to validate and verify the financial transactions [10]. Since PoW requires a lot of computing power to exploit the consensus mechanism for adding a new block to the block sequence, it is not an environmentally friendly choice. Consequently, there are other types of consensus mechanism that consumes less power such as proof of stack (PoS). According to the different requirements of IoT projects, proper consensus mechanism should be chosen. 2.3 Smart Contract Smart contract is one of the main solutions of blockchain technology that is introduced by N. Szabo in 1997 [11]. Smart contract is a self-executing contract that contains some predefine rules for an agreement between participants and eliminates the need for intermediaries. All the rules are written in code in a blockchain network that only executes when some conditions happened [12]. Here is the table comparing conventional contracts with smart contracts (Table 1). Table 1. Smart contracts versus conventional contracts Requirements

Conventional contracts

Smart contract

Security

Limited

Highly secure

Transparency

Not available

Available 7 *24

Signature

Manual

Digital

Performance time

Days

Minutes

Process

Manual

Automatic

Blockchain and Internet of Things (IoT): A Disruptive Integration

311

2.4 Blockchain Feature Blockchain technology has some unique features that make it interesting and applicable. Here is the description of some most important blockchain features [13]. • Decentralization: This feature eliminates the need for central authorities or third parties to plan for transactions in a network. Information is available to every user, which leads to direct transactions between them. • Privacy: With this feature, real identities of users are secured in the network even when they contact each other. • Immutability: This feature ensures that the information is tamper-proof in the network. Hence, after the transaction happens it cannot be edited or deleted. • Distributed ledger: Each user has one copy of the database, so the database is distributed. • Irreversibility: When one transaction is stored, it cannot be recaptured. 2.5 Blockchain Classification Blockchain network is classified into two models, permissionless and permissioned. In a permissionless blockchain, any user in the network can add a new block to the block sequence. Public blockchains such as Bitcoin are in this model. In a permissioned blockchain, only some authorized users can add a block to the blockchain. Private and consortiums are in this model. Table 2 provides a comparison between public, private, and consortium blockchain [5, 14]. Table 2. Comparison between different blockchain classifications Category of blockchain

Participant

Consensus protocols

Scalability

Public

Anonymous

PoW/PoS

High

Private

Identified

PBFT

Low

Consortium

Identified

PBFT

Low

3 Blockchain Effect on IoT The integration of blockchain and IoT has gained a large attention from both academia circles and industry. Blockchain can improve IoT issues and challenges from different aspects as follow [5]. • Security: Since security is so important in many IoT projects such as monitoring, blockchain can help to increase security by using the consensus mechanism. • Reliability: If some nodes in the IoT network are attacked, the data is still safe and reliable.

312

N. Moosavi and H. Taherdoost

• Distributed: By providing the distributed network, blockchain can make full use of the storage and computing power of distributed IoT devices in the network. Figure 2 shows how blockchain can be used in IoT network.

Fig. 2. IoT blockchain application

4 Blockchain Applications in IoT Blockchain can be applied in many IoT projects to increase the effectiveness of the network. Some of the important applications of blockchain technology in IoT are shown in Fig. 3 and briefly discussed in the following.

Smart City UAV

Smart Energy

Smart Home

Agriculture

Healthcare

Smart Industry

Fig. 3. Application of blockchain in IoT

Smart City In a smart city, information technology is used to provide better and smart services such as managing traffic in streets, smart lightning systems in city parks etc. for dwellers.

Blockchain and Internet of Things (IoT): A Disruptive Integration

313

Although these systems provide a better use of the resources that belongs to all the people in the city, they are more vulnerable to security and privacy risks. Blockchain technology can provide a secure and reliable solution for these problems. Smart Grid The improvement in distributed energy resources (DER) provides a new solution to produce energy anywhere and anytime. This new solution provides a new market, which is not centralized as before and needs to be more secure. Blockchain technology can provide a secure and distributed network for smart grid solutions [15]. Smart Home One of the main important challenges in improving smart home services is the privacy of the users. Since all the information about the home is on the network, it is so important to have a secure network. Blockchain is a good solution. Although the smart contract feature in blockchain can provide direct transactions for some services in a smart home. For example, consider that you are out of milk, then without any intervention, your refrigerator can order milk from the store by smart contract and the blockchain can control the payment [16]. Smart Industry In recent years, industries upgraded from automatic to smart. IoT systems are used in manufacturing to provide data sharing and data analytics for effective decisions. Some Blockchain features such as decentralization, tamper-proof, transparency and smart contract can be so useful to develop and promote news services in industrial IoT [17, 18]. Healthcare The improvements in healthcare personal devices can provide remote health services for people. Data such as heartbeat, diabetes, and pressure investigation are sent to doctors at any time. Providing a secure network where people can trust it and be sure about their privacy is so important [19]. Hence, The usability of blockchain in medical sector can de divided into three sections drug traceability, remote patient-monitoring, and medical record management [20]. Agriculture The use of IoT in smart agriculture can gather the information that is stored in the control system and examined by artificial intelligence (AI). This provides a cost-effective solution for in agriculture supply chain. Distributed ledger technology by blockchain improves the effectiveness, traceability, and simplicity of agricultural supply chains [21]. Uav UAVs play an important role in many smart services from military services to agriculture services. Hence, the information gathered by the UAVs is more vulnerable to jamming attacks. Attackers aim to provide interferences to the UAVs to degrade their performance. Blockchain can provide better security for UAV’s network [22].

314

N. Moosavi and H. Taherdoost

5 Conclusion In this paper, the combination and application of blockchain technology in IoT networks are introduced. At first, the blockchain architecture such as block structures and block headers, comparison between consensus mechanism, description of smart contract, most important blockchain features and blockchain classification are described with details. Then the advantages of using blockchain in IoT network which provides security, reliability and distributed network is investigated. Finally, the blockchain application and most important role in in different IoT verticals such as energy, agriculture, smart city, etc. is proposed. It is concluded that the integration of blockchain technology with IoT helps to provide sustainable and secure IoT network that is mostly needed in IoT ecosystem. It should be noted the integration of blockchain technology into IoT ecosystem is still in its infancy time and it needs to be more studied especially from the security and privacy point of view.

References 1. Da Xu, L., He, W., Li, S.: Internet of things in industries: a survey. IEEE Trans. Industr. Inf. 10(4), 2233–2243 (2014) 2. Lu, Y., Da Xu, L.: Internet of Things (IoT) cybersecurity research: a review of current research topics. IEEE Internet Things J. 6(2), 2103–2115 (2018) 3. Gamage, H., Weerasinghe, H., Dias, N.: A survey on blockchain technology concepts, applications, and issues. SN Comp. Sci. 1(2), 1–15 (2020) 4. Nakamoto, S.: Bitcoin: a peer-to-peer electronic cash system. Decentralized Business Review, p. 21260 (2008) 5. Reyna, A., et al.: On blockchain and its integration with IoT. Challenges and opportunities. Fut. Gen. Comp. Syst. 88, 173–190 (2018) 6. Dai, H.-N., Zheng, Z., Zhang, Y.: Blockchain for internet of things: a survey. IEEE Internet Things J. 6(5), 8076–8094 (2019) 7. Badshah, A., et al.: A novel framework for smart systems using blockchain-enabled internet of things. IT Prof. 24(3), 73–80 (2022) 8. Zhang, J., Zhong, C.: Differential privacy-based double auction for data market in blockchainenhanced internet of things. Wireless Commun. Mobile Comp. (2022) 9. Zheng, Z., et al.: Blockchain challenges and opportunities: a survey. Int. J. Web Grid Serv. 14(4), 352–375 (2018) 10. Xiao, Y., et al.: A survey of distributed consensus protocols for blockchain networks. IEEE Commun. Surv. Tutor. 22(2), 1432–1465 (2020) 11. Szabo, N.: Formalizing and securing relationships on public networks. First monday (1997) 12. Zheng, Z., et al.: An overview on smart contracts: challenges, advances and platforms. Futur. Gener. Comput. Syst. 105, 475–491 (2020) 13. Bhutta, M.N.M., et al.: A survey on blockchain technology: evolution, architecture and security. IEEE Access 9, 61048–61073 (2021) 14. Rajasekaran, A.S., Azees, M., Al-Turjman, F.: A comprehensive survey on blockchain technology. Sustain. Energy Technol. Assess. 52, 102039 (2022) 15. Mollah, M.B., et al.: Blockchain for future smart grid: a comprehensive survey. IEEE Internet Things J. 8(1), 18–43 (2020) 16. Kouzinopoulos, C.S., et al.: Implementing a forms of consent smart contract on an IoT-based blockchain to promote user trust. In: 2018 Innovations in Intelligent Systems and Applications (INISTA). IEEE (2018)

Blockchain and Internet of Things (IoT): A Disruptive Integration

315

17. Mistry, I., et al.: Blockchain for 5G-enabled IoT for industrial automation: a systematic review, solutions, and challenges. Mech. Syst. Signal Process. 135, 106382 (2020) 18. Huo, R., et al.: A comprehensive survey on blockchain in industrial internet of things: Motivations, research progresses, and future challenges. IEEE Commun. Surv. Tutor. (2022) 19. Hasselgren, A., et al.: Blockchain in healthcare and health sciences—a scoping review. Int. J. Med. Inform. 134, 104040 (2020) 20. Ratta, P., et al.: Application of blockchain and internet of things in healthcare and medical sector: applications, challenges, and future perspectives. J. Food Qual. (2021) 21. Lin, W., et al.: Blockchain technology in current agricultural systems: from techniques to applications. IEEE Access 8, 143920–143937 (2020) 22. Gupta, R., et al.: Blockchain-assisted secure UAV communication in 6G environment: architecture, opportunities, and challenges. IET Commun. 15(10), 1352–1367 (2021)

Emerging Technologies in Education

Implementing UX Model at Dijlah University College Omar Sabraldeen Aziz(B) , Zeena Tariq, and Zena Hussain Dijlah University College, Baghdad, Iraq {omar.sabri,zeena.tariq,zena.hussain}@duc.edu.iq

Abstract. In software systems, the User Experience (UX) is a critical success factor. Many E-Learning systems confront difficulties when it comes to UX. Practitioners, instructors, and students can better handle these difficulties in the future if they have a better grasp of them. The number of UX issues is growing every day. Despite the importance of user experience for E-Learning systems, there are few studies that have been undertaken to assess the level of acceptability and factors impacting E-Learning system acceptance in the Middle East. The majority of these research compared E-Learning systems to one or two UX factors at most. In this context, this study examines the current state of practice in E-Learning systems and the most significant success elements that E-Learning systems should possess. A survey of Dijlah University College students was performed to achieve this. A total of 100 questionnaires were delivered through a survey to Dijlah University College workers and students. Employees and students at all levels took part in the poll. The survey questions were developed based on past research findings. The findings show that six success factor determinants, namely helpful, useable, findability, desirable, accessible, and valuable, had a substantial influence on employees and students’ willingness to use an E-Learning system. The questionnaire responses were tested in two ways. The first method is to use SPSS to determine the relationships between the selected UX elements, and the second method is to use Sortsite, an online application, to analyze the compatibility of Dijlah University College’s current E-learning system vs the indicated UX factors. The findings show that understanding the identified UX factors and their relationships will assist decision makers in the future in identifying the reasons for the acceptance or rejection of using an E-Learning system among employees and students, and ultimately in assisting them in improving E-Learning system acceptance and usage. Keywords: User Experience · E-learning · Useable

1 Introduction Recently, studies and research on E-learning systems in the field of User experience (UX) have grown significantly in order to bring additional benefits to universities and enterprises. The term “user experience” is used by researchers and practitioners in many fields to improve their systems (Trevor Barker, 2010). User experience (UX) is defined © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 319–336, 2023. https://doi.org/10.1007/978-3-031-20429-6_31

320

O. S. Aziz et al.

as the interaction between the user, the system, and the context, which describes not only a professional practice, but also a resulting outcome. A user experience designer uses a set of techniques to improve a user’s requirements in order to translate them into products and services. The user experience (UX) professional seeks out appropriate tools and aspects to completely meet the user’s requirements (Buley, 2013). One of the UX implementations is making effective attempts with E-learning systems to improve students’ skills. To comprehend an E-learning system, describe it as a set of instructions presented on a digital device such as a computer or mobile device that is intended to facilitate learning. This definition includes various parts that address the what, how, and why of E-learning. The purpose of E-learning is to help individuals achieve personal learning goals or to create job-transferable knowledge and skills tied to organizational performance (Law, 2014). Many E-learning models are created as local systems and are based on some traditional UX characteristics. All of these aspects have been altered in recent models in order to deal with new factors for distributed systems over the Internet. UX also has important factors. The study looks at all of the main user experience aspects that affect E-learning systems and contribute to their usability. Several researches have explored and developed E-learning acceptance features in order to achieve them in their systems, which must be high-tech in order to transfer them to new variables of assessment and testing these E-learning systems. This study will use a table to evaluate the existing UX aspects in current e-learning systems, as well as a questionnaire to collect data. The questionnaire will cover nearly all aspects of the user experience related to an E-learning system model for limiting all problems. The study is prompted by Dijlah University college students’ interest in E-learning and their desire to learn more about their university’s website. Despite the fact that the use of E-learning at universities is quickly increasing, little is known about students’ expectations and experiences, and the majority of E-learning systems available do not match the demands of students and instructors. According to many research, a lack of understanding of UX aspects is the leading source of unhappiness and a lack of motivation to utilize these systems. Applying good UX aspects, on the other hand, is a surefire way to succeed. Many UX criteria were not used or taken into account when designing E-learning systems, resulting in poor E-learning system usage. This study intends to analyze the success elements associated with E-learning design and suggest a new UX model to fulfill the needs of students and instructors. After evaluating the current conditions and critical requirements using a questionnaire designed for this purpose, this research contributed to the investigation of UX factors that are strongly related to one another in order to determine the important factors that influence student and instructor acceptance at Dijlah University College. Our paper discusses and describes the key ideas, benefits, tools, and applications for creating a user experience model, which is crucial to the advancement of modern e-learning systems. UX elements gathered from related research are the focus of this chapter. All research studies looked at one or more factors, but none looked at as many as six as this study did. Nearly all of the linked user experience system aspects that are driving the development of the suggested UX model for an e-learning system are included in the survey. In order to construct a questionnaire and assess the present UX

Implementing UX Model at Dijlah University College

321

model with the current E-learning system utilized at Dijlah University, the study also examined these criteria.

2 Our Methodology This study will incorporate a number of acceptable UX variables that have never been considered previously in order to construct an E-learning system model. Figure 1 depicts the research approach and its steps, which are organized as follows: 1. Investigating and assessing the impact of user experience variables on e-learning as shown by prior research. The study will identify an evaluation instrument that can be used with the current system. 2. For the first phase, a questionnaire will be created based on the UX characteristics chosen in order to apply the acceptance UX model proposed in this study. 3. The findings will show whether or not a UX aspect in the acceptance model is valid.

Studying and Analyzing literatures survey

Implementing SortSite tool to test the current model

Preparing a questionnaire to measure the acceptance of the UX model proposed in this research

Data Collection

The result will demonstrate whether all UX factors in the ac-

Fig. 1. The proposed methodology

The study will construct a recommended questionnaire based on UX aspects that will focus on the E-learning design model. The design model will use novel techniques to get user approval, which is a major goal of the research. The six elements are chosen in order to improve the proposed design model, as stated in chapter one. Figure 2 depicts the proposed model, which is based on the factors chosen. Useful, useable, findability and navigation, appealing, accessible, and value are some of these qualities.

322

O. S. Aziz et al.

The six elements are briefly outlined in order to clarify their roles and impacts on the model. USEFUL

FACTORS

USABLE

FINDABLE and Navigation

A UX MODEL FOR E-LEARNING SYSTEMS

DESIRABLE ACCESSIBLE VALUABLE

Fig. 2. The proposed model

– Useful: This criterion is used to assess the value of data, and the design will be based on the user’s experience. This aspect has an impact and helps to support E-learning services (Nielsen 1996). – Usable: This aspect indicates how user-friendly the website is in terms of presenting the user experience in the E-learning industry by addressing some of the website design problems. The factors of utility and usability are critical for any model’s acceptability (Qureshi and Irfan 2009). – Findable and Navigation: This component focuses on the model’s organization and exposes the relationships between the user experience objects logically and clearly. Findability is a term used in web design to describe how easily a user can discover the information they need (Rosenbaum 2010). – Desirable: This aspect is concerned with creating a unique identity and impression in the design in order to pique the user’s interest in dealing with the website’s unique qualities (Allen 2007). – Accessible: This component is tailored to all users (abled and disabled) in order to deliver the experience in a desirable manner in order to convey the E-learning model’s objectives. Some companies disregard this issue as a waste of time and money because they are unconcerned about disabled users, yet it is critical that all people have access to the system’s benefits (Allen 2007) – Valuable: This is a goal of any product or design, and the designer should be concerned with presenting his model in the best possible way. To be valuable, the designer should deliver his work while considering all of the above factors (Rosenbaum 2010). In order to analyze UX elements such as accessibility, usability, information architecture (IA), and questionnaire structure, the study will use a testing model that includes some of the above-mentioned factors.

Implementing UX Model at Dijlah University College

323

To evaluate the E-learning model and test the user experience, the study suggested five assumptions. To validate the model, the study will use factor analysis. The following five assumptions will be examined in Chap. 5 using the SPSS framework: 1. There is a positive association between the E-learning system and the UX factor’s usefulness. 2. There is a positive relationship between the usability of the UX component and the E-learning system. 3. There is a positive association between the E-learning model and the UX factors of findability, navigation, and accessibility. 4. There is a positive association between the E-learning system and the UX aspect Desirability. 5. There is a positive relationship between the E-learning system and the value of the UX component.

3 Experiments and Rsults 3.1 Design UX Model Because this is an exploratory study, the questionnaire creator should capture nearly all relevant information about the UX elements. The questionnaire was created using a combination of UX and E-learning design considerations (Bargas-Avila and Hornbaek 2011).The survey instrument was pre-tested to ensure that the survey ideas were clear, and selected constructs were prepared prior to data collection. Face-to-face interviews with seven staff and students at random were conducted in order to identify and correct any flaws in the questionnaire. The data was collected using a structured questionnaire. 3.2 User Experience Application In this section, the research examines the characteristics of user experience (UX) and its applications, as well as the improvements that must be made after these elements are implemented in a system. Some of these characteristics are presented in the study as follows: 1. At any point during the design process, the UX designer has the highest level of recognition and foresight to translate these experiences into usable interactions and facilities. These features should be improved so that they are appropriate for all types of experience. 2. The texture of the experience should be very apparent in order for the designer to accurately transform their characteristics. 3. Any user experience has detailed references (events, papers, etc.) that should never be overlooked in order to convey the complete picture of that user experience. 4. The worth of a UX model will be a valuable model after it has performed as expected and is very beneficial. 5. As already said. Because they deal with the intricate nature of material in the world, designers found UX definitions to be a huge problem.

324

O. S. Aziz et al.

3.3 User Experience Tools The report shows the most prevalent UX tools used in E-learning systems in this part. Table 1 shows a collection of these tools, together with descriptions of their nature and function. 3.4 Evaluation Tools This section is divided into two parts: the first will cover the assessment tools used to test the existing E-learning system, and the second will examine the survey findings gathered through user questionnaires. The research looked at a variety of websites on the internet and on the market to determine the best evaluation tools. Some of these resources are available for free, while others are not. The research preferred to go over the useful tools stated in the study and choose the ones included in the Table 2. The first of the two tools in the Table 2 is based on the number of users in order to collect the e-learning system’s common for use part. The SortSite is an online evaluator system that can be installed on the user’s computer and used to assess all aspects of an E-learning system. The findings are presented in the form of a report that examines the weak and strong points. The SortSite tool was chosen for the study because it is effective, free, and simple to use.

4 Analysis and Test This section implements the SortSite test on the current E-learning system at Dijlah University College and studies the results of testing and evaluating 62 questionnaire forms obtained from 62 users of the E-learning system, according to the phases of methodology in chapter three. The following objectives are the emphasis of this chapter: 1. To assess and improve the University’s current E-learning system. 2. To investigate and test the impact of UX aspects on the E-learning system and all users (students and staff members). 3. To investigate the issues that have arisen as a result of the growth of E-learning systems. 4. To compile the results of the surveys that were sent out in order to determine the qualities of a new model. 5. To answer a study topic that is built on past UX system expertise. 4.1 The Results of SortSite Comparisons Two more E-learning systems for various universities were tested in the study. Isra University college is one of them. Baghdad University is the other. The three evaluation findings are compared in this study to see if there are any differences. It became evident how to interpret the SortSite findings. The study compared the findings of universities in Table 3 to highlight the percentages of weakness and strength

User Experience testing, Usability testing, accessibility testing

Usability testing platform offers both self-moderated and moderated tests

User testing, Website testing, prototype testing

UserZoom1

UserTesting2

Validately3

Loop114

1

2

3

4

3 Validately: https://validately.com 4 Loop11: https://www.loop11.com

1 UserZoom tester: https://www.userzoom.com 2 User Testing: https://www.usertesting.com

Function

Provides user with quantities, qualitive data, and great for uncovering trends as well as synthesizing and analysing post test data

Tool name

Seq

Table 1. Evaluation tools Description

(continued)

It is an online user testing and survey tool that lets you run your own usability studies without the need for a usability lab, specialized equipment or moderator. The practitioner determines the tasks they want to test on their website, recruits participants and launches the study

Validately is a comprehensive solution for user research that includes a dedicated testing platform, recruiting services, and automated reporting. The team help a user spend more time actually focused on research

The software is held wildly divergent political opinions about the Affordable Care Act. Without a wide range of customer feedback, there was a great risk that the TurboTax team would design an experience that matched the needs of only a small portion of their customers

Influence the powerful combination of User Videos and UX Analytics, for the quantitative and qualitative data needed to interpret and improve the relationship between user experience and business KPIs

Implementing UX Model at Dijlah University College 325

User testing, Navigation Testing, performance testing

A/B testing

Usability testing

Accessibility testing

UX, Performance testing

UsabilityHub5

Crazy Egg6

Userlytics7

Functional Accessibility Evaluator8

Hotjar or Mouseflow9

5

6

7

8

9

9 Mouseflow: https://mouseflow.com/

7 User lytics: https://www.userlytics.com 8 Function Accessibility Evaluator: https://fae.disability.illinois.edu

5 Usability hub: https://usabilityhub.com 6 Crazy Egg: https://www.crazyegg.com

Function

Tool name

Seq

Table 1. (continued)

(continued)

Mouseflow tracks clicks, mouse movement, scrolls, forms, and more. It shows an anonymized recording of the activity from each visitor on your site - just like CCTV

The Functional Accessibility Evaluator (FAE) evaluates a website or a single web page based on the W3C Web Content Accessibility Guidelines (WCAG) 2.0 Level A and AA requirements

It provides remote unmoderated usability testing for both desktop and mobile

Crazy Egg is a heat map software designed to help you understand your users better: why visitors are leaving your website, how users click and scroll through your content, where they’re coming from to begin with, and who clicks on what the most

It is an online usability testing platform designed to capture user feedback through a variety of tests

Description

326 O. S. Aziz et al.

UX, Usability, conversion testing

UX, Usability, Accessibility

ClickTale10

Usabilla11

Eye tracking12

10

11

12

12 Eyetracking: https://www.eyetracking.com

11 Usabilla: https://usabilla.com/

10 Clicktale: https://www.clicktale.com/

Function

UX, Usability testing

Tool name

Seq

Table 1. (continued)

Eye tracking is analysing process for individual’s optical focus to measure focus time, point of focus, and eye movement. Eye tracking pursue the look at the eye as it focuses on various objects

This testing tool allows you to test both conversion and user experience. With nine individual applications that you’d otherwise pay for separately, Usability Tools allows you to access a user research panel, create feedback forms, record visitor activity, test web forms and more

A market leader in user testing, ClickTale allows you to watch video playback of how users experience your site on both desktop and mobile devices. Doing so will help you to eliminate pain points, increase conversions and grow your business

Description

Implementing UX Model at Dijlah University College 327

328

O. S. Aziz et al. Table 2. The evaluation tools

Seq Tool name

Function

Number of users

1.

Eye tracking Evaluating usability, navigation and accessibility factors 100 by Eye movements

2.

Loop11

Evaluating

100

3.

SortSite

Evaluating Accessibility, Usability, and Design faults

Online test

of SortSite’s criteria. For example, Dijlah University college received the greatest percentage of accessibility because SortSite discovered six accessibility issues on six pages, followed by Isra University college and Baghdad University. Baghdad University has the largest percentage of students answering to the Quality criteria, followed by Dijlah University college and Isra University College. Dijlah University college has the highest Usability percentage, followed by Baghdad University and Isra University college. 4.2 Classification and Descriptive The results of the questionnaire were evaluated for the E-learning system design and UX factors were implemented on the system. The study tested the results for ensuring that the project is achieving its objectives. The study also calculated standard deviation and Skewness values to measure the distributions of collected data from the questionnaire. These results are very important to evaluate the questionnaire standardization. The Table 4 illustrates the results of descriptive. The values of standard deviation, Skewness factor, and Kurtosis factor are shown in Table 4, and they are labeled as high or low values since the values of all the factors are normally between + 2, − 2, but the best values are closest to Zero. All of these values will be analyzed and discussed in the study to ensure that they are accomplished in the proper order. Table 4 lists 62 standard deviation values that are less than + 1, with the exception of a few that are greater than 1. As a result, the findings are close to zero, indicating a wide range of values. The table also demonstrates that the Skewness and Kurtosis ranges of values are obtained from normal distributions and contain relatively low spread values, indicating that the study produced acceptable results. 4.3 Data Distribution The relationships between the participant’s answers and users’ preferences were investigated in order to determine which UX factor is more successful toward Dijlah E-learning system. The study performs a statistical analysis using the SPSS program to illustrate the rate of data ratio of the six factors at Dijlah e-learning system in the following subsections. The diagram below explains and distinguishes what can be improved in the future, including the completion of studies and research, as well as the collecting of a continual number of questionnaires to expose new information and facts.

53% have issues, worse than average

12% have issues, better than average

42% have issues, worse than average

4 pages with search engine issues

9 pages with quality issues

3 pages with privacy issues

2 pages with browser specific issues

17 pages and files checked

Quality

Valuable – Errors 2 pages with broken links or other errors

7 pages have W3C standards issues

Findability

Valuable -Standards

Valuable Privacy

Desirability Compatibility

Total

12% have issues, worse than average

42% have issues, worse than average

24% have issues, worse than average

6% have issues, better than average

1 pages with usability issues

Usability

48% have issues, worse than average

51% have issues, worse than average

51% have issues, worse than average

10% have issues, better than average

53% have issues, worse than average

45% have issues, worse than average

49% have issues, worse than average

99 pages and files checked

47 pages with browser specific issues

43 pages with privacy issues

50 pages have W3C standards issues

9 pages with broken links or other errors

52 pages with quality issues

44 pages with search engine issues

48 pages with usability issues

50% have issues, worse than average

Benchmark

Pages

36% have issues, worse than average

6 pages with accessibility problems

49 pages with accessibility problems

Pages

Benchmark

E-Learning - Baghdad University

E-learning – Dijlah University College

Accessibility

Factors

Benchmark

16% have issues, worse than average

0% have issues, better than average

40% have issues, worse than average

26% have issues, worse than average

40% have issues, worse than average

31% have issues, worse than average

40% have issues, worse than average

40% have issues, worse than average

99 pages and files checked

15 pages with browser specific issues

0 pages with privacy issues

39 pages have W3C standards issues

25 pages with broken links or other errors

39 pages with quality issues

30 pages with search engine issues

39 pages with usability issues

39 pages with accessibility problems

Pages

E- learning – Isra University College

Table 3. The results of SortSite for the comparison of three universities

Implementing UX Model at Dijlah University College 329

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

Q1.2

Q1.3

Q1.4

Q1.5

Q1.6

Q1.7

Q1.8

Q1.9

Q1.10

Q2.1

Q2.2

Q2.3

Q2.4

Q2.5

Q2.6

Q3.1

Q3.2

Q3.3

Q3.4

Q3.5

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

1.00

2.00

2.00

1.00

2.00

1.00

1.00

1.00

1.00

1.00

3.00

2.00

2.00

Statistic

Statistic

62

Minimum

N

Q1.1

Question

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

Statistic

Maximum

226.00

216.00

215.00

243.00

218.00

176.00

245.00

226.00

233.00

253.00

278.00

223.00

230.00

228.00

221.00

202.00

181.00

239.00

270.00

262.00

262.00

Statistic

Sum

3.6452

3.4839

3.4677

3.9194

3.5161

2.8387

3.9516

3.6452

3.7581

4.0806

4.4839

3.5968

3.7097

3.6774

3.5645

3.2581

2.9194

3.8548

4.3548

4.2258

4.2258

Statistic

Mean

0.92500

0.95371

1.11205

0.99669

1.11240

1.08935

0.93085

1.17483

1.15497

0.83565

0.71842

1.10824

0.89419

1.11288

0.96871

0.97401

1.17764

0.95552

0.54613

0.58448

0.66331

Dev. Statistic

Std

Table 4. Descriptive results

0.304 0.304

−0.421 −0.635

0.304

−0.784

0.304

0.304

−0.832

−0.360

0.304

−1.023

0.304

0.304

−0.503

0.304

0.304

−1.586

−0.861

0.304

−0.775

−0.411

0.304

−0.521

0.304

0.304

−0.797

0.174

0.304

−0.356

−0.749 0.304

0.304

−0.014 0.304

0.304

−0.579

−0.328

0.304

−0.632

0.099

Std. Error 0.304

Statistic

Skewness

0.737

0.135

−0.532

0.263

−0.461

(continued)

0.599

0.599

0.599

0.599

0.599

0.599 0.599

−0.389

0.599

0.599

0.599

0.599

0.599

0.599

0.599

0.599

0.534

0.022

0.528

−0.522

2.930

0.082

−0.343

0.226

−0.352

0.599

0.599

−0.889 −0.078

0.599

0.599

0.599

0.599

Std. Error

0.309

−0.822

2.370

0.908

Statistic

Kurtosis

330 O. S. Aziz et al.

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

62

Q4.2

Q4.3

Q4.4

Q4.5

Q5.1

Q5.2

Q5.3

Q5.4

Q5.5

Q5.6

Q6.1

Q6.2

Q6.3

Q6.4

Q6.5

Q6.6

Q6.7

1.00

2.00

1.00

2.00

2.00

1.00

1.00

1.00

1.00

2.00

1.00

1.00

2.00

2.00

1.00

2.00

1.00

1.00

1.00

Statistic

Statistic

Q4.1

Minimum

N

Q3.6

Question

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

5.00

Statistic

Maximum

241.00

240.00

195.00

259.00

218.00

234.00

251.00

250.00

213.00

245.00

230.00

225.00

239.00

273.00

219.00

242.00

149.00

235.00

210.00

Statistic

Sum

3.8871

3.8710

3.1452

4.1774

3.5161

3.7742

4.0484

4.0323

3.4355

3.9516

3.7097

3.6290

3.8548

4.4032

3.5323

3.9032

2.4032

3.7903

3.3871

Statistic

Mean

0.88900

0.91408

0.93820

0.71344

0.88228

1.04676

0.91306

0.86778

1.04992

0.89493

0.99815

1.01196

0.93820

0.77797

1.08216

0.90009

1.27343

1.04233

0.89360

Dev. Statistic

Std

Table 4. (continued)

0.304

−0.502

0.304

0.304

−0.899

0.304

0.304

−0.841

−0.786

0.304

−0.526

−0.535

0.304

−0.470

0.304

0.304

−0.299

−0.176

0.304

−0.757

0.304

0.304

−0.563

0.304

0.304

−1.285

−0.554

0.304

−0.567

0.097

0.304

−0.502

0.304

−0.729 0.304

0.304

−0.289 0.667

Std. Error

Statistic

Skewness

0.860

−0.396

0.049

0.599

0.599

0.599

0.599

0.599

−0.659 0.163

0.599

0.599

0.599

0.599

0.599

0.599

0.599

0.599

0.599

0.599

−0.522

0.857

1.118

−0.252

−0.546

−0.485

0.530

−0.439

1.338

−0.240

0.599

0.599

−0.429

0.599

−0.514

0.599

Std. Error

0.134

0.468

Statistic

Kurtosis

Implementing UX Model at Dijlah University College 331

332

O. S. Aziz et al.

The shape of the box in the lesson demonstrates the torsion signal’s type of distribution as follows: 1. The distribution is moderate if the medium is in the middle of the rectangle, implying that there is no torsion or symmetry (symmetrical distribution). 2. If the medium is closer to the rectangle’s base, the distribution is crooked to the right, indicating positive torsion, which indicates that many data values are low. 3. If the medium is at the top of the rectangle, the distribution is curved to the left, indicating negative torsion, which indicates that many data values are high.

Usability Boxplot chart

Findability & Navigation Boxplot chart

Accessibility Boxplot chart

Useful Boxplot chart

Desirability Boxplot chart

Valuable Boxplot chart

5 Discussion The study’s discussion focuses on the characteristics of the factors in order to provide the agreement and disagreement for each element. The results of the usability factor have a

Implementing UX Model at Dijlah University College

333

high level of agreement, indicating that the stakeholders have a high level of acceptance of the usability features. The study is also concerned with the usefulness of E-learning since the questionnaire contains many questions that evaluate the degree of helpful features, resulting in a high level of acceptance in the useful component because stakeholders reflect the usefulness agreement status. Because the questionnaire collected low values of agreement in various aspects, the study should improve the findable and navigation features to increase the degree of acceptance. The questionnaire has a high value of desirable to improve the E-learning system because some desirable features have low acceptance values, thus the study must improve the desirable component in the current E-learning system. The accessibility factor has a high level of acceptance because it has a high level of agreement, which means that the accessibility factor’s qualities are reflected with a high degree of agreement. The valuable factor also has high acceptance values since it has high agreement values, therefore the valuable factor’s attributes are mirrored in a high degree of agreement.

6 Conclusion The main problem in software engineering has traditionally been to offer the techniques, methodologies, and accompanying tools needed to build high-quality software. Given that E-learning systems face several obstacles these days, one of the most significant is identifying success criteria. In this context, our research was prompted by the unidentified causes of E-inability learning’s to meet the needs of employees and students. A set of research connected to E-learning systems were studied and classified during the literature review, with a focus on success factors and difficulties that affect their success. The goal was to figure out what the most essential success characteristics for E-Learning systems should be. These notions are well-supported by an exploratory research of the existing state-ofthe-practices in E-Learning systems. A survey was done with 62 participants to collect data from Dijlah University College workers and students in order to determine their level of acceptability of the E-learning system. The survey questions were developed based on past research findings. According to the findings, six success factor determinants, namely helpful, usable, findability, desirable, accessible, and valuable, had a substantial impact on employees and students’ willingness to use an E-Learning system. A description of the survey’s findings was included in Experimental part. Based on the foregoing rationale, this study has developed a UX model that is intended to meet the study’s goals. The proposed UX model presents a novel notation for analyzing the amount of acceptability of an E-learning system based on six UX criteria that have never been combined in one research previously, as most studies have only looked at one or two UX factors at most. This study has created a UX model based on the aforementioned logic in order to achieve the study’s objectives. Most studies have only looked at one or two UX variables at most, therefore the proposed UX model gives a novel notation for measuring the level of acceptability of an E-learning system based on six UX criteria that have necxver been integrated in one research before.

334

O. S. Aziz et al.

References Tarhinia A., Hone, K., Liu, X.: User acceptance towards web-based learning systems: investigating the role of social. In: Organizational and Individual factors in European Higher Education, 1877–0509 by Elsevier B.V. (2013) Althobaiti, M.M., Mayhew, P.: Assessing the usability of learning management system: user experience study. In: International Conference on E-Learning, E-Education, and Online Training. Springer, Cham (2015) Sivajia, A., TzuaanSooa, S.: Understanding, Enhancing and Automating HCI Work Practices: Malaysian Case Studies, pp. 656–665. Elsevier Ltd. (2013) Buley, L.: The User Experiennce Team of One. Brooklyn: Rosenfeld (2013) Goi, C.L., Ng, P.Y.: E-learning in Malaysia: success factors in implementing e-learning program 20(2), 237–246 (2009). ISSN 1812–9129 Draper, C.D.: Ships Passing in the Night E-Learning Designers’ Experiences with User Experience. Brigham Young University, PhD thesis (2015) Clark, R.C., Mayer, R.E.: E-Learning and the Science of Instruction: Provenguidelines for Consumers and Designers of Multimedia Learning, 3rd edn. Pfeiffer, San Francisco (2011) Cooper, A., Reimann, R., Cronin, D.: About Face 3: The Essentials of Interaction Design, 3rd edn. Wiley Publishing Inc., Indianapolis, IN (2007) Law, E.L.-C., Roto, V.: Understanding, scoping and defining user eXperience: a survey approach. ACM 719–728 (2009) Eklund, J., Kay, M., Lynch, H.M.: E-learning: Emerging issues and key trends. Australian Flexible Learning Network, Melbourne, Australia (2003) Fraenkel, J.R., Wallen, N.E.: How to Design and Evaluate Research. Mc. Fraw-Hill Inc, USA (1996) ISO.: Ergonomics of human system interaction - Part 210: Human-centred design for interactive systems (formerly known as 13407). Switzerland: International Organization for Standardization (2008) Bargas-Avila, J.A., Hornbæk, K.: Old wine in new bottles or novel challenges? A critical analysis of empirical studies of user experience. ACM (2011). 978-1-4503-0267-8/11/05 Gilbert, J., Morton, S., Rowley, J.: e-Learning: the student experience. British J. Educ. Technol. 38(4) (2007) Tarigan, J.: Factors influencing users satisfaction on e-learning systems. jurnal manajemen dan kewirausahaan 13(2), 177–188 (2011) Robinson, J., Lanius, C., Weber, R.: The Past, Present, and Future of UX Empirical Research, pp. 10–24. ACM (2017) Krug, S.: Rocket Surgery Made Easy: The Do-It-Yourself Guide to Finding and Fixing Usability Problems. New Riders, Berkeley (2009) Harry, B., Santoso, M.S.: Measuring user experience of the student-centered e-learning environment. J. Educ. Online-JEO, 58–79 (2016) Hassenzahl, M.: The thing and I: Understanding the relationship between user and product. In: Blythe, M., Overbeeke, C., Monk, A.F., Wright, P.C. (eds.) Funology: From Usability to Enjoyment, pp. 31–42. Kluwer, Dordrecht (2003) Mohammadi, H.: Investigating users’ perspectives on e-learning: an integration of TAM and IS success model. Comput. Human Behav. 45, 359–374 (2015) Chuang, H.-C., Wang, C.-Y., Chen, G.D., Liu, C-C.: Design and evaluation of an affective interface of the e-learning systems. In: 10th IEEE International Conference on Advanced Learning Technologies (2010) Law, E.L.: Attitudes towards user experience (UX) measurement. Int. J. Human-Comput. Stud. 526–541 (2014)

Implementing UX Model at Dijlah University College

335

Rauschenberger, M., Olschner, S., Cota, M.P., Schrepp, M., Hochschule, J.T.: Measurement of user experience. CISTI 471–476 (2012) Martínez-Argüelles, M., Castán, J., Juan, A.: How do students measure service quality in elearning? A case study regarding an internet-based university. In: Proceedings of the European Conference on E-Learning (2009) Allen, M.W.: Designing Successful e-Learning, Pfeiffer, 989 Market Street, San Francisco, CA 94103-1741 (2007) Mayer, R.C.: E Learning and the Science of Instruction. Pfeiffer, San Francisco (2011) Moebs, S.: QoS-based Experience-aware Adaptation in Multimedia e-Learning - A Learner, is a Learner, is a User, is a Customer. Dublin City University, Dublin (2011) khasawneh, M., Yaseen, A.B.: Critical success factors for e-learning satisfaction, jordanian universities’ experience. J. Busin. Manag. (coes&rj-jbm) issn (e): 2306–7179 (2017). ISSN: 2306–8043 Morville, P., Rosenfeld, L.: Information architecture for the World Wide Web. O’Reilly Media, Sebastopol (2007) Nicholson, P.: A history of e-learning. In: Fernandez-Manon, B., Sanchez-Perez, J.M., GomezPulido, J.A., Vega-Rodriguez, M.A., Bravo-Rodriguez, J. (eds.) Computers and Education, pp. 1–11. Springer, Dordrecht, Netherlands (2007) Nielsen, J.: Usability metrics: tracking interface improvements. IEEE Softw. 13(6), 78–86 (1996) Nurdiana Azizan, M.R.: Islamic knowledge: factors effective Islamic knowledge transfer for visually impaired users via Malaysia e-learning websites. In: The International Seminar on Best Practices for Best Islamic Sharing of Islamic Libraries, pp. 1–19 (2016) Owston, R., York, D., Murtha, S.: Student perceptions and achievement in a university blended learning strategic initiative. Internet Higher Educ. 18, 38–46 (2013) Paechter, M., Maier, B., Macher, D.: Students’ expectations of, and experiences in e-learning: their relation to learning achievements and course satisfaction. Comput. Educ. 54(1), 222–229 (2010) Panagiotis Zaharias, C. P. (2016). Quality management of learning management systems: a user experience perspective. CURRENT ISSUES IN EMERGING ELEARNING, 60–83 Kashfi, P., Nilsson, A., Feldt, R.: Integrating user experience practices into software development processes: implications of the UX characteristics (2017) Parrish, P.: Aesthetic principles for instructional design. Educ. Tech. Res. Dev. 57(4), 511–528 (2009). https://doi.org/10.1007/s11423-007-9060-7 Sun, P.C., Tsai, R.J., Finger, G., Chen, Y-Y., Yeh, D.: What Drives a Successful e-Learning? An Empirical Investigation of the Critical Factors Influencing Learner Satisfaction. Elsevier Ltd (2008) Hammouri, Q., Abu-Shanab, E.: Exploring factors affecting users’ satisfaction toward e-learning systems. Int. J. Inform. Commun. Technol. Educ. (IJICTE) 14(1), 44–57 (2018) Rodrigues, J.J.P.C., Sabino, F.M.R., Zhou, L.: Enhancing e-learning experience with online social networks. IET Commun. 5(8), 1147–1154 (2011) Rosenbaum, S.: Improving The User Experience of Evidence. Avhandling, Oslo (2010) Santoso, H.B., et al. (2016) Measuring user experience of the student-centered e-learning environment. J. Educ. 13(1), 58–79 (2016) Lee, S.: Design of a Learner-Directed E-learning Model. University of Hertfordshire, Ph.D. thesis (2013) Sugant, R.: An Assessment of Factors of Service Quality of E- Learning. Applied Research Series, pp. 1–42 (2016) Ozkan, S., Koseler, R., Baykal, N.: Evaluating Learning Management Systems: Adoption of Hexagonal E-Learning Assessment Model in Higher Education, Al Bostan Rotana, Dubai, UAE. ISBN: 978-1-902316-58-1

336

O. S. Aziz et al.

Alrawashdeh, T.A., Al-Mahadeen, B.M.: Extended UTAUT to examine the acceptance of web based training systems by public sector employees. iJIM 7(1) (2013) Trevor Barker, S.L. (2010). Approaches to student modeling in the context of e-learning 2.0. In: The 9th European Conference on e-Learning, Porto, Portugal., pp. 50–62 Wieschnowsky, T., Paulheim, H.: A visual tool for supporting developers in ontology- based application integration. In: Paper Presented at the 7th International Workshop on Semantic Web Enabled Software Engineering (ISWC) 2011. Bonn, Germany (2011) Wu, J.-H., Tennyson, R.D., Hsia, T.-L.: A study of student satisfaction in a blended e-learning system environment. Comput. Educ. 55(1), 155–164 (2010)

Integration of Face-to-Face Instruction and CALL in Computer-Assisted Cooperative Learning Amr Abdullatif Yassin1,2(B) , Norizan Abdul Razak1 , and Tg Nor Rizan Tg Mohamad Maasum1 1 Faculty of Social Sciences and Humanities, Universiti Kebangsaan Malaysia, Bangi, Selangor,

Malaysia [email protected] 2 Centre of Languages and Translation, Ibb University, Ibb, Yemen

Abstract. This study aimed to investigate how face-to-face cooperative learning and CALL are integrated when teaching through face-to-face Computer Assisted Cooperative Learning (CACL). Although CACL is not a new term and the integration of technology with face-to-face instruction has been discussed in previous literature, there is no clear evidence for a model that can be used as a guide in the process of integrating the two methods, namely face-to-face cooperative learning and CALL, in teaching through CACL. This study employed the qualitative research design, and semi-structured interviews and observation were carried out to investigate the process of integration in face-to-face CACL during teaching reading skills. The study used Neumeier (2005) model as a theoretical framework; however, the findings showed that this model needs to be adapted to suit teaching through face-to-face CACL. The main adaptation is in the first two parameters because face-to-face CACL is considered one mode. It is the leading mode, mainly because face-to-face cooperative learning and CALL are used simultaneously during teaching, making it difficult to count the time of each mode separately. The study concluded with a model for integration in face-to-face CACL, which can be adapted for teaching through synchronous or asynchronous CACL. Also, the paper provided implications for teachers and educators interested in the design and implementation of CACL. Keywords: Computer assisted cooperative learning · Face-to-face cooperative learning · CALL · Integration model

1 Introduction The field of education is moving toward integrating technology into the field of education as it proved to be effective in teaching [1–4]. Integration of technology into the field of education should focus on providing learners with authentic and meaningful learning experiences [5, 6]. Technology integration into learning is more effective when students understand how to use technology to support their learning, which requires focusing on the content to be presented to students [7–10]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 337–360, 2023. https://doi.org/10.1007/978-3-031-20429-6_32

338

A. A. Yassin et al.

However, teachers and educators still need to understand and practice the effective integration of technology in teaching to maximize the benefit to students [11, 12]. That is, the term “integration” is used in different studies to refer to the blending of face-to-face instruction with technology [12–15]. One of the methods of teaching that integrates face-to-face instruction with CALL is Computer Assisted Cooperative Learning (CACL). Many scholars have tried to utilize the advantages of CALL and cooperative learning under the term CACL, which started in the field of education in the 1980s like the studies of [16–19]. These studies showed that CACL is effective in the process of teaching. In this method of teaching, CALL and cooperative learning instructions have complementary advantages for students. That is, students can do different exercises easily on the computer, and they can discuss the lesson with each other [16]. Moreover, using CACL instruction is effective in teaching reading skills, taking into consideration that oral interaction among students helps them to support each other learning [16, 20]. Regarding the integration of cooperative learning and CALL instruction in CACL, Johnson and Johnson [16] stated that good integration planning leads to effective teaching. They recommended the implementation of the principles of cooperative learning in CACL. Also, Brush [21] stated that in integrating cooperative learning with CALL instruction, the teacher should focus on different elements during the process of teaching, including the balance between the individual tasks and the group tasks, the level of students which can be homogenous or heterogeneous according to the need of the teacher, and the teacher guidance to implement cooperative learning with CALL instruction. However, the integration process in face-to-face CACL in learning activities is still vague in previous literature. That is, CACL has two elements, namely cooperative learning and CALL. Planning the integration of these two modes in teaching has not been based on an explicit model in previous literature. In other words, there is no clear evidence for a model that explains the implementation of face-to-face CACL activities. Investigation of the integration in CACL will help teachers and scholars in the implementation process by focusing on the aspects of each mode to meet teaching goals. Therefore, the current study investigates the integration of face-to-face cooperative learning and CALL in face-to-face CACL to develop a model that can be used as a guide for scholars and educators interested in teaching, describing, or designing CACL activities. Accordingly, this study aims to answer the following question: How are face-to-face instruction and CALL integrated with CACL to teach reading skills?

2 Literature Review The idea of integrating technology into education requires more than the facilities and hardware parts of the technology to be used in the process of teaching [22–24]. Teachers need to focus on other factors besides the facilities to provide a successful learning experience for students through giving suitable content, support, training, technology skills, and clear learning objectives. The success of integration is even related to the process of teaching itself [6, 25, 26]; however, such integration might be failed due to complexity [27, 28], when the process of teaching is not clear for the teacher and the students.

Integration of Face-to-Face Instruction and CALL

339

Many studies have focused on the effective integration in CACL. The study of Heba and Nouby [29] showed that the integration of cooperative learning with CALL instruction makes the process of teaching more effective. However, the authors stated that integrating cooperative learning with technology requires more focus on “delivery methods that specifically focus on the role of the adopted instructional design (such as ADDIE), participants’ characteristics, the process of paring/grouping, face-to-face student-student interaction and student-tutor interaction, the appropriate balance between e-learning and face-to-face approaches, and peer-tutoring” [29]. Another study that has focused on providing a model for online collaborative teaching is by Persico et al. [30], who developed a model to monitor online collaborative activities, and this model has four dimensions. The first one is the participative dimension; the second dimension is the social dimension; the third dimension is the cognitive dimension, and the fourth dimension is the teaching dimension. So, this collaborative system focused on group work or collaborative work, and it is clear that behaviourist CALL is a main feature in the system because the system sores the activities of the students for the reference of the teacher and the students. Also, it focused more on the interaction among students and with the teacher. The above two studies have focused on the integration in CACL even though the studies have also used other terms such as CSCL. The two studies showed that integrating cooperative learning with CALL instruction is more effective when it is planned, yet there is no clear evidence for a model that explains how cooperative learning activities are carried out with CALL instruction. One of the prominent works that investigated the integration of face-to-face instruction and technology is Neumeier [13]. In her work, Neumeier came up with a model that describes the successful integration of face-to-face instruction and technology. This model has six parameters, and each parameter has descriptors. She has explained the integration of face-to-face and technology, and she has covered different points for the design or description of the activities and the teaching method. What is effective in this model is the focus on the integration of the two modes, namely face-to-face interaction and CALL instruction. Although the model can be used to describe the integration of technology with collaborative activities, it did not account for cooperative learning activities based on social interdependence theory and has different strategies to be implemented in the process of teaching along with the five principles of cooperative learning. Previous studies, such as [16, 20, 31, 32], investigated the effectiveness of CACL in the process of teaching, but they did not investigate the integration of face-to-face instruction and CALL under the term face-to-face CACL. Therefore, the current study uses Neumeier’s [13] model and adapts it to suit face-to-face CACL, and the findings of the study will provide a model that suits the integration in face-to-face CACL. To sum up, integrating technology into the field of education is effective in the process of teaching, and the focus should be shifted to the effective integration of technology in the process of teaching. Face-to-face CACL integrates face-to-face cooperative learning and CALL; however, previous studies did not provide a clear model for the process of integration in face-to-face CACL in teaching, especially for EFL/ESL students. The investigation of the integration in face-to-face CACL is essential to make teaching more effective and provide educators with the optimal integration of the two modes.

340

A. A. Yassin et al.

3 Theoretical Framework: Integration of Face-to-Face Instruction and CALL Until writing this study, there is no evidence for a clear model that has explained the process of integration in face-to-face CACL. One of the prominent models that have explained the integration of face-to-face instruction with CALL is by Neumeier [13]. This framework looks at how to integrate technology with face-to-face instruction. The main idea of this framework is to investigate the integration of face-to-face instruction and technology. Therefore, this model can be adapted and modified to explain the process of integrating face-to-face cooperative learning and CALL. Thus, Neumeier [13] stated that this model is used to investigate the “combination of face-to-face (F2F) and computer assisted learning (CAL) in a single teaching and learning environment”. This shows that the framework can be applied to face-to-face CACL because it has two components: faceto-face instruction and CALL. However, the missing element is cooperative learning, which requires implementing cooperative learning principles and cooperative learning strategies. Therefore, this framework will be used in this study. It will be adapted and modified according to the findings from the integration process according to the students’ experiences. Neumeier’s [13] framework has six parameters. These parameters can be used to investigate the integration of face-to-face instruction and CALL. The first parameter is the mode which aims to investigate the leading mode in the process of teaching. The leading mode can be face-to-face interaction, and CALL is the secondary mode or vice versa. This parameter also investigates the distribution of modes which refers to the time spent in each mode. Another descriptor of this parameter is the choice of the modes which are related to the components of both face-to-face instruction and CALL. The second parameter is the model of integration which has two descriptors. The first one is the sequencing of modes which refers to how the two modes (face-to-face instruction and CALL) are sequenced. According to Neumeier [13], the two modes can overlap each other. She argued that sequencing modes aim to reduce the transactional distance, which can be achieved through collaborative activities. The second descriptor is the level of integration, which refers to the flexibility of using modes. Some modes can be obligatory as face-to-face instruction, and others are optional, like some CALL features. The third parameter is the distribution of learning content, objectives, and assignment of purpose, which refers to the use of the two modes in teaching the content to the students. The teaching of the content can be parallel or isolated. It can be parallel in the sense that the teacher teaches a particular skill face-to-face and in the lab. Also, it can be isolated in the sense that the teacher can introduce the skill face-to-face, but he does not use the lab to teach that skill. Another point related to this parameter is the objectives of the content, which is the aim of teaching or the teaching course. The fourth parameter is the teaching method, which depends on the teacher to choose an appropriate method of teaching. This parameter has different forms, such as using collaborative learning and a communicative approach. Still, it did not give an account for cooperative learning where the teacher has to implement the five principles of cooperative learning.

Integration of Face-to-Face Instruction and CALL

341

The fifth parameter is the involvement of learning subjects which has different descriptors. The first one is interactional patterns. The second descriptor is the roles of the learners and the teacher. The third descriptor is the level of learner autonomy, which refers to understanding the process of learning through face-to-face instruction and CALL. The sixth parameter is the location which refers to the place of the class. It is important to offer the students a place for learning which is suitable for them. This also depends on the needs of the teacher, so the face-to-face instruction can be in a classroom, and the CALL activities can be at home. Also, the teacher can use CALL in a lab in the school or the university. Therefore, this study will investigate these parameters in face-to-face CACL setting as modifications might appear to suit this teaching method. Changes will be made according to the data the researchers will collect from participants to come out with a new model that explains the integration of face-to-face CACL activities.

4 Methodology 4.1 Research Design This study employed a qualitative approach design as the data were collected through semi-structured interviews and observations [33]. This design is suitable for the study because the researchers needed to describe the process of integration in face-to-face CACL. This helps to get an in-depth understanding of the teaching process and learning activities in the class. Moreover, although the study used Neumeier’s [13] model as the theoretical framework, it is clear that face-to-face CACL is different from the description of the model. Therefore, it was not clear to the researchers how face-to-face cooperative learning and CALL instruction is integrated with CACL. This required collecting data from participants so that the researcher could clarify what is different in the current study from the description available in [13]. 4.2 Participants The current study used purposive sampling because the participants were selected according to their active participation inside the classroom, their ability to discuss ideas with their teacher and classmates, and their educational level [33]. This will help to get an in-depth discussion with them since they can express their ideas and elaborate on their opinions during the interviews. Also, the different levels of education help to get an idea during the interviews from a different perspective instead of focusing on one education group of students. Accordingly, background information of the five participants, who volunteer to participate in the study, is shown in Table 1. In terms of the teacher, he is 26 years old, and he holds a bachelor’s degree in English Language Studies. Also, he has CELTA certificate and is expert in teaching English with technology to EFL learners.

342

A. A. Yassin et al. Table 1. Students’ background information

No

Stage of study

Gender

University

Major

1

Postgraduates

M

Uniza

Pharmacology

2

M

Limkokwing

MBA

3

M

UKM

Molecular biology

4

M

UM

Architecture

5

M

APU

Telecommunication engineering

6

Undergraduates

M

UTeM

Software engineering

7

M

UPM

BA

4.3 Process The researchers first designed a website in order to teaching reading skills through faceto-face CACL. The design of the website followed ADDIE model. This model has five steps, namely Analysis, Design, Development, Implementation, and Evaluation. The analysis was for finding the needed reading skills to be studied by the students, the Design was for selecting the materials, the Development was for creating the website, the Implementation was for teaching, and the Evaluation was for the outcomes of the implementation. The process of teaching was through the implementation of the five principles of cooperative learning, namely positive interdependence, promotive interaction, individual responsibility, social skills, and group processing. Also, the teaching used Students-Team Achievement Divisions (STAD), which included introducing the skills, group exercises, individual exercises, and rewarding the top team. The process of teaching is shown in Fig. 1. 4.4 Data Collection The data was collected through semi-structured interviews with five participants and the teacher. Also, the researchers have collected data through observation. That is, the first author has made in-depth interviews with five students until reaching the saturation point, where no new ideas emerged from the interviews (Creswell 2009). The interviews with the students resulted in 51 pages. Also, the interview with the teacher lasted for 42 min. In terms of observation data, the first author has attended the 16 lectures of the course as a non-participant observer during which he noted all the activities inside the class as well as the teaching and learning of reading skills through face-to-face CACL.

Integration of Face-to-Face Instruction and CALL

343

Face-to-Face CACL

Web-based CALL

Cooperave learning Principles

Introducing the skill

Tutorials

F2F cooperavely: Principles 1,2,3,4

Cooperave exercises

Exercises

F2F cooperavely: Principles 1,2,3,4

Individual exercises

Individual exercises

Individual pracce: Principle 3

STAD

Group Processing & Reward

F2F cooperavely: Principle 5

Fig. 1. Teaching reading using face-to-face CACL and STAD

4.5 Data Analysis The semi-structured interviews were carried out in Arabic and English according to the request of the participants to give them a chance to express themselves well. After that, all the interviews were transcribed, and the interviews in Arabic were translated into English. All the interviews were sent back to the participants to revise and modify them according to their convenience. All the participants responded that the interviews are identical to their answers in the interviews. After that, all the interviews were decoded and categorized into themes (Creswell 2009). The observation data is used for triangulation to support the themes that appeared in the interviews. To ensure the trustworthiness of the data, member checking was utilised so that the data conveys the participants’ ideas, and the three researchers have revised the themes together, which is another method to ensure the trustworthiness of the data.

344

A. A. Yassin et al.

4.6 Findings This study aimed at investigating the process of integrating face-to-face cooperative learning and CALL and used the model of Neumeier (2005) as a theoretical framework. This model has six parameters, and there are descriptors for every parameter. This model describes the integration of face-to-face instruction and CALL. Moreover, the situation of face-to-face CACL is different since there are cooperative learning principles and cooperative learning strategies that should be implemented in the classroom. Therefore, this question will investigate how face-to-face cooperative learning and CALL are integrated taking into consideration the six parameters as well as cooperative learning principles and STAD strategy principles. The analysis will follow the parameters of Neumeier’s model, and the findings will be described according to the experience of students in studying reading skills through face-to-face CACL. 4.7 Mode The first parameter is the mode which refers to using face-to-face cooperative learning and CALL. However, this study uses face-to-face CACL as one method, and the students stated that they cannot separate between face-to-face cooperative learning and CALL. They found it as one method because both modes, namely face-to-face cooperative learning and CALL, are used from the beginning of the class until the end. The students’ expressions concerning this point are provided below. S1: “[face-to-face cooperative learning and CALL] cannot be separated from each other. … they are together.” S2: “I think they [face-to-face cooperative learning and CALL] are integrated with each other. Computer and cooperative learning are integrated with each other. It is true that the student might use the computer alone, aaa but aaa he will not get the benefit which he came to get. For example, in the reading, he will read normally as if he is reading a book, but cooperative learning gives you the information in a nice way as groups and as a group activity. The student might lack things, and this thing is available with his classmate. This makes it cooperative.” The comments of the students showed that it is difficult to separate between CALL and cooperative learning in face-to-face CACL during the classes because both are used together during the whole process of learning. This leads to an important point that it is not obligatory in face-to-face CACL to differentiate between the two modes or decide on the leading mode. This is because face-to-face CACL should be treated as one mode, and it is the leading mode. This also leads to another important finding concerning the descriptor of the first parameter namely sequencing of mode. That is, it is difficult to separate between CALL and face-to-face cooperative learning in face-to-face CACL, so it is also difficult to discuss the distribution of modes. That is, it is difficult to count the time of using each mode separately inside the class because the activities of face-to-face cooperative learning and CALL are used together from the beginning of the class until the end.

Integration of Face-to-Face Instruction and CALL

345

This theme is supported by observation because the students used to study using the two modes simultaneously from the beginning of the session until the end. So, the study of reading skills was not dependent on one mode only. Accordingly, the time of each mode cannot be counted in face-to-face CACL. That is, it is difficult to count the time of face-to-face cooperative learning alone or the time of CALL alone. The two modes are integrated to form one mode. This shows that the first parameter should be modified to suit face-to-face CACL because they are used simultaneously. So, the mode should be face-to-face CACL, which is also the leading mode.

5 Model of Integration This section discusses two factors namely sequencing of modes and level of integration. These are the two descriptors of the second parameter given by Neumeier (2005) for integrating face-to-face learning and CALL. In the context of this study, the discussion is on the integration of face-to-face cooperative learning and CALL. a. Sequencing of Individual Modes This descriptor refers to the use of both face-to-face cooperative learning mode and CALL mode, and it investigates whether these two modes are parallel or overlap. The two modes in this study, face-to-face cooperative learning and CALL, are used together which makes it difficult to distinguish between the two modes. This is discussed in the previous parameter, but what is important is the effect of the integration on the transactional distance. In other words, this descriptor aims to minimize the transactional distance among students. The transactional distance refers to the physical distance between the students when they study together. The students’ comments on this theme are shown below. S3: “I think this is the benefit of group study and the theoretical and the computer activities. This prevents boredom so you do not have time to get distracted.” S4: “aaaa … this method of teaching is good to get the desired benefit, and the benefit which they will get. The benefit will be higher than all the other methods. Also … aaa there will not be boredom because this method keeps you active during the whole class … you will not be distracted with yourself… you will not be distracted with the other students.” Accordingly, one of the advantages of the integration is to make the study interactive and to lessen the distraction among the students. Using face-to-face CACL instruction keeps the students active during the whole class which minimizes distraction to a great extent. This is because the transactional distance between the learners is not available, and the steps of the learning are linked together to minimize the distance of communication and interaction between the learner.

346

A. A. Yassin et al.

This leads to an important finding that face-to-face CACL is an effective method to reduce the transactional distance between learners because cooperative learning activities and CALL instructions are used simultaneously. So, the improvement of reading skills among students is also attributed to the simultaneous implementation of face-to-face cooperative learning and CALL instruction, which helped to reduce the transactional distance among the students. To put it simply, the integration of face-to-face cooperative learning and CALL in teaching reading shows that face-to-face CACL uses the two modes simultaneously which allows the students to study together using the web-based CALL. This integration allows the students to interact with each other face-to-face, and this reduces the transactional distance to a great extent. b. Level of Integration This descriptor investigates the optional and the obligatory sub-modes in doing the activities. That is, it investigates if some features of web-based reading are optional for the students. In this study, the website provides tutorials and exercises which are obligatory for the students. Moreover, the study used face-to-face CACL in which faceto-face cooperative learning and CALL modes are obligatory during the whole class. That is, CALL is obligatory in all the activities, and students must do all the activities inside the classroom cooperatively using the tutorials and exercises in the web-based CALL. This descriptor aims to make the learners know how to study using face-to-face CACL, so they know how to study and what are their responsibilities. This is also related to learning autonomy, which means that the students should understand the process of learning and how to use the two modes autonomously inside the classroom. This helps to make the integration of the two modes easier and more fruitful for the students. This theme is shown in the students’ comments below. S2: “yes, at the beginning there was a complete dependency but with the passage of time this weak student should participate. Thus, when we answer a passage, the group members keep the answers and we move to the weak student and ask this student about his answer. He replied that my answer is this. We ask him, how did you reach to this answer? If the excellent student starts with his answer, the argument is finished. Therefore, the weak student starts with his answer and then the excellent student gives his answer. Discussion happened after that why the weak student chose A as the answer or why the excellent student chose C as the answer.” S3: “If we did not know what to do in every stage, it would be difficult. Okay … but the situation was easy because it was divided. You start receiving the skill introduction … face to face. Then you study in group. Then you do exercises in groups and sometimes as individuals. The division for time in the class made the style easy so that we accept them together … the group, the computer and the teacher."

Integration of Face-to-Face Instruction and CALL

347

6 Distribution of Learning Content and Objective and Assignment of Purpose This parameter looks at whether the two modes are isolated or parallel in teaching reading skills. In other words, it investigates how the two modes are used in teaching reading skills. This study used face-to-face CACL which made the two modes integrated so that they are used simultaneously in teaching reading skills. That is, the teacher introduces the skill using the projector, then the students read the tutorials in the computer in order to make sure that they have understood the materials properly. After that the students move to the practice phase, in which they practice reading exercises cooperatively. After cooperative practice, the students move to individual exercises to make sure that they have mastered the skills. At the end, there is a reward for the top team. Therefore, teaching reading skills and doing exercises in the website were taught through face-toface CACL, in which the two modes are used simultaneously from the beginning until the end of the class. The students explained the process of teaching reading skills, which depended on face-to-face CACL instruction from the beginning of the class until the end as it is shown in their comments below. S4: “It is Computer Assisted Cooperative Learning so we are using different things. Like for example, we are using different things … we are using our laptops/the computers, and we are interacting with the lecturers, we are interacting with the groups, we are solving problem in groups then in individual. So, it was helpful for me.” S5: “At the beginning you get instructions [about the reading skill] from the teacher, then you study with your group using the computer, and then you practice the exercises using the computer individually.”

7 Teaching Method In this parameter, there are three sources that influence the teaching method which are the self-access online material, the online tutor, and the face-to-face teacher. However, in the current study, there is no online tutor. So, the focus will be on reading materials, teaching methods, and STAD strategy. These three elements are discussed in this section with reference to excerpts from the interviews. a. Materials The website materials introduced a wide variety of reading skills on different topics. In terms of content, students stated that the materials are new and suitable for them, which helped them to gain a lot of benefits and to improve their reading skills. The main theme related to the content is the diversity of the topics of the passages. That is, the content was suitable because the students came to read different passages with different topics. The theme of the diversity of the topics is discussed by the students as it is shown in the comments below. S2: “aaa it was interesting. What I noticed is that aaa the content was not directed to a specific field. It was diverse because some of the passages were in the scientific

348

A. A. Yassin et al.

field and some were in the literary field, and some of them aaaa we can say that aaa some of them are about political sides aaa such as the leaders of America and the European countries. There were different topics, and this diversification was enriching as it helped us to aaa avoid the fear of reading in different fields. Therefore, it was very excellent and the content was excellent.” Teacher: “I believe that the content will help them to study at the university … the content is very good to help them to achieve their goals in reading at the university.” Therefore, the content was suitable for the students because it provides different topics related to different majors. This made learning interesting for the students and helped them to be involved in challenging reading tasks. Also, the students could come to know many new vocabularies. Accordingly, the design of the CALL materials is vital to making face-to-face CACL effective. In other words, behaviourist CALL and cognitive CALL are highlighted in the comments of the students above ssince the materials provide different exercises with challenging content, topics, and vocabularies. These elements made face-to-face CACL instruction helpful to improve the students’ reading skills. Besides, this section shows that the ADDIE model was suitable to design the web-based CALL as it guided the researchers to provide the materials according to the needs of the students. b. Teaching Method The teaching method is face-to-face CACL in which face-to-face cooperative learning is used with CALL instruction. In this method, both face-to-face cooperative learning and CALL are used together in all the classes. This theme is discussed by the participants, who explained the process of learning. Based on their discussion, it is clear that the process of teaching depended on the integration of face-to-face cooperative learning and CALL. In other words, the teaching of reading skills was through face-to-face CACL. This theme is shown in the expressions of the students below. S2: “Yes … we studied by using computer. At the beginning, the teacher asks us to open the laptops, aaa and we open the laptops. There is a specific website in which we open the skills. There are different reading skills. Every day we study aaaa one of the skills by using the computer. We study one or two passages related to the skill. Then we move to aaaa exercises which contain passages about the same skill. Thus, when we study the skill we answer as groups, and then we take exercises individually.” S3: “the teacher explains the skills, after the explanation of the skill, we move to the tutorials using the computer and study the tutorials as groups … we discuss the skill and understand it together. Then we move to do the exercises some of which are in groups and some of them are individually.” Teacher: “it is neither cooperative learning nor CALL, so it is integrated, which helps in giving the best lessons, aaa and aaa getting the best results.” Therefore, it is clear that face-to-face CACL utilizes CALL instructions to provide the students with different exercises. Also, cooperative learning helped the students to discuss the materials and negotiate the answers in groups. The comments of the students

Integration of Face-to-Face Instruction and CALL

349

show the integration of the two modes in face-to-face CACL, which is the method of teaching reading skills. c. STAD Strategy The method used in this study is face-to-face CACL according to the responses of the students above. Moreover, the STAD strategy is used during the implementation of faceto-face CACL to teach reading skills. This is according to the explanations of the student below. S4: “And, aaa Computer Assisted Cooperative Learning was a new skill for and a new learning method for me, to be in a group and aaa do several things such as doing many activities inside the classroom. And, aaa we start by doing tutorials then we having exercises in groups, then individual, then aaa there was … there were rewards given to us as a motivation and warming up activities at the beginning making us excited for the classes. It is a new learning method for me.” Accordingly, the strategy, which the teacher used in teaching reading skills with face-to-face CACL, is STAD. In this strategy, the teacher introduces the skill. Then the students work cooperatively on exercises. After that, each student works on exercises individually, and finally, the top team is rewarded by the teacher.

8 Involvement of Learning Objects This parameter investigates the interaction patterns, and what is required from the students and teachers to do inside the classroom. This parameter is important because interaction patterns shape the process of learning, especially with the advancement of technology. Also, this parameter is important to cater to the teachers’ need to introduce new teaching methods that involve face-to-face interaction and CALL instruction. Accordingly, the interaction among students with the computer and with the teacher was through the implementation of the principles of cooperative learning and through the roles of students and teachers according to the STAD strategy. This will be clarified in the discussion of the principles of cooperative learning and STAD strategy below. Also, the parameter level of integration is discussed at the end of this section. a. Cooperative Learning Principles The interaction among the students with the computer was through the implementation of the five principles of cooperative learning. The discussion of the implementation of these principles clarifies the interaction among students in groups with the computer and the individual learning with the computer. One of the roles of students was to achieve the five principles of cooperative learning. The first principle of cooperative learning is positive Interdependence, in which the students should understand the materials and help the other group members to understand the materials as well. Participants have achieved this principle as shown in the comment below.

350

A. A. Yassin et al.

S3: “yes, the level of the students in every group was gradual from the weak to the higher and so on. So, the weak student can get from the student who is stronger than him and everyone gets benefit from the others. The division of the students in the groups played a main role in the course.” The second principle of cooperative learning is promotive interaction, which means that the students should exchange ideas and materials in order to help each other. The first theme related to the promotive interaction principle is the exchange of ideas to answer the reading exercises. The theme of exchanging ideas among group members during the classes is shown in the students’ comments below. S1: “yes, this happened. It happened that some of my classmates gave me links and they told me that these links will give you more benefits. I also gave them some websites. One of my group members gave me a handout.” S4: “yeah, the skills when we are solving the questions in groups, in the group you will explain to your friends why did you chose this answer, why not the other answer. So, you are giving them the idea how did you choose this answer. Why this should be the correct, and why this is wrong?” The third principle is individual responsibility which refers to the importance of the individual to participate in the success of the group, especially through doing the individual exercises in the web-based CALL. This principle highlights the use of the students to do exercises in CALL individually, which is the individual exercise with CALL. Also, this principle highlights the responsibility of each student for the success of his team. This interaction pattern is vivid in the expression of the student below. S2: “No, we did not depend on the teacher. There was some dependency on the others when we start to answer questions as groups among the students. For example, one of the students might be excellent. He might have an excellent vocabulary background, so the other groupmates might depend on him. But, the other students aaaa quickly reach to self-confidence and work on the aaa individual exercises alone.” The fourth principle is social interaction which refers to the social skills as such skills are important to make cooperative learning more fruitful. The students showed that the language of interaction among the students was the English language as shown in the excerpts below. S2: “in terms of the language, we used to speak English all the time. We did not use Arabic language at all.” S3: “the communication was interesting because the students who participated in the course were almost in the same level. There were not big differences between the students as the level was the same. This makes the maximizes the benefit. The usage of English language in the communication was excellent and there was no difficulty."

Integration of Face-to-Face Instruction and CALL

351

The fifth principle of cooperative learning is group processing which refers to the reflection of the students on their learning of reading skills. This reflection can be within the group and with the whole class. The students showed that the reporter used to discuss the difficulties with his groups and the whole class, and they used to get feedback at the end of the course from the teacher and the other students. The theme of group processing is shown in the comments below. S3: “we used to get feedback from the teacher and the other students as well because they used to give comments during the activities. Every group discuss its difficulties so that we come to know the difficulties which all the groups faced.” Teacher: “Students got many types of feedback. The first aaa the first feedback was within the group. Aaaa the second one with the whole class … the whole class with the teacher and from the other groups as well.” The observation in all the sessions shows that the students used to work together and discuss ideas. Also, students used to help each other in answering group activities, and then students used to do individual activities to test their understanding of reading skills. Besides, students used to communicate in English during the whole class, and they used to do group reflection at the end of the class. Furthermore, they used to do whole class group processing as the students used to reflect at the end of all the sessions. These observations support the implementation of the five principles of cooperative learning in all the sessions of studying reading skills. b. STAD: Role of the Teacher and Students This section also discusses the interaction of learning objects, and the focus is on the duties of the students according to the instructions of the STAD strategy. The teacher also has different roles in teaching reading skills through face-to-face CACL. In terms of the students, they have four roles according to the STAD strategy. The students discussed their duties and the duties of the teacher according to their experience in studying reading skills through face-to-face CACL. In terms of the role of the teacher, the students highlighted different roles which were important for the successful integration of face-to-face CACL. These roles are carried out mainly to make the interaction among students and with the teacher effective during the study of reading skills, including teaching, supervision, and guidance. The excerpt below shows the teacher’s roles. S4: “starts … he explains … he is giving us the activities which make us more interactive with the others. So, we do not feel like to be isolated during doing the course. He gave us every day different warming up activities. We interact with people and those who does not like to participate with the others, he will participate because the teacher gives activities. After that he will start to explain to us what is the course, what is the main skill of the course, what is the objective, and what we will study. Then, he will aaa ask us to study in groups, to solve the group passages and the individual passages. After that, he will give us feedback. Why aaa this is correct? Why this is wrong? We will have a discussion with the lecturer and we

352

A. A. Yassin et al.

ask him even from out of the topic which we have studied, and he will answer to us.” In terms of the roles of students, there are four roles given to the students alternatively in every session. These roles are summarizer, recorder, facilitator, and reporter. The students’ comments on their roles show that they are aware of these four roles as it is shown in the excerpts below. S2: “the recorder writes the scores on the sheet which we used to get at the beginning of the class. We have also the summarizer who summarizes what we have studied during the whole day with the other group members. The reporter takes information from him, and the reporter is the one who stands in front of the classmates to speak about what we have studied depending on the explanation … depending on the lesson and the information of the summarizer and all the group.” S3: “one was the recorder who used to document the scores. The scores of every student and the whole group. There is also the summarizer who summarizes what we have studied. The reporter gets benefit from the ideas of the summarizer aaa to report them to the other groups after that and talk about them. And, aaa the facilitator who keeps the group focused and speak in English.” Teacher: “They [students] are divided into groups. So, one is a facilitator, one is a reporter, one is a summarizer, and the fourth one is a recorder.” According to observation inside the class, the teacher used to divide the roles of facilitator, recorder, reporter, and facilitator at the beginning of all the sessions. The teacher used to give the students these roles alternatively at the beginning of the class. c. Level of Learning Autonomy The level of learning autonomy refers to the students’ ability to understand their roles and responsibilities inside the classroom. This is also highlighted in the second parameter “model of integration”, as the main aim of the third descriptor “level of integration”. Hence, this section investigates how learning autonomy is important in the integration of face-to-face cooperative learning and CALL in order to make the process of learning smooth and more effective. The comments of the students show that they could improve autonomous learning when they studied reading skills through face-to-face CACL. S2: “at the beginning in the first class or the first lesson, it was difficult for the students to understand that they have to bring their laptops. After that the idea was very normal.” S3: “aaa it was perfect since the beginning because the instructions were clear and the students have understood the study process.”

9 Location The location parameter is very important for the success of learning, and students should study in an environment with which they are familiar and in which they feel comfortable. Accordingly, the location of the course was in an institute, which is close to all the

Integration of Face-to-Face Instruction and CALL

353

participants. In this regard, the students stated that the class was suitable for the study as shown in their expressions below. S1: “the classroom was suitable for our number. It is not crowded. We used to do the activities inside the classroom. Sometimes, we have activities that require us to move to another classroom. So, the hall was suitable, the light was suitable, and the ventilation was good.” S4: “yeah. It was suitable … it was cold … yeah … we have chairs, we have the projector, we have everything that we need, we have the lecturer.” Moreover, the students had to bring their laptops with them to the classroom to study. The students stated that it is normal and this might be better than studying in a computer lab because they could move easily to do group activities and individual activities as well. The students’ comments on this theme are shown below. S2: “actually, my point of view concerning bringing laptops … at the beginning in the first class or the first lesson, it was difficult for the students to understand that they have to bring their laptops. After that the idea was very normal because if we studied in a computer lab, the computer will be the basic tool without cooperative learning among the students.” S3: “I mean the communication will not be there between students if every student has a computer. Also, the communication will not be easy because every student will not be able to explain to the others and the students will not be close enough to each other. Every student will have his own space and I think that this is not helpful.” This is supported by observations as the students used to move freely and comfortably to do group activities, and then take their own space to do individual exercises. The class and the use of laptops help the students to form circles for group work, and they felt free to move to a remote place to work on the individual activities.

10 Discussion This study aimed at investigating the process of integrating face-to-face cooperative learning and CALL. The importance of these questions is related to the success of the design and the description of the process of integrating face-to-face cooperative learning and CALL. This study adapts Neumeier’s [13] model; however, integrating face-to-face cooperative learning and CALL requires modifying these parameters and their descriptors to suit face-to-face CACL. This first parameter is the mode, and it has three descriptors, namely focus on mode, distribution of modes, and choice of modes. According to Neumeier [13], there should be a leading mode that the students use when integrating CALL and face-to-face learning. The findings of the study showed that both modes are used together from the beginning until the end of the class, which made face-to-face CACL one method of teaching, and it is the leading mode. Moreover, the second descriptor is the distribution of modes which counts the time spent in each mode. However, in face-to-face CACL, it is difficult to

354

A. A. Yassin et al.

count the time spent in face-to-face cooperative learning and the time spent in CALL because both modes are used together. This raises a new point when using face-to-face CACL, which is that the two modes cannot be separated from each other, and face-to-face CACL is considered the leading mode. This is against the argument of Kerres (2001) as cited in Neumeier [13] that when there are face-to-face instruction and CALL, it is important to decide on the leading mode. Therefore, this study suggests that the first parameter in integrating face-to-face cooperative learning and CALL should be face-toface CACL which is also the leading mode. Also, the findings of this study show that the second and third descriptors, namely distribution of modes and choice of modes, cannot be discussed in face-to-face CACL instruction. In terms of the second parameter “level of integration”, face-to-face CACL also cancels the existence of the descriptor sequencing of individual modes, because the two modes are integrated as one method in the process of teaching and learning. In other words, the two modes are used simultaneously in teaching reading skills. This descriptor requires counting the time of each mode; nevertheless, in face-to-face CACL the time of each mode cannot be counted separately without being related to the activities of the other mode. According to [13], sequencing of individual modes should reduce the transitional distance that affects communication, and this transactional distance is the gap between the learners. The transactional distance is used to eliminate the feeling of isolation among the students during the activities [13]. Accordingly, face-to-face CACL minimized the transactional distance to a great extent since the students studied using face-to-face cooperative learning and CALL together during the whole class. In other words, although face-to-face CACL is considered the mode of teaching, the physical distance was reduced to a great extent during reading activities. In terms of the descriptor “level of integration” of the second parameter, it refers to the sub-modes that might be optional for the students. In this study, both modes, faceto-face interaction, and CALL, are obligatory. Neumeier [13] argued that designing the integration of modes is important to improve learning autonomy among the students. In face-to-face CACL, both modes, face-to-face cooperative learning and CALL, are one unit that makes the integration of both modes obligatory in all the sessions. According to the experience of the students, they could increase learning autonomy with time and depend on themselves in terms of understanding understand the process of learning reading skills. The findings of this descriptor support Neumeier [13] because face-to-face CACL activities improved the students’ learning autonomy. The third parameter, according to Neumeier [13], looks at whether the two modes are isolated or parallel. However, in the current study, the aim was to teach reading skills, but speaking was an important element of cooperative learning. Therefore, face-to-face CACL uses both cooperative learning and CALL simultaneously in all the classes to teach reading skills. In other words, the best description for the integration of faceto-face cooperative learning and CALL in this study is that the two modes are used simultaneously in face-to-face CACL. The findings of this study are different from those of Adair-Hauck et al. [34] because it taught reading only without the need for discussion among students. However, in the current study, the students needed to study cooperatively which helped the students to interact with each other and support the learning among students.

Integration of Face-to-Face Instruction and CALL

355

In terms of content, the findings of this study are different from those of Chenoweth et al. [35]. That is, these studies presented that the online materials are not enough which led them to copy other materials and use them for their study. However, in the current study the students stated that the content was enough, especially that the students exchanged materials with each other, and this study also showed that learning theories are important for learning activities [36]. Also, the content was suitable for the students in terms of their level of reading as it provides different passages with different topics and many new vocabularies. The findings of the online materials in this study show that CALL content and the features of CALL activities increase the benefit to the students [36]. The fourth parameter is the teaching method. According to Neumeier (2005), there are three sources, influencing the teaching method which are the self-access online material, the online tutor and the face-to-face teacher. However, in the current study, there is face-to-face cooperative learning among the students and a face-to-face teacher instead of an online teacher. Therefore, the teaching method is face-to-face CACL in which learning depends on face-to-face cooperative interaction, instead of an online tutor. Also, the content delivery was through the STAD strategy. This leads to an important point that the integration in face-to-face CACL does not isolate the students, and this method makes the role of the teacher essential inside the classroom. Also, the STAD strategy was effective with face-to-face CACL because the students could do the reading activities easily without facing any misunderstandings in the process of learning [37]. The fifth parameter is the involvement of learning subjects including students, teachers, and computers. According to Neumeier [13], this parameter investigates the interaction patterns, and what is required from the students and teachers to do inside the classroom. Accordingly, the findings of the study showed a wide variety of interactions in which cooperative learning and CALL are available. The interaction patterns include teacher to students with a computer, teacher to student with a computer, students to the student with a computer, and students individually with a computer. This supports the statement of Neumeier [13] that the interaction patterns are important so that the students can improve their level of learning autonomy. Another descriptor related to this parameter is that the roles of the students and teachers should be determined for the success of the learning process. This study supports the statement of Neumeier [13] because planning the roles of the students and teachers leads to the success of using face-to-face CACL. In this method, the students have to achieve the five principles of cooperative learning and the principles of STAD. The findings of the study showed that the students achieved the five principles of cooperative learning, namely positive interdependence, promotive interaction, personal responsibility, social interaction, and group processing, which made face-to-face CACL more effective [38]. Also, the students took the roles alternatively in all the classes as required for STAD namely recorder, facilitator, summarizer, and reporter. In addition, they followed the principles of STAD in all the classes which are introducing the skill, doing exercises as groups, doing exercises individually, and rewarding the top team in every class. The findings of the study showed that following a specific teaching method and assigning roles to students ensure the success of learning and also increase the level of learning autonomy which is in line with Neumeier [13]. Besides, the implementation of the five

356

A. A. Yassin et al.

principles of cooperative learning made the interaction among the students during the reading activities more dynamic and more productive. Therefore, this made the interaction among the students more effective, and the use of the technology was an additional advantage for teaching reading skills through face-to-face CACL. The sixth parameter, according to [13], is the location which is very important for the success of the integration of face-to-face cooperative learning and CALL, because the students should study in an environment with which they are familiar and in which they feel comfortable. Accordingly, the location of the course was an institute in a close place to all the participants, and the students brought their laptops with them to all the classes. The location was suitable for all the students and studying with laptops made cooperative learning easier. Hence, the integration in face-to-face CACL was smooth because the use of laptops made the students able to work in groups cooperatively and move to do individual exercises with the computer easily. Therefore, there is a need to adapt Neumeier’s [13] model to suit face-to-face CACL in the process of designing and describing face-to-face CACL courses, and integrating face-to-face cooperative learning and CALL. According to the findings of this study, Table 2 summarizes the findings of this question and the adaptation of Neumeier’s [13] model to suit face-to-face CACL. Table 2. Integrating f2f cooperative learning and CALL Parameter

Individual descriptors

1. Mode (face-to-face cooperative learning and CALL)

– Face-to-face CACL (the leading mode)

2. Model of integration

– Sequencing of individual modes: simultaneous – Level of integration: face to face cooperative learning and CALL are obligatory

3. Distribution of learning content and objectives and assignment of purpose

– Simultaneous

4. Language teaching method

– Cooperative learning principles – Cooperative learning strategy – CALL

5. Involvement of learning subjects (students and teachers)

– Interactional patterns: cooperative language learning activity – Variety of teacher and learner roles (cooperative learning principles and students’ role according to the need of cooperative learning strategy) – Level of autonomy

6. Location

Classroom, computer lab, institutional setting

Integration of Face-to-Face Instruction and CALL

357

11 Conclusion and Implications This study aimed to investigate the integration of face-to-face CACL in teaching reading skills. The integration focused on the two modes in CACL, namely face-to-face cooperative learning and CALL instruction. The study used the model of Neumeier [13] as the study theoretical framework. Although this model discussed collaborative learning, it did not account for cooperative learning, which requires the implementation of the five principles of cooperative learning. This is the main difference between collaborative and cooperative learning. The findings of the study have supported many parameters and descriptors described by Neumeier; however, the findings also showed that there are different elements that need to be adapted to suit the integration in face-to-face CACL. The main difference is that face-to-face CACL requires implementing face-to-face cooperative learning and CALL instruction together, which made it difficult to count the time of each mode. The simultaneous implementation of the two methods makes face-to-face CACL one mode, especially since they are used together in teaching from the beginning until the end of the class. Accordingly, face-to-face CACL is considered to be the leading mode with no descriptors to investigate the sequencing of modes. Besides, the simultaneous implementation of face-to-face CACL made the process of learning smooth and easier for the students, and this is one of the main factors that made teaching reading skills to students more effective. Finally, the findings of this study show that teaching students through face-to-face CACL requires clear and planned integration of the teaching activities. Doing so helps the students to improve learning autonomy in terms of understanding the process of learning through this method, which helps to make learning more effective. Moreover, the outcomes of the study show that the ADDIE model can be used along with the adapted model of intreating face-to-face cooperative learning and CALL in Table 2. That is, there are five phases of the ADDIE model, which aim mainly to design the materials and the activities for CALL lessons. However, the face-to-face CACL model adds the direction toward integrating face-to-face cooperative learning with CALL. Therefore, each model complements the other model. In other words, points are missing in the integration model in face-to-face CACL such as investigating the needs of the students, designing the materials according to their needs, and evaluating reading skills. However, these elements are available in the ADDIE model. Similarly, the ADDIE model is mainly used for CALL design, but there is no indication for integrating face-to-face cooperative learning and CALL. This makes CALL lack specific elements in terms of the integration in face-to-face CACL, including focusing on modes, distribution of modes, teaching method, and level of integration. These elements are available in the model of integration in face-to-face CACL. Accordingly, the ADDIE model and model of integration in faceto-face CACL have complementary processes for each other to have a complete picture for the design and implementation of face-to-face CACL. The model for the design and implementation of face-to-face CACL is shown in Fig. 2.

358

A. A. Yassin et al. Analysis

Design

Needs analysis

-CALL theories -Social Interdependence Theory -Cooperative learning strategy -Instructional design for the activities 1-leading mode 2-Model of integration 3-Distribution of learning content and objectives and assignment of purpose 4. Language teaching method 5-Involvement of learning subjects (students and teachers) 6-Location

Development

-Content development / Selection -CALL development and its features

Implementaon

-F2F CACL (CL principles, STAD, face to face interaction patterns)

Evaluaon

Pre-test and post-test; interviews

Fig. 2. Face-to-face CACL design and implementation model

References 1. Keengwe, J., Onchwari, G., Wachira, P.: Computer technology integration and student learning: barriers and promise. J. Sci. Educ. Technol. 17(6), 560–565 (2008) 2. Joshi, A., Pan, A., Murakami, M., Narayanan, S.: Role of computers in educating young children: US and Japanese teachers’ perspectives. Comput. Sch. 27(1), 5–19 (2010) 3. Al-Qaysi, N., Mohamad-Nordin, N., Al-Emran, M., Al-Sharafi, M.A.: Understanding the differences in students’ attitudes towards social media use: a case study from Oman. In: 2019 IEEE Student Conference on Research and Development (SCOReD), pp. 176–179. IEEE (2019) 4. Qasem, Y.A.M. et al.: A multi-analytical approach to predict the determinants of cloud computing adoption in higher education institutions. Appl. Sci. 10 (14), 4905 (2020). https://www. mdpi.com/2076-3417/10/14/4905 5. Delialioglu, O., Yildirim, Z.: Design and development of a technology enhanced hybrid instruction based on MOLTA model: Its effectiveness in comparison to traditional instruction. Comput. Educ. 51(1), 474–483 (2008) 6. Ghavifekr, S., Rosdy, W.A.W.: Teaching and learning with technology: effectiveness of ICT integration in schools. Int J Res Educ Sci 1(2), 175–191 (2015) 7. Cady, D., Terrell, S.R.: The effect of the integration of computing technology in a science curriculum on female students’ self-efficacy attitudes. J. Educ. Technol. Syst. 36(3), 277–286 (2008) 8. Harris, J., Mishra, P., Koehler, M.: Teachers’ technological pedagogical content knowledge and learning activity types: curriculum-based technology integration reframed. J. Res. Technol. Educ. 41(4), 393–416 (2009)

Integration of Face-to-Face Instruction and CALL

359

9. Razak, N.A., Yassin, A.A., Maasum, T.N.R.T.M.: Formalizing informal CALL in learning english language skills. In: Enhancements and Limitations to ICT-Based Informal Language Learning: Emerging Research and Opportunities, pp. 161–182. IGI Global (2020) 10. Al-Emran, M., Al-Maroof, R., Al-Sharafi, M.A., Arpaci, I.: What impacts learning with wearables? An integrated theoretical model. Interact. Learn. Environ. 1–21(2020) 11. Keengwe, J., Anyanwu, L.O.: Computer technology-infused learning enhancement. J. Sci. Educ. Technol. 16(5), 387–393 (2007) 12. Yassin, A.A., Razak, N.A., Maasum, T.N.R.T.M.: Innovation attributes of F2F computerassisted cooperative learning in teaching reading skills. Int. J. Web-Based Learn. Teach. Technol. (IJWLTT) 17(3), 1–17 (2021) 13. Neumeier, P.: A closer look at blended learning—parameters for designing a blended learning environment for language teaching and learning. ReCALL 17(2), 163–178 (2005) 14. Jones, J.G., Morales, C., Knezek, G.A.: 3-Dimensional online learning environments: examining attitudes toward information technology between students in Internet-based 3-dimensional and face-to-face classroom instruction. Educ. Media Int. 42(3), 219–236 (2005) 15. Saleh, S.M., Asi, Y.M., Hamed, K.M.: Effectiveness of integrating case studies in online and face-to-face instruction of pathophysiology: a comparative study. Adv. Physiol. Educ. 37(2), 201–206 (2013) 16. Johnson, R.T., Johnson, D.W., Stanne, M.B.: Comparison of computer-assisted cooperative, competitive, and individualistic learning. Am. Educ. Res. J. 23(3), 382–392 (1986) 17. King, A.: Verbal interaction and problem-solving within computer-assisted cooperative learning groups. J. Educ. Comput. Res. 5(1), 1–15 (1989) 18. Hooper, S.: Cooperative learning and computer-based instruction. Education Tech. Research Dev. 40(3), 21–38 (1992) 19. Neo, M.: Cooperative learning on the web: a group based, student centred learning experience in the Malaysian classroom. Australas. J. Educ. Technol. 20(2) (2004) 20. Sioofy, M., Ahangari, S.: The effect of computer assisted cooperative language learning on Iranian high school students’ language anxiety and reading comprehension. Int. J. Foreign Lang. Teach. Res. 1(3), 45–59 (2013) 21. Brush, T.A.: Embedding cooperative learning into the design of integrated learning systems: rationale and guidelines. Educ. Tech. Res. Dev. 46(3), 5–18 (1998) 22. Tondeur, J., Devos, G., Van Houtte, M., Van Braak, J., Valcke, M.: Understanding structural and cultural school characteristics in relation to educational change: the case of ICT integration. Educ. Stud. 35(2), 223–235 (2009) 23. Al Sharafi, M.A., Herzallah, F.A.T., Alajmi, Q., Mukhtar, M., Abdullah, R., Eleyan, D.: Information culture effect on E-commerce adoption in small and medium enterprises: a structural equation modeling approach. Int. J. Bus. Inf. Syst. 1(1) (2020). https://doi.org/10.1504/ijbis. 2020.10022194 24. Al-dheleai, Y.M., Tasir, Z., Al-Rahmi, W.M., Al-Sharafi, M.A., Mydin, A.: Modeling of students online social presence on social networking sites and academic performance. 2020, Online Soc. Presence SNS, Acad. Perform. 15(12), 16, 2020–06–26 (2020). https://doi.org/ 10.3991/ijet.v15i12.12599 25. Kurt, S.: Technology use in elementary education in Turkey: A case study. New Horizs. Educ. 58(1), 65–76 (2010) 26. Ertmer, P.A., Ottenbreit-Leftwich, A.T., Sadik, O., Sendurur, E., Sendurur, P.: Teacher beliefs and technology integration practices: a critical relationship. Comput. Educ. 59(2), 423–435 (2012) 27. Groff, J., Mouza, C.: A framework for addressing challenges to classroom technology use. AACE Rev. (Form. AACE J.) 16(1), 21–46 (2008)

360

A. A. Yassin et al.

28. Levin, T., Wadmany, R.: Teachers’ views on factors affecting effective integration of information technology in the classroom: developmental scenery. J. Technol. Teach. Educ. 16(2), 233–263 (2008) 29. Heba, E.-D., Nouby, A.: Effectiveness of a blended e-learning cooperative approach in an Egyptian teacher education programme. Comput. Educ. 51(3), 988–1006 (2008) 30. Persico, D., Pozzi, F., Sarti, L.: Monitoring collaborative activities in computer supported collaborative learning. Distance Educ. 31(1), 5–22 (2010) 31. AbuSeileek, A.F.: Cooperative vs. individual learning of oral skills in a CALL environment. Comput. Assist. Lang. Learn. 20(5), 493–514 (2007) 32. AbuSeileek, A.F.: The effect of computer-assisted cooperative learning methods and group size on the EFL learners’ achievement in communication skills. Comput. Educ. 58(1), 231– 239 (2012) 33. Creswell, J.W.: Research Designs: Qualitative, Quantitative, and Mixed Methods Approaches. Sage, Callifornia (2009) 34. Adair-Hauck, B., Willingham-McLain, L., Youngs, B.E.: Evaluating the integration of technology and second language learning. CALICO J. 269–306 (2000) 35. Chenoweth, N. A., Ushida, E., Murday, K.: Student learning in hybrid French and Spanish courses: An overview of language online. CALICO J. 115–146 (2006) 36. Yassin, A.A., Razak, N.A., Maasum, N.R.M.: Investigating the need for computer assisted cooperative learning to improve reading skills among Yemeni university EFL students: a needs analysis study. Int. J. Virtual Pers. Learn. Environ. (IJVPLE) 9(2), 15–31 (2019) 37. Yassin, A.A., Razak, N.A., Maasum, T.N.R.T.M.: Cooperative learning: general and theoretical background. Adv. Soc. Sci. Res. J. 5(8) (2018) 38. Roger, T., Johnson, D.W.: An overview of cooperative learning. Creat. Collab. Learn. 1–21 (1994)

Smart Techniques for Moroccan Students’ Orientation Morad Badrani1(B) , Adil Marouan1 , Nabil Kannouf2,3 and Abdelaziz Chetouani1

,

1 LaMAO Laboratory, ORAS Team, ENCG, Mohammed First University, Oujda, Morocco

{m.badrani,a.chetouani}@ump.ac.ma 2 LSA Laboratory, ENSA Alhoceima, Abdelmalek Essaâdi University, Tetouan, Morocco 3 LaMAO Laboratory, MSC Team, Mohammed First University, Nador, FP, Morocco

Abstract. Dropping out of school is a complex and multifactorial phenomenon that does not happen overnight. For example, to move from middle school to secondary school, the orientation is currently based only on the students’ choice without taking into account their points. The students are oriented towards branches that do not correspond to their capacities, in addition to the fact that the majority of parents, with a widespread mentality in society, wish to orient their children towards the scientific branch even if their capacities do not allow it, which finally leads to the school abandonment. This paper explores the possibility of making recommendations to students to help them orient themselves well, by analyzing real data collected from many institutions in the Moroccan city of Nador, the data are real and include the period of classes (2018–2019 and 2019–2020 and 2020–2021) of 7720 students of the third college year who have chosen their orientation for the following year, this data was trained on machine learning algorithms. Finally, to make the decision, we compare three classification algorithms (decision tree, naïve Bayes, and k-nearest-neighbor) by the precision, recall, and f1 score, our obtained result shows that the decision tree is the most suitable for this procedure. Keywords: Artificial intelligence · Student’s orientation · Decision tree · Naïve-bayes · KNN

1 Introduction The orientation of students is an important task in their carrier, because it’s very complicated to make choices and decide for the human being, this task (orientation of student) relies on the marks of the student and his desire, because if he like a branch, but he hasn’t the capacities for him, he will not be capable to keep up the program of that branch, which causes an emotion of unjustness, which can lead to the school dropout. Unfortunately, the middle school learners are still faced with this guidance issue, because when they are in college they cannot yet decide their orientation choices, even if their choice is between two branches: Sciences and letters; And according to statistics made public by the Ministry of National Education, more than 161,837 students of © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 361–366, 2023. https://doi.org/10.1007/978-3-031-20429-6_33

362

M. Badrani et al.

the middle school left education during 2020–2021, these statistics show that 62,244 students have dropped out after the third year of this cycle, which shows that the difficult transition of college students to the high school. And the studies that could facilitate this process are rarely. This document is structured as follows: related work are presented in Sect. 2. Section 3 presents the Background and Methods. Our implementation and results are presented in Sect. 4. And we end with some conclusions and future work.

2 Related Work There is little research on student orientation in middle schools, such as the study [1] using Big Data and machine learning to help the student in their orientation by comparing four classification algorithms: Naïve-Bayes, SVM, Neural Networks, and Random forest and conclude that Naïve-Bayes is more suitable for student’s guidance, another work [2] using machine learning under MapReduce with Hadoop to compare three algorithms: Naïve-Bayes, Knn (K-nearest-neighbors), and Neural Networks, by execution time and classification accuracy and also concluded that naïve-bayes is more useful for the student orientation, in another work [3] propose a framework for smart recommendation using data mining, to make good decisions For improved learning and guidance for students of the preparatory classes (CPGE)-Morocco, in another work of [4] propose an automated system for the guidance of Moroccan students in the university using the electronic portfolio (ePortfolio). Another of [5] compared some classification algorithms employing the Mapreduce model, and they deduce that the algorithms of classification launched on the models of MapReduce work agreeably in gigantic datasets, another work of [6] compare four classifier algorithms: ZeroR, J48, Naïve Bayes, and Bayes Net employing data of healthcare, and deduce that j48 more powerful than the other algorithm. In this work we based on the effects of dropping out of school in Morocco, caused by the problem of the wrong orientation of students, using real data to train and test three algorithms of supervised machine learning algorithms.

3 Background and Methods Artificial intelligence [7] corresponds to a set of technologies that allow simulating intelligence and automatically accomplishing tasks of perception, comprehension, and decisions these techniques particularly call upon the use of computer science, electronics, mathematics (especially statistics), neuroscience, and cognitive science. As Fig. 1 shows, artificial intelligence is present and applied in different activities: • In health: It has been used to predict disease, improve clinical workflow, and even identify patients’ risk for hospital-acquired infections. • In Education: there are also many problems in the field of education that can be solved by using artificial intelligence, such as: Automatic Exam Correction… • In the company: Robots used to perform complex or repetitive tasks… • In finance: is applied to the collection of personal data and provide financial advice. • In social media: While browsing the socials Medias, data is collected on the use to identify customer needs and which products may potentially interest each customer.

Smart Techniques for Moroccan Students’ Orientation

363

• In agriculture, artificial intelligence is used to improve the quality of production and reduce certain production costs. • Automobile: At this stage, many safety devices in modern vehicles use artificial intelligence that processes data from a multitude of sensors. • In Surveillance: Artificial intelligence is increasingly used in this field and it uses technologies such as facial recognition in security cases.

Fig. 1. Application of artificial intelligence

Artificial Intelligence is particularly present and applied in these different fields thanks to machine learning. Machine learning, abbreviated as ML, is a subdomain of AI (Artificial Intelligence) based on the programming of computers of different forms to be able to perform tasks and execute the commands assigned to them according to the data. at their disposal and their analysis while limiting the human intervention that directs it or removes it completely. Machine learning algorithms as Fig. 2 shows, can be categorized according to the mode of learning they use: supervised learning, unsupervised learning, and reinforcement learning. 3.1 Unsupervised Learning In unsupervised [8] learning, the output class is undefined. In this case, machine learning happens autonomously. These machine learning algorithms organize data into a set of clusters to describe its structure and make complex data appear simple and organized for easy analysis. 3.2 Reinforcement Learning Reinforcement learning [8] for an autonomous agent (robot, etc.), In this type of learning, the machine is made to interact with the environment around it and discover errors on its own, as this method is based on the idea of reward and punishment.

364

M. Badrani et al.

Fig. 2. Type of machine learning

3.3 Supervised Learning The machine is interested in input and response data already known. This processing of the known data will allow the algorithm to train itself and develop a predictive model adapted to the data to be processed later. To develop these predictive models, supervised learning [8] is based on classification techniques, the application of classification techniques can be found in different fields: digital, medical, banking. In this paper we will present: K-Nearest Neighbor (KNN), Naïve-Bayes, Decision Tree, this algorithms are the most classifiers commonly used to generate predictive models for student’s Orientation: Decision Tree (DT). [9] This algorithm is used in operations research problems and decision studies, and is an ideal tool for both economic and management operations. It is also used in machine learning and in most cases its results depend on predictions. And the final structure that results is like to that of a tree by containing a root and branches. Naïve-Bayes. [10] Is one of the most widely used algorithms in machine learning, mainly used for classification tasks. It is founded on the popular Bayes probability theorem: P(X|N) = P(N|X)P(X)/P(N)

(1)

This algorithm is used by researchers to recognize classes of objects on labeled data sets. Then, the algorithm is trained on unlabeled data. Once this cycle is complete, the researchers associate the labels and retrain. The KNN (k-nearest neighbors). [11] Is one of the most important and simple methods in the classification technology of supervised machine learning. Each sample can be represented by its nearest neighbor; The idea of this algorithm is: if the most similar k of most samples of a sample (that is, the nearest neighbors in the feature space) belong to a certain class, then the sample also belongs to this class. The kNN (k-NearestNeighbor) method associates only a very small number of adjacent samples when making the class decision.

Smart Techniques for Moroccan Students’ Orientation

365

4 Implementation and Results To guide the students to specializations (Science, Literature), we compare three classification algorithms (decision tree, Naïve Bayes, and k-nearest-neighbor) as per the precision, recall, and f1 score. Precision. [12] Allows us to know the number of positive predictions well done. For our problem, this would be the measure of the students we correctly identify as having the scientific branch among all students. Mathematically: Precision =

TrueP TrueP + FalseP

Recall. [13] Is the measure of our models correctly identifying true positives (in our case, the true positives are the students who actually chose the scientific branch); recall tells us how many we correctly identified as having chosen the scientific branch. Mathematically: Recall =

TrueP TrueP + FalseN

F1 score Precision-Recall values can be very useful in understanding the performance of a specific algorithm and also help produce requirements based results. But when it comes to comparing multiple algorithms trained on the same data, it becomes difficult to comprehend which algorithm is best suited for the data based on Precision-Recall values alone. Fortunately for us, a metric to combine precision and recall exists: the F1 Score [13], it allows a good evaluation of the performance of our model. It’s calculated: Score = 2.

Precision . Recall Precision + Recall

After training our data on machine learning algorithms. Finally, to make the decision on which method is the best for students’ guidance, and as Table 1 shows we found that the decision tree is the most suitable for this procedure. Our obtained results. Table 1. Performance comparaison of algorithms Algorithms

Precision (%)

Recall (%)

F1-score (%)

DT

82.44

84.31

83.3

Naïve-Bayes

68.96

51.67

83.25

Knn

85.34

81.42

58.3

5 Conclusion and Future Work In this paper, we compare the performance of three classification algorithms, to discover the suited algorithm for student guidance, using real student marks and their choices of

366

M. Badrani et al.

orientation, these three classification algorithms are: Decision Tree, Naïve Bayes, KNN, after the experience and as Table 1 show, that decision tree’s algorithm give the best performances and it’s better for students’ orientation. In our future work, we will try to apply our approaches to help Moroccan students of high school in their orientation.

References 1. Ouatik, F., et al.: Students’ orientation using machine learning and big data,pp.111–119, (2021) 2. Ouatik, F., Erritali, M., Jourhmane, M.: Student orientation using machine learning under MapReduce with Hadoop. J. Ubiquitous Syst. Pervas. Netw. 13(1), 21–26 (2020) 3. Mimis, M., El Hajji, M., Es-saady, Y., Oueld Guejdi, A., Douzi, H., Mammass, D.: A framework for smart academic guidance using educational data mining. Educ. Inf. Technol. 24(2), 1379–1393 (2018). https://doi.org/10.1007/s10639-018-9838-8 4. Mohammed, A., et al.: Implementation of a computerized system for the orientation of the Moroccan student in the university. Procedia—Soc. Behav. Sci. 182, 381–387 (2015) 5. Pakize, S.R., Gandomi, A.: Comparative study of classification algorithms based on MapReduce model. Int. J. Innov. Res. Adv. Eng. (IJIRAE) 1(7), 251–254 (2014) 6. Rghioui, A., Lloret, J., Oumnad, A.: Big data classification and internet of things in healthcare. Int. J. E-Health Med. Commun (IJEHMC) 11(2), 20–37 (2020) 7. Haton, J.P., Haton, M.C. : L’intelligence artificielle. Presses universitaires de France (1989) 8. Chinnamgari, S.K.: R Machine Learning Projects: Implement Supervised, Unsupervised, and Reinforcement Learning Techniques Using R 3.5. Packt Publishing Ltd (20190 9. Charbuty, B., Abdulazeez, A.: Classification based on decision tree algorithm for machine learning. J. Appl. Sci. Technol. Trends 2(01), 20–28 (2021) 10. Rish, I.: An empirical study of the Naïve Bayes classifier. In: IJCAI 2001 Workshop on Empirical Methods in Artificial Intelligence, vol. 3. No. 22 (2001) 11. Bijalwan, V., et al.: KNN based machine learning approach for text and document mining. Int. J. Database Theory Appl. 7(1), 61–70 (2014) 12. Vakili, M., Ghamsari, M., Rezaei, M.: Performance analysis and comparison of machine and deep learning algorithms for IoT data classification (2020). arXiv preprint arXiv:2001.09636 13. Powers, D.M.W.: Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation (2020). arXiv preprint arXiv:2010.16061

A General and Theoretical Background of English Academic Writing with Reference to Saudi EFL Context Asma Abdullah Alasbali1,2(B) , Harmi Izzuan Baharum1 , and Zuhana Mohamed Zin1 1 Department of Language Academy, Faculty of Social Sciences, Universiti Technology

Malaysia, Iskandar Puteri, Malaysia [email protected] 2 English Department, Arts and Sciences Faculty, King Khalid University, Abha, Saudi Arabia

Abstract. Academic writing is an essential skill for learners of English writing, especially learners of English as a second (ESL) or English as a foreign language (EFL). This highlights the importance of paying more attention to English academic writing, particularly in higher education institutions. Hence, the research provides a general and theoretical background about English academic writing with reference to teaching writing to EFL students in the context of Saudi higher education institution. The study convers different topics and sub-topics, including the definition of academic writing, teaching approaches of academic writing, the nature of English writing, and the factor that influence academic writing. This study also discusses the theories related to English academic writing, which are: the contrastive rhetoric theory, social constructionism theory, and connectivism theory. Other topics discussed in this study are related to the issues faced by EFL students of academic writing, such as the linguistic issues, cultural and psychological issues, instructors’ teaching practices, teaching methods, learning environment issues, and issues related to the use of technology in learning academic writing. Keywords: Academic writing · Theories · Issues of academic writing · EFL learners · Saudi higher education

1 Introduction English language proficiency, primarily academic writing skills, has become an indispensable requirement for scholars and university students to succeed in Higher Education (HE) [1, 2]. English academic writing, which is considered the main “publication tool” [3], is an imperative skill for both students and academics to demonstrate their knowledge in their chosen discipline [4, 5]. However, learning academic writing skills is challenging, especially for those who study English as a Foreign Language (EFL) [6–8]. Academic writing is challenging for Higher Education EFL students and for English native speakers who work in academia [9]. To write academically well, the students first need to learn how to speak the language in their minds [10]. They need to understand the language per se and its vocabularies sensibly. However, English academic insufficiency © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 367–381, 2023. https://doi.org/10.1007/978-3-031-20429-6_34

368

A. A. Alasbali et al.

has been cited in the literature and found to prohibit EFL students from making the most out of their courses because they fear that their English will not be good enough [11– 13] indicate that writing insufficiency can significantly influence EFL Higher Education students’ success and prevent them from engaging in classes, making them feel cautious, frustrated, and depressed. Furthermore, the barriers in English academic writing are found in the literature to be the most prevalent challenge facing EFL students in Higher Education settings [14–19]. Difficulties in English academic writing can arise from several factors, such as similarity to the first language (L1), in this case, is the Arabic language, understanding the grammar of vocabulary, sounds-spelling mismatches or level of productive vocabulary control [11, 16–20]. Previous studies also indicate other factors, such as the first language [21], prior academic culture [22] and cognitive comprehension [23], meta-cognition [24], and writing structure and style [25, 26]. These factors may play a fundamental role in shaping academic writing for most EFL students, and in many cases, struggling with the rhetorical and linguistic aspects of writing [27]. Studies also show that factors, such as students’ learning engagement, can significantly influence EFL students’ academic writing development. Learning engagement has been defined in many different ways. Some are very general, and some are very specific. In the broadest sense, it covers any engagement in the classrooms and out of classrooms, even out-of-school contexts [28]. Learning engagement is the time and effort students are afforded in activities related to learning and teaching interactions [28]. Reference [28] revealed that learning engagement has an important impact on language teaching and communicative competence. In addition, learning engagement has a crucial influence on self-efficacy and students’ expectations [29]. Engagement in the learning environment might be represented in different types: cognitive engagement, which refers to the extent to which the students interacted in the learning process. Behavioural engagement involves being involved in mutual revisions with peers to push the learning process forward. Affective engagement includes students’ attitudinal reactions and the emotional situation of the students that emerge and evolve as they receive, process, and interact with the academic learning process [29]. The interest in English academic writing development and the challenges faced by English as a Foreign Language (EFL) students have been ongoing in the literature [30, 31]. In most cases, Higher Education EFL students often find academic writing a daunting task, particularly if they are studying in different academic environment settings, as in the context of Saudi Higher Education [32]. In Saudi Arabia, numerous studies have shown/indicated that Higher Education teachers encounter many problems while teaching English as a foreign language to students [32–34, 35]. Past studies have also revealed that Saudi university students majoring in English are weak/incompetent in writing skills and make various academic writing mistakes [36, 37]. EFL Students do not consider learning the English language essential since it is not an official language in the country [38]. This proposed study intends to investigate the challenges (in terms of the difficulties and obstacles) Saudi Arabian EFL students face in academic writing in Higher Education. The study will explore diverse academic aspects of EFL students’ academic writing learning. These aspects encompass learning skills, pedagogical methods,

A General and Theoretical Background of English

369

students’ background, and technology-based environment that might have influenced EFL students’ learning of academic writing.

2 Academic Writing According to [39], academic writing success depends on how well writers understand what they write about and how they approach the writing task. Academic writing can be seen as an evaluation task that authors should prove their understanding of knowledge and demonstrate their disciplinary skills of thinking, interpreting, and presenting. Reference [40] define Academic writing as an orderly writing technique employed in colleges, universities, and scholarly papers. Writing, such as in newspapers, reports, and publications, students must compose their research dissertation in an academic style [40]. Students are supposed to present for courses and what teachers and scholarly researchers use to write various educational materials. 2.1 Definition of Academic Writing in Higher Education Academic writing is a recognised standard of the writing style used in schools, colleges and universities across academia. Academic writing has become an essential component of learning as learners and tutors recognise the importance of critical reasoning skills for students in Higher Education. Academic writing is a clear, compact, concentrated, structured type of writing supported by proof. The intention is to help the learner understand [40]. It has an orderly tone and way, but it is not complicated and does not demand lengthy sentences and complex vocabulary. However, few general academic writing aspects are consistent over every discipline [41]. 2.2 English Writing Teaching Approach in the Saudi Context There has been an unawareness of learning English as an adopted language; the investigation handled by [42] confirms the unawareness of teaching English via the utilisation of various learning strategies. Studies research carried out by professors in Saudi Arabia openly exhibit a few remarkably qualified teachers. Reference [43] suggested that these instructors have strong attention to multiple determinants concerning pupils in their schooling. Moreover, they hold that learners were not skillful at English stuff in Saudi Arabia, and the majority preferred the choice of the deductive strategy to enrich the education process. Reference [44] insists that tutoring English is a method correlated with fluency in speaking, the influence of education language rules, and what to explain in English teachings. 2.3 The Nature of English Writing Saudi Arabia is controlled by Islamic culture and beliefs. These extreme conservative customs have made the natives resist changes, especially embracing new lifeways, including studying English. Arabic is the dominant language used in numerous schools

370

A. A. Alasbali et al.

across the nation, and English is deemed a foreign language in Saudi Arabia. Reference [45] point out a general misunderstanding that learning English changes students’ capacity to learn Arabic more at younger ages. It may unfavourably impair Saudi culture and customs [46]. This misunderstanding has hugely affected English writing because students think that it will comprise their cultural beliefs. Public foundational schools do not offer English as a subject when their young students. The hesitance to provide learners with the English language at a very young age will eventually affect their academic writing at a later stage in their academic mission [45]. However, the state has recognised the essence of English in global information. It has made it a compulsory part of the curriculum, which might help learners hone their academic writing abilities. 2.4 The Complexity of Writing for EFL Students Arabic is the principal means of expression and has a distinct writing method and a notably rich literary record. It is simple to maintain that poor English abilities are the sole challenge of scholarly writing for EFL scholars. Nevertheless, the subject’s honesty is far more intricate than this [47]. While a lack of accuracy in the language is unquestionably a significant constituent, there are multiple other concerns that learners also face, from the incapacity to utilise knowledge in the needed circumstances to inadequate comprehension of academic writing basics. However, worrying is that these difficulties influence the learners’ academic attainment and their consequent work chances [48]. EFL learners have a tough time coming up with academic writings since they have to spend extra time deciding the appropriate words to build their sentences than an indigenous English talker. Therefore, plenty of issues fall within the notion of structure: vocabulary, grammar, and sentence structure. Reference [48] note that the usual way for an EFL learner to overwhelm this problem is by learning the language in more inclusive depth. A scholar will need to keep advancing their English and multiply their time to discover how to write it. Furthermore, according to [48], insignificant aspects were also unveiled concerning difficulties in learners’ writing, namely paragraph organisation, dictions, and dictionary misspelling. Considering its complexity in finishing writing assignments, the scholars required more intense guidance from the lecturer, such as continuous consultation concerning the lecturer’s feedback in their composition draft. 2.5 Differences in Writing Between English and Arabic Language Differences between Arabic and English are pretty plain. Arabic holds 28 consonants (English 24) and eight vowels (English 22). Concise vowels are unnecessary in Arabic and certainly do not arise in writing [49]. Arabic writings are constituted and written from right to left, utilising cursive handwriting, which is equated to English, transcribed by applying Latin script and recited from left to right. There is no differentiation among lower and upper cases, and punctuation laws are more insecure than English. Unsurprisingly, these significant disparities between the Arabic and English writing modes make Arab scholars’ unusual difficulties. They generally require much more time studying or writing than their English-learning competitors from the Indo-European linguistic communities.

A General and Theoretical Background of English

371

Reference [49] further note that Arabic does not have the verb to be present tense, and no assistant does. Moreover, there is a singular present tense in Arabic instead of English, which are elementary and continuous forms. These variations result in errors when students are writing their academic assignments in English. 2.6 Possible Factors Influencing English Academic Writing 2.6.1 Similarity to L1 Vocabulary plays aids in promoting second language acquisition for L2 scholars in various colleges and university undergraduates. Undergraduates, in distinct, are explained and concentrated on progressing their English to compose better scholarly compositions [50]. Specifically, L2 scholars readily get discouraged from tutorials; hence, instead of merely reading, proposing an inventive approach in education would intensify more vocabularies in a more practical strategy by utilising vocabulary plays. 2.6.2 Sounds-Spelling Mismatches Spelling is just the putting together of some letters to produce words. Sound-spelling matches influence English academic writing since the very letter does not perpetually describe the same character; the very sound is not perpetually represented by the same letter. Additionally, some letters are not vocalised at all [51]. People in their speaking, at some points, articulate sounds in some places where there is no letter. Based on the abovediscussed issues, writing poses a significant danger of having numerous misspelled and unfinished sentences. 2.6.3 Understanding the Grammar of Vocabulary English is considered a general language and is extensively applied, and vocabulary is a prime aspect of English learning. L2 scholars observe learning English strenuously because it is not their local language. When individuals do not speak excellent English, they are seen as more inferior in the human hierarchy [50]. Notwithstanding such circumstances, there are other learning methods to comprehend everything, such as vocabulary plays. This helps develop vocabulary knowledge, but it also assists with thoughtfulness enhancement. 2.6.4 Level of Productive Vocabulary Control Vocabulary knowledge is a benchmark of learning in writing, reading, listening, and speaking. Knowledge of vocabulary managed by a learner affects the nature of their writing. It is granted by [51] that the loss of vocabulary control will alter the quality of writing compared to other factors such as attitude, motivation, and preparation for the exam. Therefore, learners who study a second language should dominate the vocabulary of quantity to know and express meaning in writing [51].

372

A. A. Alasbali et al.

3 Theoretical Background of English EFL Writing Second language writing comprehension is highly intricate, dynamic, multicomponential, and composite because it requires different strategic, rhetoric, and linguistic needs, significant facets, and learning issues. Reference [13] clarifies that the writing group can be significant determining factors towards learning English as a second language. Learners may opt to adopt either the product group or the process group. There is in place varying linguistic competencies and insights regarding language. Reference [52] refer to linguistic competency as the capability to use a particular language. In such a case, use could mean not only writing but reading too. For an individual to be considered fluent, one must operate the language accurately and fluently. These are also influenced by native language background, genre, pragmatics, metalinguistic knowledge, metacognition, strategy use, motivation and other contextual factors. In the theoretical background of English EFL writing, it is believed that a greater degree of what writers do is similar when they comprehend such syntax and structure. In their native language, they merely transliterate what they do [52]. However, writing in the second language often becomes less successful than the native language because it is often riddled with many errors. Many advanced learners with other native languages other than English would be compelled to learn English through ‘children’s movies to understand all about the language and its cultural entities, including idiomatic expressions. This might be more effective for the spoken language but not for academic writing. Therefore, reading scientific research articles is better because learners acquire the English language and the vernacular of the field the student needs. However, reading articles is not sufficient either. Practice, followed by reliable and acceptable feedback, is necessary. Writing is generally a complex endeavour, and the more researchers investigate the routes and commitment for writing, the more intricate the associations between cognition and generating writing appears. Theories about writing continuously change from emphasising mechanics and form to stressing creativity and sociability [29]. This subheading analyzes major writing theories, highlighting how they develop into practice and the strategies supporting each theory. 3.1 Contrastive Rhetoric Theory Contrastive rhetoric theory revealed with Robert Kaplan’s 1966 study, in which Kaplan made the pronouncement that “each language and each culture has a paragraph order unique to itself, and that part of the learning of a particular language is the mastery of its logical system” [53]. His study had provided insights into problems EFL students encountered in learning English -in general, and learning academic writing specifically [54]. In 1966, Robert Kaplan published his seminal paper “Cultural Thought Patterns in Intercultural Education,” which marked the birth of contrastive rhetoric. Influenced by the weak version of the Sapir-Whorf hypothesis that “language influences thought,” Kaplan put forward his idea based on three main assumptions: Speech and writing are cultural phenomena; Each language has a set of writing conventions unique to it; Linguistic and rhetorical conventions of a first language interfere with writing in a second language. The

A General and Theoretical Background of English

373

impact of contrastive rhetoric studies on teaching EFL academic writing and practice is always remarkable, and the significance cannot be disregarded [54]. Contrastive Rhetoric Theory describes how people’s second language writing can be influenced by their first language and culture [54]. L1 and culture are shaping &describing their interaction with the social culture elements around them. The main aspects of this theory are L1, Culture, Communication, Persuasion, and activity [55]. Contrastive Rhetoric Theory describes how people’s second language writing can be influenced by their first language and culture. The theory posits that the distal processes shape people and describe their interactions with the social-cultural elements nearest to them. Social relationships influence learners they encounter daily [55]. Numerous Scholastic composing exists, and they are very significant in that the actual worth is in work embedded in the writing. Genuinely scholarly work is the product of somebody’s hard work. Numerous handbooks were the outcome of several decades of in-depth analysis and investigation [44]. If students can put that sort of energy into a manuscript, they can be assured that the script might serve its intended purpose. 3.2 Social Constructionism Theory Compared to the cognitive process theory, which highlights the mental processes of composition majorly, sociocultural theory stresses motivation, impact, and social effects as writing components. The theory came up from the works of Vygotsky, who argued that infants acquire knowledge about their environment from interactions with others or those that have much more master of the subject matter. From such a background, researchers came up with the social constructionism theory to stress the importance of language to sociocultural interactions [56]. Such a focus on language allowed future researchers to connect the sociocultural model to writing instructions and growth. Generally, composing down communication has been of great importance; however, with the significance of sociocultural theory, today, writing is backed up by a cooperative, social activity, which beginner writers can learn more from more experienced ones. These theories came up as writing became a learning tool [56]. Writing is thought of as a discipline that students have to master. However, with sociocultural theory, writing is seen as an instrument of learning [57]. Writing is perceived as a discipline that students should master, such as writing to persuade others or writing towards informing. The sociocultural theory posits that writing expands beyond the classroom and provides context to comprise prior knowledge, understanding of knowledge, multiple genres, motivations, and potential role of learning environment engagement. In social constructivism, knowledge is individually constructed and socially mediated through the engagement of community members (the learners) with others in various social activities. The learners, thus, internalise the results that are activated by working together, and the learning process takes place. The main aspects in this theory are learner’s time, learner’s interacting, learner’s background, and knowledge [55]. In other words, this theory is essential to expose the identity of the EFL students who are learning academic writing at the university with consideration given to what “selves” they exhibit in their writing and how much their learning background and experiences shaped those selves in a targeted writing course [58]. According to [58], the students were overwhelmed by the assumed teachers’ expectations and their lack of experience required

374

A. A. Alasbali et al.

to meet those expectations. These missing aspects might be significant aspects that will be viewed through the lens of the theory to get more understandable interpretations. 3.3 Connectivism Theory Connectivism is a theory of learning founded by George Siemens, and later Downs joined him in developing the theory following the changes in the digital era [59, 60]. The theory explains the way that knowledge is conveyed across a network of connections and communications among the users of technology via four key principles for learning: autonomy, connectedness, diversity, and openness [60]. The theory is relevant to this study because it supports learning aspects such as learner’ role, learning environment and learning interactivity through technology related to using technology-based environments [61]. The theory informs the study concerning the role that technology-based learning environment engagement might act as a mediator element in the experience of students’ academic writing learning.

4 Issues Faced by EFL Students in Academic Writing EFL students have difficulty writing academic essays because they have to employ more time and energy to find the correct information to form their sentences than a native English speaker. This assignment alone can create disappointments with the minimal English vocabulary they usually have. Writing could be a difficult skill to be studied or taught due to the reality that it is not a mild cognitive activity; instead, it is considered to be a complex mental production that needs “careful thought, discipline, and concentration” [62]. According to [63], EFL students face other several challenges. These challenges mainly involve differentiating between written and spoken words and expressions, reviewing morphology, including subject-verb agreement, and connecting two or more sentences to make a logical paragraph. Reference [63] still argues that when learners complain about how challenging it is to write in a second language, they argue not mainly about the difficulty of finding the right words and using the correct morphology but also about the complexity of obtaining and representing approaches in a new language. 4.1 Linguistic Issues Linguistics academic writing problems emerge as the principal constraints for EFL learners to write an excellent English essay in academic writing. In the tutoring of writing, the flow of activities typically includes (1) familiarisation: students learn grammar and vocabulary, normally through a text; (2) controlled writing: learners emulate given guides; (3) supervised writing: students handle model texts; and (4) free writing: students apply the models. They have developed to write a letter, a paragraph, a composition, and writing is not just a mechanism for communication. Still, additionally, it serves as a medium of learning, thinking, and planning expertise or ideas. Besides, writing is a complicated exercise requiring some stages of composition task completion. Unquestionably, this skill, especially in an EFL setting, has been considered one of the most

A General and Theoretical Background of English

375

complex skills for students to master [64, 65]. The difficulty is due to the need to generate and design ideas using the proper vocabulary, sentence, and paragraph organisation choices and turn such conceptions into arable text along with a selective rhetoric pattern. Linguistic concerns have been the main obstacle challenging EFL learners in their academic writing in Saudi Arabia. 4.2 Cultural and Psychological Issues Writing in a second language feigns several obstacles for most English as a foreign language, EFL learners, specifically in Saudi Arabia. Saudi Arabia learners know writing in the traditional techniques where it abides by laws and a singular structure. They are not capable of resolving various rhetorical approaches. Reference [66] likewise challenge Arab learners to formulate words and sentences in their L1 and then transpose them into the L2. Reference [66] argue that topic familiarity and cultural propriety are essential factors inducing negative L1 transfer into L2 writing. 4.3 Instructions’ Strategies and Practices Issues It is essential for Saudi Arabia that educational institutions need to aid ‘student’s growth of creativity as learning involves thinking. Many school-going people decline to apply comprehensive factors to attain solutions to obstacles. Moreover, most secondary school and college students fail to utilise the higher-order thinking skills they need for their eventual achievement in post-secondary learning or their profession in later life. Finally, enhanced thought does not just unexpectedly arise as a result of learning. Reference [67] asserts that EFL learners studying at Saudi Arabia colleges undergo various languagerelated information literacy challenges. Emirati lives in an Arabic-speaking community; though, they are involved in an English-speaking academic community. The research presented here examines the challenges these learners face to understand their skills of information literacy thoroughly. A gap exists in library and information science study investigating EFL students’ experiences of information knowledge using phenomenography, a relational method. 4.4 Teaching and Learning Methods Teaching and learning styles issues are several common concerns affecting EFL students in Saudi Arabia [68]. The absolute concrete-sequential learning style is compatible with the commonly preferred teaching approach (hands-on training). The high portion of EFL learners whose answers suggested mixed learning styles insinuates that they can support various teaching methods. It is urged that teachers acknowledge the learners’ different learning styles and employ diverse teaching techniques to develop an optimal learning setting for them and improve their academic writing abilities [68]. 4.5 Learning Environment Issues Writing is an essential language skill and a fundamental productive exercise, particularly for foreign English (EFL) language students [69]. Because of the intricacy of

376

A. A. Alasbali et al.

writing and its diverse specification, both beginner and progressed EFL understudies typically have negative contemplations towards writing, commonly known as writing doubt or composing uneasiness. In favoured learning, settings may acquire two benefits. Initially, they could get writing assistance during the modification and shifting steps in constructing reactions and comments from their allies and teachers [69]. Besides, they would have the option to acquire data from the Internet, disperse that information, and compose capability through close and personal discussion. 4.6 Adaption of Technology in the Higher Education Learning Environment Technology adoption has become integral in today’s educational institutions, and it has influenced all facets of life at the social, organisational and individual levels. In many developed countries, technology-based learning environments have been incorporated within the national policies and educational strategies. Not only has the educational sector utilised the technology in teaching and learning, but many governmental sectors have also adopted it for the provision of services. The public and private sectors rely heavily on technology [70]. Hence, educational providers need to meet the changes in the market to prepare students to pursue their careers with the right skills and mindset. Many academic institutions have also adopted technology around the globe in teaching, assessment, curriculum development and professional development [71, 72]. The technology enables educators to plan and deliver their materials cost-effectively. It has become an integral element in students’ higher education learning cycle and a validated teaching tool for teachers and educational providers. 4.7 Technology-Based Environment In the 21st century, technology has been integrated into almost all aspects of human life, including education. Skinner acknowledged the design of teaching machines designed to assist correct learning [73]. New communities will be virtual as well as physical. A technology-based environment means utilizing the students to study independently, with freedom in the environment of technology [74, 75]. In this case, students are required to use technology to get information and knowledge they did not get from their teacher. Teachers should take advantage of the technology and excitement that students bring with them to learn from them and connect technology to their teaching [76]. Technology-based environments can enhance contrastive sides of language learning, such as student motivation, vocabulary acquisition, writing competence, and the use of authentic language [77, 76]. 4.8 Academic Writing and Technology-Based Environment Learning to write is fundamental to becoming literate. Academic writing is an important skill to master, even more for undergraduate students who should write academically. It is difficult to make students write academically, and a complex process is needed [78]. Technology has greatly improved students’ written communication skills [79]. There is increasing interaction across online platforms such as Twitter, Facebook, Instagram, and

A General and Theoretical Background of English

377

Whatsapp, as these applications enable students to communicate in written form on the one hand. Students also conveniently use digital applications such as writing and editing software on the other hand, as well as automated grammar-checking tools to improve their written communication skills [80]. While technology has played a prominent role in freeing students time to simplify tedious tasks within technology-based learning environments, digital platforms such as citation generators allow students to adopt writing formats and reference styles like APA or Harvard. The writing process has become faster; thus, students can save the time needed to improve their academic writing skills and do their homework more comprehensively. In addition, editing and proofreading apps such as Grammarly and WhiteSmoke provide automated writing services by making corrections to identify errors in real-time to speed up revision before assignments are submitted [81]. It can be argued that the adoption of digital tools in academic writing within technically supported learning environments led to the improvement of students’ writing skills by increasing creativity and speed of achievement. By engaging in technologyenabled learning environments, students discover new information online and access others’ ideas about the topics they want to write about. This opportunity also enables them to be creative and innovative in their writing through collaborative work by sharing knowledge [82]. Technology-based learning platforms encourage communication, inspiration, online writing practice and encourage collaboration among students. By way of illustration, engaging in learning via digital social media platforms allows students to complete group academic writing projects and develop teamwork skills. Engaging in a technology-based learning environment makes academic writing assignments more meaningful as students share various information necessary for their writing skills [82]. It is also possible through discussion forums where EFL students deal with the associated challenges by sharing skills and learning by working with others. Technology-based academic learning environments may also improve performance as students become more attentive and thoughtful about their written information [83]. These environments, and effective involvement in them, enable students with lower writing proficiency to interact with better and more efficient students so that they tend to achieve better achievement [84]. Creating academic written content promotes accuracy and relevance as it develops students’ attention to detail to improve their overall writing skills. Engaging in technology-based learning and its applications such as grammar checkers, plagiarism detectors, urban dictionaries, and citation generators improve students’ writing skills for language proficiency [83, 84].

5 Conclusion This research provided a general and theoretical background about academic writing. It has highlighted that academic writing is an essential skill for EFL/ESL learners. This study also highlighted that to master writing skill is the responsibility of both the teachers and the students, and this requires paying attention to the linguistic, cultural, and pedagogical issues in order to support the students’ learning and enhance their writing skills. Besides, even though technology is a good tool for students to improve

378

A. A. Alasbali et al.

their writing skills, teachers and students need to adapt to the right tools and get support through collaboration or using technology tools to check their spelling and grammar mistakes. All to all, good learning of writing skills started from the classroom, and the teacher play a main role in this regard.

References 1. Dearden, J.: The changing roles of EMI academics and English language specialists. In: Key Issues in English for Specific Purposes in Higher Education, pp. 323–338. Springer (2018) 2. Khalili Sabet, M., Salamatbakhsh, M.: Predictive validity of English proficiency levels for academic achievement. Iran. J. Engl. Acad. Purp. 9(3), 32–45 (2020) 3. Buckingham, L.: Development of English academic writing competence by Turkish scholars. Int. J. Dr. Stud. 3 (2008) 4. Gheyathaldin, L., Shishakly, R.: Impact of English language proficiency on the academic performance of business students. Asian ESP J. 16 (21), 245–272 (2020) 5. Oducado, R.M., Sotelo, M., Ramirez, L.M., Habaña, M., Belo-Delariarte, R.G.:English language proficiency and its relationship with academic performance and the nurse licensure examination. Nurse Media J. Nurs. 10, 46–56 (2020) 6. Ceylan, N.O.: Student perceptions of difficulties in second language writing. J Lang Linguist Stud 15(1), 151–157 (2019) 7. Mukhroji, M.: Exploring the academic writing needs to improve the academic literacy of the Indonesian EFL undergraduate and masters Students. Int J Innov, Creat Chang 10(10), 252–269 (2020) 8. Noori, A.: An investigation of Afghan undergraduate English major students’ academic writing difficulties. J. Foreign Lang. Teach. Learn. 5 (2) (2020) 9. Zhao, J.: Native speaker advantage in academic writing? Conjunctive realizations in EAP writing by four groups of writers. Ampersand 4, 47–57 (2017) 10. Durga, V.S.S., Rao, C.S.: Developing students’ writing skills in English-A process approach. J. Res. Sch. Prof. Engl. Lang. Teach. 6(2), 1–5 (2018) 11. Mayan, O.H.: Understanding students experiences with E-learning 2.0 in Australian higher education: a grounded theory study. Online Submission (2019) 12. Fernsten, L.A.: Writer identity and ESL learners. J. Adolesc. Health. 52(1), 44–52 (2008) 13. McKinley,J.: Critical argument and writer identity: social constructivism as a theoretical framework for EFL academic writing. Critic. Inquiry Lang. Stud. 12(3), 184–207 (2015) 14. Li, H., Fox, R.F., Almarza, D.J.: Strangers in stranger lands: Language, learning, culture. Int. J. Progress. Educ. 3(1), 1–44 (2007) 15. Littlemore, J., Chen, P.T., Koester, A., Barnden, J.: Difficulties in metaphor comprehension faced by international students whose first language is not English. Appl. Linguis. 32(4), 408–429 (2011) 16. Marlina, R.: “I don’t talk or I decide not to talk? Is it my culture?”—International students’ experiences of tutorial participation. Int. J. Educ. Res. 48(4), 235–244 (2009) 17. Preisler, B., Klitgård, I., Fabricius, A.: Language and learning in the international university, From English Uniformity to Diversity and Hybridity. Multilingual Matters, Bristol (2011) 18. Qiu, W.: Language adjustment of international students in the US: a social network analysis on the effects of language resources, language norm and technology. Michigan State University (2011) 19. Ramachandran, N.T.: Enhancing international students’ experiences: an imperative agenda for universities in the UK. J. Res. Int. Educ. 10(2), 201–220 (2011)

A General and Theoretical Background of English

379

20. Rohmatillah, R.: A study on students’ difficulties in learning vocabulary. English Education: jurnal tadris bahasa Inggris 6(1), 75–93 (2014) 21. Almuhaysh, M.A.: Reiteration relations in Arab ESL student writing: effect of first language transfer. Engl. Lang. Teach. 13(7), 8–18 (2020) 22. Rahmat, N.H.: Problems with rhetorical problems among academic writers. Am. J. Soc. Sci. Hum. 4(4), 506–515 (2019) 23. Xie, Q.: Diagnosing university students’ academic writing in English: Is cognitive diagnostic modelling the way forward? Educ. Psychol. 37(1), 26–47 (2017) 24. Anderson, T.: The Theory and Practice of Online Learning. Athabasca University Press (2008) 25. Nunan, D.: Second Language Teaching and Learning. ERIC (1999) 26. Setyowati, L., Sukmawan, S.: EFL Indonesian students’ attitude toward writing in English. Arab. World Engl. J. (AWEJ) 7 (4) (2016) 27. Javadi-Safa, A.: A brief overview of key issues in second language writing teaching and research. Int. J. Educ. Lit. Stud. 6(2), 12–25 (2018) 28. Suherdi, D.: SMSLEFA: an alternative synergistic multilayered analysis of students’ learning engagement in EFL context. Indones. J. Appl. Linguist. 8(1), 11–20 (2018) 29. Hyland, K.: Second Language Writing. Cambridge University Press (2019) 30. Ravari, Z.K., Tan, K.E.: A qualitative investigation of strategies and experiences of non-native students writing master’s theses. J. Appl. Res. High. Educ. (2019) 31. Roman, K.M.L.: Intergration of Academic Writing in Adult Classes in English as a Second Language: A Single-Case Study. Northcentral University (2020) 32. Al-Khairy, M.A.:Saudi English-major undergraduates’ academic writing problems: a Taif University perspective. English Lang. Teach. 6(6), 1–12 (2013) 33. Al-Sobhi, B.M.S., Preece,A.S.: Teaching English speaking skills to the Arab students in the Saudi school in Kuala Lumpur: problems and solutions. Int. J. Educ. Liter. Stud. 6(1), 1–11 (2018) 34. Alhamad, R.: Challenges and induction needs of novice English as a foreign language teachers in Saudi Arabia. Int. J. Educ. Lit. Stud. 6(1), 50–63 (2018) 35. Ashraf, T.A.: Teaching English as a foreign language in Saudi Arabia: struggles and strategies. Int. J. Engl. Lang. Educ. 6(1), 133–154 (2018) 36. Javid, C.Z., Umer, M.: Saudi EFL learners’ writing problems: a move towards solution. Proceeding of the Global Summit on Education GSE 2014, 164–180 (2014) 37. Obeid, R.: Second language writing and assessment: voices from within the Saudi EFL Context. Engl. Lang. Teach. 10(6), 174–181 (2017) 38. Qayoom, N., Saleem, M.: A study of Saudi advanced academic writing students’ perceptions of research essays, and gaps in their knowledge. Int. J. Learn., Teach. Educ. Res. 19(6), 223–237 (2020) 39. Irvin, L.L.: What is “Academic” writing? Writing spaces: Readings on writing 1, 3–17 (2010) 40. Zemach, D.E., Rumisek, L.: Academic Writing. Macmillan (2016) 41. McCarthy, L.P.: A stranger in strange lands: A college student writing across the curriculum. In: Landmark Essays. Routledge, pp. 125–155 (2020) 42. Khusniddinovich, S.M.: The importance of grammar in teaching the foreign language. Doctieni nayki i obpazovani 2 (7) (29), 64–66 (2018) 43. Minasyan, E., Midova, V.: Grammar as one of key competences of language acquisition. Russian linguist. Bull. 3(7), 107–108 (2016) 44. Nafisah Endahati, N.: The Implementation of Inductive Approach in Pre-Intermediate Structure Subject to English Education Study Program Students PGRI. University of Yogyakarta (2015) 45. Mahboob, A., Elyas, T.: English in the kingdom of Saudi Arabia. World Englishes 33(1), 128–142 (2014)

380

A. A. Alasbali et al.

46. Alrabai, F.: Learning English in Saudi Arabia. In: English as a Foreign Language in Saudi Arabia, pp. 102–119. Routledge (2018) 47. Alharbi, M.A.: Saudi Arabia EFL university students’ voice on challenges and solution in learning academic writing. Indones. J. Appl. Linguist. 8(3), 576–587 (2019) 48. Ariyanti, A., Fitriana, R.: EFL students’ difficulties and needs in essay writing. In: International Conference on Teacher Training and Education 2017 (ICTTE 2017), pp. 32–42. Atlantis Press (2017) 49. Betti, M.J., Ulaiwi, W.A.: Stress in English and Arabic: a contrastive study. Engl. Lang. Lit. Stud. 8(1), 83–91 (2018) 50. Coxhead, A.: Academic vocabulary, writing and English for academic purposes: perspectives from second language learners. RELC J. 43(1), 137–145 (2012) 51. Milton, J.: Measuring the Contribution of Vocabulary Knowledge to Proficiency in the Four Skills. Bardel, C., Lindqvist, C., Laufer, B. (eds.), vol. 2, pp. 57–78 (2013) 52. Scontras, G., Fuchs, Z., Polinsky, M.: Heritage language and linguistic theory. Front. Psychol. 1545 (20150 53. Kaplan, R.B.: Cultural thought patterns in inter-cultural education. Lang. Learn. 16(1–2), 1–20 (1966) 54. jiangli, S.: Contrastive rhetoric and teaching of ESL writing. Educ. Q. Rev. 2 (2) (2019) 55. Connor, U., Ene, E., Traversa, A.: Intercultural rhetoric. In: The Routledge Handbook of English for Academic Purposes, pp. 294–306. Routledge (2016) 56. Pathan, H., Memon, R.A., Memon, S., Khoso, A.R., Bux, I.: A critical review of Vygotsky’s socio-cultural theory in second language acquisition. Int. J. Engl. Linguist. 8(4), 232 (2018) 57. Obeidat, M.M., Alomari, M.D.A.: The effect of inductive and deductive teaching on EFL undergraduates achievement in grammar at the hashemite University in Jordan. Int. J. High. Educ. 9(2), 280–288 (2020) 58. McKinley, J.: Identity construction in learning English academic writing in a Japanese University. J. AsiaTEFL 14(2), 228–243 (2017) 59. Siemens, G.: Connectivism: a learning theory for the digital age. Int. J. Instruct. Technol. Distan. Learn. Retrieved from http://www.idtl.org/Journal/Jam_05/article01.html,2005 60. Downes, S.: Places to go: connectivism & connective knowledge. Innov.: J. Online Educ. 5(1), 6 (2008) 61. Siemens, G.: Connectivism. Foundations of learning and instructional design technology (2017) 62. Grami, G.M.A.: The effects of integrating peer feedback into university-level ESL writing curriculum: a comparative study in a Saudi context. Newcastle University (2010) 63. Alfaki, I.M.: University students’ English writing problems: diagnosis and remedy. Int. J. Engl. Lang. Teach. 3(3), 40–52 (2015) 64. Abdulkareem, M.N.: An investigation study of academic writing problems faced by Arab postgraduate students at Universiti Teknologi Malaysia (UTM). Theory and Practice in Language Studies 3(9), 1552 (2013) 65. Razak, N.A., Yassin, A.A., Maasum, T.N.R.T.M.: Formalizing informal CALL in learning English language skills. In: Enhancements and Limitations to ICT-Based Informal Language Learning: Emerging Research and Opportunities, pp. 161–182. IGI Global (2020) 66. Hussein, A., Mohammad, M.: Negative L1 impact on L2 writing. Int. J. Humanit. Soc. Sci. 1(18), 184–195 (2011) 67. Shukri, N. A.: Second language writing and culture: Issues and challenges from the Saudi learners perspective. Arab. World Engl. J. 5(3) (2014) 68. Kalyani, D., Rajasekaran, K.: Innovative teaching and learning. J. Appl. Adv. Res. 3(1), 23–25 (2018) 69. Hussin, A.A.: Education 4.0 made simple: Ideas for teaching. Int. J. Educ. Lit. Stud. 6(3), 92–98 (2018)

A General and Theoretical Background of English

381

70. Alshammari, T., Messom, C., Cheung, Y.: Citizens’ adoption of digital technologies during COVID-19 (2020) 71. Rico-Bautista, D., Medina-Cardenas, Y., Coronel-Rojas, L.A., Cuesta-Quintero, F., MaestreGongora, G., Guerrero, C.D.: Smart university: key factors for an artificial intelligence adoption model. In: Advances and Applications in Computer Science, pp. 153–166. Springer, Electronics and Industrial Engineering (2021) 72. Alfiras, M., Yassin, A.A., Bojiah, J.: Present and the future role of the internet of things in higher education institutions. J. Posit. Psychol. Wellbeing 6(1), 167–175 (2022) 73. Skinner, B.F.: Teaching Science in high school—What Is wrong? scientists have not brought the methods of science to bear on the improvement of instruction. Science 159(3816), 704–710 (1968) 74. Salinas, Á., Nussbaum, M., Herrera, O., Solarte, M., Aldunate, R.: Factors affecting the adoption of information and communication technologies in teaching. Educ. Inf. Technol. 22(5), 2175–2196 (2016). https://doi.org/10.1007/s10639-016-9540-7 75. Alfiras, M., Bojiah, J., Yassin, A.A.: COVID-19 pandemic and the changing paradigms of higher education: A Gulf university perspective. Asian EFL J. 27(5.1), 339–347 (2020) 76. Walker, A., White, G.: Technology Enhanced Language Learning: Connecting Theory and Practice-Oxford Handbooks for Language Teachers. Oxford University Press (2013) 77. Strickland, B., Brien, M.O.: A Review of the Literature on Technology in Second and Foreign Language Learning. The Language Research Centre at the University of Calgary, p. 1 (2013) 78. Sajjad, I., Sarwat, S., Imran, M., Shahzad, S.K.: Examining the academic writing challenges Faced by university students in KFUEIT. PalArch’s J. Archaeol. Egypt/Egyptol. 18(10), 1759– 1777 (2021) 79. Saeed, M.A. Al Qunayeer, H.S.: Exploring teacher interactive e-feedback on students’ writing through Google Docs: factors promoting interactivity and potential for learning. Lang. Learn. J. 1–18 (2020) 80. Defede, N., Magdaraog, N.M., Thakkar, S.C., Bizel, G.: Understanding how social media is influencing the way people communicate: verbally and written. Int. J. Mark. Stud. 13(2), 1 (2021) 81. Huynh, T.-N.: A complex dynamic systems approach to foreign language learners’ anxiety in the emergency online language classrooms. Comput. Assist. Lang. Learn. Electron. J. 22(3), 200–229 (2021) 82. Khandan, R., Shannon, L.: The effect of teaching–learning environments on student’s engagement with lean mindset. Educ. Sci. 11(9), 466 (2021) 83. Dal, B.Ö., Güler, D., Tufan, E., Kaban, A.L.: Using padriseup as a collaborative writing tool in higher education EFL classes. In: EDEN Conference Proceedings, vol. 1, pp. 1–19 (2021) 84. Doyle, E., Buckley, P., McCarthy, B.: The impact of content co-creation on academic achievement. Assess. Eval. High. Educ. 46(3), 494–507 (2021)

Measuring Educator Satisfaction of Learning Analytics for Online Learning Systems in Malaysia Nur Maisarah Shahril Khuzairi(B) , Zaihisma Che Cob, and Thaharah Hilaluddin Universiti Tenaga Nasional, Jalan Ikram-Uniten, 43000 Kajang, Selangor, Malaysia [email protected]

Abstract. The use of online learning systems has proliferated, which has caused an explosion of generated data. The use of learning analytics to sort through this data for valuable performance-enhancing insights is fast becoming the norm. Although many research works are published on learning analytics, there is still a lack of research that identifies learning analytics features as a significant influencing factor in improving educator satisfaction. The purpose of this paper is to examine the influence of learning analytics features that leads to a successful implementation of learning analytics for online learning systems. A survey involving 134 educators from 17 public and private higher education institutions in Malaysia was conducted. Data were analyzed using the structural equation modeling method, and the results showed a significant influence of learning analytics on educator satisfaction. This empirical evidence gives insight into the development of successful learning analytics for online learning systems. Keywords: Learning analytics · Online learning systems · e-learning · Learning management systems · Educator satisfaction · DeLone and McLean · IS Success model

1 Introduction Teaching and learning processes have rapidly moved to an online setting due to technological advancements and environmental and social factors. To facilitate this transition to online teaching and learning, higher education institutions have developed and adopted many online learning systems. This has caused an explosion of generated data. The use of learning analytics to sort through this readily accessible data for valuable performance-enhancing insights is fast becoming the norm. However, there is a glaring issue in how we examine the successful implementation of online learning systems with learning analytics that add value to educators for effective decision making. With this in mind, the purpose of this paper is to investigate the role of learning analytics features in influencing educator satisfaction. In order to achieve the intended research outcomes effectively, it is crucial to evaluate various factors of an online learning system [1]. Specifically, more attention needs to be focused on system quality in an online learning system when learning analytics is © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 382–391, 2023. https://doi.org/10.1007/978-3-031-20429-6_35

Measuring Educator Satisfaction of Learning Analytics

383

involved. There is a dearth of research that methodically models as well as validates the influence of learning analytics for its successful implementation in online learning systems in higher education institutions, mainly from the educators’ perspective. Hence, the influence of learning analytics features on online learning systems from the perspective of educator satisfaction is explored in this study. The Information System success model [2] by DeLone and McLean was applied to solve this research problem. The research outcome has a high potential to add valuable insight for researchers, policymakers, and practitioners when implementing learning analytics for online learning systems in the higher education domain of Malaysia. Section 2 will define online learning systems, learning analytics, DeLone and McLean’s Information System success model, and the key constructs in this study: Learning Analytics Features, System Quality, and Educator satisfaction. Next Sect. 3 moves forward by defining the research hypotheses, methodology, and data analysis presentation. Finally, in Sects. 5 and 6, we discuss the findings that posit recommendations and conclusions.

2 Literature Review 2.1 Online Learning Systems Online learning is defined as the use of information-communication technology, specifically the internet, for education purposes [3], whether one is conducting lecture classes, student learning, or assessment. In other words, online learning is basically education made available through devices that have access to the internet [4, 5]. Since its introduction, online learning has been coined with many other terms such as “e-learning,” “blended learning,” and “distance learning,” but all are fundamentally the same, which is the use of the internet for education. The primary tools spearheading online learning are learning management systems (LMS) and course management systems (CMS). These software and hardware infrastructures are the cornerstone of an educational institution’s online learning deployment to connect and facilitate the teaching and learning process between educators and students. Online learning systems have been a consistent focus of education research for the past two decades, and their benefits are well-documented [6]. In spite of the well-documented examples and benefits of online learning systems, it has raised some questions about its impact on educator satisfaction when learning analytics is introduced into the system. 2.2 Learning Analytics Learning analytics is defined as “the measurement, collection, analysis, and reporting of data about learners and their contexts, for purposes of understanding and optimizing learning and the environments in which it occurs” [7]. Since the shift to online learning, there has been rapidly growing interested and research activity in learning analytics. Through the years, online learning systems have been improved with learning analytics to measure and analyze teaching and learning data to provide actionable insights for decision-making. However, there is a concern within this emerging field of learning

384

N. M. Shahril Khuzairi et al.

analytics for online learning systems, which is how to determine when information is communicated successfully via learning analytics to its target user and how to evaluate that success [8]. 2.3 Information System (IS) Success Model One of the highly referenced and used research works in technology is the Information System (IS) success model by DeLone and McLean [9]. It is a model to examine the factors affecting IS success and its impact. The work by DeLone and McLean [9] is a comprehensive model with 6 significant variables that measure IS success from various aspects which are information quality, system quality, user satisfaction, system use, and organizational/individual impact. Then years later, changes were made to the model with the introduction of ‘service quality, a new determinant of user satisfaction and system use [2]. The new determinant ‘service quality’ highlights on the quality of support such as end-user support to the system user and ease of maintenance of the overall system. With that, there are three major aspects of qualities that contribute to user satisfaction and system use. However, the strength of information quality, system quality and service quality depends heavily on the context of the studies and also the type of statistical analyses used. As a general rule, to measure the success of information system in an individual level, focus can be on information and system quality while service quality inclines towards the organization level. Various research works have employed and tested the causal relationships between each variable in the model, especially in the field of online learning [3, 10–16]. The purpose of these research works are to investigate the factors that contribute to the successful implementation of an online learning system, and based on the findings, the hypothesized relationships are generally found to be positively significant. System Quality. System quality in IS Success model’s [9] plays an important role in determining IS success. System quality encompasses general features such as system availability, reliability, and usefulness of the functionalities. Although system quality is not posited as a direct determinant but does have a significant effect on user satisfaction with an information system is well proven and cannot be denied [2, 17, 18]. Additionally, the traditional software development life cycle (SDLC) methodology also stresses the importance of system quality. Thus in this study, system quality is proposed as a construct affecting educators’ satisfaction. System quality relates to the system’s characteristics, such as its functionalities, availability, and reliability of the system, along with analytics features in place to improve user efficiency. Learning Analytics Features. The importance of learning analytics features is well documented [19, 20, 21]. A recent study by Bakharia and colleagues [20] has set out to verify various types of analytics, and their role played in supporting educators for the evaluation of learning designs. The five types of analytics identified were namely temporal, comparative, cohort dynamics, tool specific, and contingency. Other studies also affirmed that learning analytics features significantly affect learning analytics tool design [22, 23, 24]. Notably, the previous work has not quite given much consideration to the aspect of learning analytics features influencing the system quality of an information

Measuring Educator Satisfaction of Learning Analytics

385

system. It is asserted that a system with implemented analytics is to make sure that large amounts of data are presented in a way that is easy to understand and used with relative ease by educators from all walks of life increases the positive satisfaction of the said system, particularly for those with limited technological expertise. Therefore the influence of learning analytics features in terms of how it affects the overall system quality is believed to be an important determinant for supporting educator satisfaction in an online learning system. Educator Satisfaction. User satisfaction in the IS success model is considered as one of the factors in measuring the success of an information system. It is identified as a factor predicting one’s actual usage of the learning application [13], particularly among higher education institutions. Awang et al. [25] examined the influence of user satisfaction among educators, and their findings indicated that user satisfaction is a significant predictor of the use of new technology. Also, when investigating the utilization of online learning, user satisfaction also appeared as an important factor influencing one’s attitude, whether it is positive or negative towards usage of the online learning [26]. In this study, user satisfaction is included and defined as the perception of educators that learning analytics features in the system will satisfy their needs and how well it lives up to their anticipations; thus, it is measured specifically as educator satisfaction instead.

3 Research Hypotheses and Methodology 3.1 Hypotheses To examine the impact of learning analytics for online learning systems on educator satisfaction, this study used DeLone and McLean’s IS success model, focusing on system quality, and educator satisfaction. Additionally, the learning analytics features construct was put forth as an antecedent to system quality, leading to educator satisfaction. Hence, the research hypotheses for this study are: H1: Learning Analytics Features are positively associated with System Quality. H2: System Quality is positively associated with Educator Satisfaction. The research model of the present study and the hypotheses are presented in the figure below (Fig. 1).

Fig. 1. The research model.

3.2 Methodology A quantitative means via survey was employed in this study. According to Creswell (2008), the use of quantitative data is most prevalent in educational research as there

386

N. M. Shahril Khuzairi et al.

are significant results to justify the findings. The advantage of this approach is that the researcher can employ probability models to obtain conclusive results to answer each specified research question [27]. Study measurements. In this study, a survey questionnaire consisting of 19-item was designed to gauge user perceptions of learning analytics reporting in online learning systems based on the model’s constructs. The five-point Likert scale with a range of 1 = “strongly disagree” to 5 = “strongly agree” was used as a measurement. Participants. This study targeted public (IPTA) and private (IPTS) higher education institutions in Malaysia. Furthermore, since the targeted respondents were educators of higher education who have access to online learning systems or related systems, those with at least six months of experience using such systems were selected for the survey. Survey email invitations were sent out to the registrar department in 17 public and private universities in Malaysia, with an average of 100 educators per institution. A total of 204 questionnaires were collected; this is an average response rate of 20%. However, only 134 responses are valid and usable for further analyses.

4 Data Analyses Quantitative data collected from the self-administrated survey were analyzed using the selected software SmartPLS Version 2.0.M4. The SmartPLS was utilized to conduct Confirmation Factor Analysis (CFA) as well as PLS-SEM. The purpose of these analyses is to confirm the proposed model’s validity and reliability. Additionally, to assess the structural model and also validate the hypotheses proposed. In quantitative data analyses, it is crucial to examine the consistency and reliability of the dataset. According to Hair et al. [28], reliability is the overall degree of consistency among the indicators. Thus, to verify the reliability of the dataset, Cronbach Alpha (CA) and Composite Reliability (CR) were employed. CA values of 0.7 and above are acceptable for social science studies [29], whereas those above 0.6 for CR are considered acceptable. Based on the results, all the CA values are in the acceptable range above 0.7, and all CR values are above 0.6, in the range of 0.83 and 0.94 (as shown in Table 1). Average Variance Extract (AVE) values were used to measure the convergent validity. Bagozzi and Yi [30] and Fornell and Larcker [31] suggest adequate convergent validity only when AVE values exceed 0.5. As shown in Table 1, all AVE values are above 0.5, ranging from 0.542 to 0.717, inferring adequate convergent validity. Next, discriminant validity is assessed by Fornell and Larcker’s [31] criterion via the cross-loadings of the indicators. The discriminant validity indicates the measure of constructs are truly distinct and should not be highly related to each other. The square root of AVE and its loadings of an item on its assigned latent construct should be greater than its cross-loadings and all other constructs measured in the study [31, 32]. The results of the discriminant validity are presented in Table 1. Once the measure of both validity and reliability were proved, the structural model assessment was performed, which includes the model’s hypothesized relationships. This was assessed via the significance of path coefficients and t-values between each construct, as presented in Table 2. As shown in the results, all the path coefficient findings were

Measuring Educator Satisfaction of Learning Analytics

387

Table 1. Convergent and discriminant validity results. Constructs

CA

CR

Discriminant validity AF

SQ

Learning Analytics Features (AF)

0.894

0.914

0.736

System Quality (SQ)

0.870

0.907

0.601

0.814

Educator Satisfaction (ES)

0.865

0.927

0.531

0.673

AVE

ES 0.542 0662 0.847

0.717

positive, and the proposed hypotheses are supported, with a significance level p-value less than 0.01. Additionally, all two (2) hypotheses associations are significant in relation to learning analytics features, and system quality that predicted a positive effect on educator’s satisfaction. Table 2. Research hypothesis testing. Hypotheses

Path coefficient, β

t-value

p-value

Decision

H1: AF −> SQ**

0.601

9.577

0.000

Supported

H2: SQ −> ES**

0.641

9.775

0.000

Supported

Note: ** p < 0.01

5 Discussion This study investigates the role of learning analytics features in influencing educator satisfaction for the successful implementation of online learning systems with learning analytics in higher education institutions. Although a significant majority of teaching and learning have been moved to online learning systems due to the pandemic, its effectiveness and successful implementation are determined by educator satisfaction. Two factors were selected in an attempt to explain educator satisfaction with learning analytics for online learning systems: system quality and learning analytics features. Findings from this study found system quality is a strong predictor of educator satisfaction, with a value of (β: 0.641, t-value: 9.775, p < 0.01). These findings suggest that if learning analytics for online learning systems is available, accessible, reliable, accurate, and fast, educators will generally be satisfied to use it in their teaching and learning. Such findings are consistent with the literature on the success of an information system in various fields. The significance of system quality construct in the IS success model to improve user satisfaction are indisputable, it has gained recognition over the years, and they have been extensively used and validated in research related to information system and e-learning [3, 33, 34]. Although system quality is not posited as the key predictor of an information system’s success, the result has shown that it influences overall educators’ satisfaction with an online learning system.

388

N. M. Shahril Khuzairi et al.

Additionally, the relationship between learning analytics features and system quality was also found to be significant, with a value of (β: 0.601, t-value: 9.577, p < 0.01). This finding could indicate that in order to incorporate learning analytics into an information system, it is important to include different learning analytics features needed by users as it will increase the system quality of an online learning system, subsequently leading to enhanced educator satisfaction. Reviews of past literature did not examine the relationship between learning analytics and system quality, particularly when learning analytics is put in the picture of Malaysia’s higher education context. For example, mainstream analytics reporting modules expanded their functionalities beyond merely presenting and reporting basic educational data. Attributes of commonly used learning management system qualities of an analytics reporting module such as Moodle enable users to view students’ results and added functionalities such as tracking their engagement data, interaction in a forum, and course access to heatmap overlay to see students’ counts of visits. These learning analytics data contribute to the basic functionalities in improving overall system quality. The proven significance of the learning analytics features constructs towards system quality points strongly to the importance of learning analytics to enhance educator satisfaction. This is especially vital given the rapid proliferation of online learning due to the growth of education technology and the effect of the Covid-19 pandemic. A study by 2021 Horizon Report has also identified learning analytics as a key technological trend to shape the future of teaching and learning, especially during this time of crisis [35]. Apart from that, with large amounts of educational data readily available via online learning systems, higher education institutions are forced to enter the era of “big data” [36], where the use of learning analytics will be front and center to support educators and institutions to make better evidence-informed decisions instead of ’trend-guessing’. Thus understanding learning analytics significance towards online learning systems is crucial to successfully implementing online learning in higher education institutions.

6 Conclusion With education data easily collected and made available via online learning systems, the field of learning analytics is thriving with the latest research work in relation to its tools and techniques to improve online teaching and learning efficiency and effectiveness. Though the benefits of learning analytics are clear, they must be supported with proper theory. We must draw attention to its impact and how it influences the overall system in order to have a successful uptake instead of merely adopting learning analytics as a latest technological add-on. Hence, the present study is significant because it examines the influence of learning analytics on the overall system based on the IS success model. It investigates the effect of system quality and learning analytics features on educator satisfaction. The findings showed that researchers and practitioners, and developers should put much additional consideration into learning analytics features to increase system quality. Good and relevant analytics reports will improve system quality, leading to higher satisfaction levels which increase the use of an application or system. Today, due to the growth of educational technology, online learning is growing at an unprecedented rate, a better understanding of the effective implementation of learning

Measuring Educator Satisfaction of Learning Analytics

389

analytics for online learning systems that impact educator satisfaction will help narrow the research gap concerning the lack of studies about online learning and learning analytics. By empirically validating the IS success model from a different context, particularly from a developing country Malaysia, the study also contributes to the body of knowledge on the successful implementation of learning analytics for online learning systems.

References 1. Seliana, N., Suroso, A.I., Yuliati, L.N.: Evaluation of e-learning implementation in the university using DeLone and McLean success model. J. Appl. Manag. 18(2), 345–352 (2020) 2. DeLone, W.H., McLean, E.: The DeLone and McLean model of information system success: a ten-year update. J. Manag. Inf. Syst. 19(4), 9–30 (2003) 3. Aldholay, A., Abdullah, Z., Isaac, O., Mutahar, A.M.: Perspective of Yemeni students on use of online learning: extending the information systems success model with transformational leadership and compatibility. Inf. Technol. People 33(1), 106–128 (2019) 4. Haynes, A.S.: Establishing considerations for universal design for learning and accessibility in online courses. In: Research Anthology on Developing Effective Online Learning Courses. IGI Global, Columbia (2021) 5. Ates-Cobanoglu, A., Yucel, Z.E., Kilic, M.: Online course design tips for boosting learner autonomy with synchronous and asynchronous tools. In: Handbook of Research on Managing and Designing Online Courses in Synchronous and Asynchronous Environments. IGI Global, Turkey (2022) 6. Singh, V., Thurman, A.: How many ways can we define online learning? A systematic literature review of definitions of online learning (1988–2018). Am. J. Distan. Educ. 33(4), 289–306 (2019) 7. EDUCAUSE, https://er.educause.edu/articles/2011/9/penetrating-the-fog-analytics-in-lea rning-and-education , last accessed 2022/05/10 8. Hung, Y., Parsons, P.: Affective Engagement for communicative visualization: quick and easy evaluation using survey instruments. In: Visualization for Communication (VisComm) Workshop 2018. IEEE, Berlin Germany (2018) 9. DeLone, W.H., McLean, E.R.: Information system success: the quest for the dependent variable. Inf. Syst. Res. 3(1), 60–95 (1992) 10. Molokeng, P.M., Van Belle, J.-P.: Investigating the use of learning analytics at South Africa’s higher education institutions. In: Antipova, T. (ed.) DSIC 2021. LNNS, vol. 381, pp. 59–70. Springer, Cham (2022). https://doi.org/10.1007/978-3-030-93677-8_6 11. Clark, J. A., Tuffley, D.: Learning Analytics implementations in universities: Towards a model of success, using multiple case studies. In: Chew, Y.W., Chan, K.M., Alphonso, A. (eds.) Australasian Society for Computers in Learning in Tertiary Education, ASCILITE, pp. 82–92. Singapore (2019) 12. Uddin, M.D.M., Isaac, O., Alrajawy, I., Maram, M.A.: Do user satisfaction and actual usage of online learning impact student performance? Int. J. Manag. Human Sci. 3(2), 60–67 (2019) 13. Yakubu, M.N., Dasuki, S.I.: assessing elearning system success in Nigeria: an application of the Delone and McLean information systems success model. J. Inf. Technol. Educ.: Res. 17, 184–203 (2018) 14. Thongsri, N., Shenb, L., Bao, Y.: Investigating factors affecting learner’s perception towards online learning: evidence from ClassStart application in Thailand. Behav. Inf. Technol. 38(12), 1243–1258 (2019)

390

N. M. Shahril Khuzairi et al.

15. Balaban, I., Stacin, K., Sobadic, A.: Analysis of correlations between indicators influencing successful deployment of ePortfolios. In: Skala, K., Koricic, M., Grbac, T.G., Cicin-Sain, M., Sruk, V., Ribaric, S., Gros, S., Vrdoljak, B., Mauher, M., Tijan, E., Pale, P., Janjic, M. (eds.) International Convention on Information and Communication Technology, Electronics and Microelectronics 2018, pp. 0788–0793. IEEE, Opatija, Croatia (2018) 16. Park. S.Y.: An analysis of the technology acceptance model in understanding university students’ behavioural intention to use e-learning. Educ. Technol. Soc. 12(3), 150–162 (2009) 17. Lin, W.S., Wang, C.H.: Antecedences to continued intentions of adopting e-learning system in blended learning instructions: a contingency framework based on models of information systems success and task-technology fit. Comput. Educ. 58(1), 88–99 (2012) 18. Detlor, B., Hupfer, M.E., Ruhi, U., Zhao, L.: Information quality and community municipal portal use. Gov. Inf. Q. 30(1), 23–32 (2013) 19. Tempelaar, D., Rienties, B., Mittelmeier, J., Nguyen, Q.: Student profiling in a dispositional learning analytics application using formative assessment. Comput. Hum. Behav. 78, 408–420 (2018) 20. Bakharia, A., Corrin, L., Barba, P.D., Kennedy, G., Gasevic, D., Mulder, R., Williams, D., Dawson, S., Lockyer, L.: A conceptual framework linking learning design with learning analytics. In: Gasevic, D., Lynch, G. (eds.) Proceedings of the 6th International Learning Analytics and Knowledge Conference, pp. 329–338. ACM, New York (2016) 21. Lockyer, L., Heathcote, E., Dawson, S.: Informing pedagogical action: aligning learning analytics with learning design. Am. Behav. Sci. 57(10), 1439–1459 (2013) 22. Sarmonpal, S.: Learning Analytics From Research to Practice: A Content Analysis to Assess Information Quality on Product Websites. Pepperdine University, Malibu California (2018) 23. Greller, W., Drachsler, H.: Translating learning into numbers: a generic framework for learning analytics. Educ. Technol. Soc. 15(3), 42–57 (2012) 24. Verbert, K., Manouselis, B., Drachsler, H., Duval, E.: Dataset-driven research to support learning and knowledge analytics. Educ. Technol. Soc. 15(3), 133–148 (2012) 25. Awang, H., Aji, Z., Osman, W.R.S., Ishak, M.S.: Examining virtual learning environment success using DeLone and McLean IS success model. In: Pacific Asia Conference on Information Systems. Association for Information Systems, Yokohama Japan (2018) 26. Seta, H.B., Wati, T., Muliawati, A., Hidayanto, A.N.: E-learning success model: an extension of DeLeon & McLean IS success model. Indonesian J. Electr. Eng. Inf. 6(3), 281–291 (2018) 27. Smith, J.K., Heshusius, L.: Closing down the conversation: the end of the quantitativequalitative debate among educational inquiries. Educ. Res. 15(1), 4–12 (1986) 28. Hair, J.F., Hult, G.T.M., Ringle, C.M., Sarstedt, M.: A Primeron Partial Least Squares Structural Equation Modeling (PLS-SEM). Sage, Thousand Oaks California (2014) 29. Nunnally, J.C.: Psychometric theory. McGraw-Hill, New York (1978) 30. Bagozzi, R.P., Yi, Y.: On the evaluation of structural equation models. J. Acad. Mark. Sci. 16, 74–94 (1988) 31. Fornell, C., Larcker, D.F.: Evaluating structural equation models with unobservable variables and measurement error. J. Mark. Res. 18(1), 39–50 (1981) 32. Chin, W.W.: How to write up and report PLS analyses. In: Vinzi, V.E., Chin, W.W., Henseler, J., Wang, H. (eds.) Handbook of Partial Least Squares, pp.655–690. Springer Cham, Berlin. (2010) 33. Al-Kofahi, M.K., Hassan, H., Mohammad, R.: Information system success model: a review of literature. Int. J. Innov. Creativ. Change 12(8), 104–128 (2020) 34. Hsu, M.H., Chang, C.M., Chu, K.K., Lee, Y.J.: Determinants of repurchase intention in online group-buying: the perspective of DeLeone & McLean IS success model and trust. Comput. Hum. Behav. 36, 234–245 (2014)

Measuring Educator Satisfaction of Learning Analytics

391

35. Pelletier, K., Brown, M., Brooks, D.C., McCormack, M., Reeves, J., Arbino, N., et al.: 2021 EDUCAUSE Horizon Report Teaching and Learning Edition. EDU, Boulder (2021) 36. Wibawa, B., Siregar, J.S., Asrorie, D.A., Syakdiyah, H.: Learning analytics and educational data mining for learning science and technology. In: Meiliasari, M., Rahmawati, Y, Delina, M., Fitriani, E. (eds.) AIP Conference Proceedings, vol. 2331, pp. 060001–1–060001–7. AIP Publishing Jakarta (2021)

Digital Support, Teacher Support, and Blended Learning Performance: Investigating the Moderating Effect of Gender Using Multigroup PLS-SEM Analysis Mohammed Ali Al-Awlaqi1(B) , Ahmed Mohammed Taqi2 Nor Hasliza Binti Md Saad2 , and Nezar Al-Samhi1

,

1 School of Business, Lebanese International University, Sana’a, Yemen

[email protected], [email protected] 2 School of Management, Universiti Sains Malaysia, 11800 USM Penang, Malaysia

[email protected], [email protected]

Abstract. This study measured the effect of digital support and teacher support on learning performance. Moreover, it measured the moderating effect of gender on this relationship. The study utilized data from 209 students who enrolled in a blended learning program. It used partial least square structural equation modeling PLS-SEM to test the direct relationship between digital support and teacher support on learning performance, while a multigroup PLS-SEM technique was used to test the moderating effect of gender on the relationship between digital support, teacher support and learning performance. The digital and teacher support variables showed a significant impact on learning performance. Digital support showed negative impact, while teacher support showed a positive impact. Gender showed a significant moderating effect on the relationship between digital, teacher support and learning performance. Keywords: Digital support · Teacher support · Blended learning · Multigroup analysis · Yemen

1 Introduction Blended learning combines two elementary learning systems, the traditional in class system or face-to-face and the modern online learning system [1]. It offers a more private and stress-free learning environment [2]. To have successful implementation, blended learning should balance the support for the two systems. It is an important linkage in the chain of the global transformation of teaching methods toward advanced technological approaches [3]. Blended learning was found to be more effective than other types of teaching approaches [3]. Students in Yemen are not familiar with new learning systems such as blended learning. Yemeni students usually link their success in learning to the traditional faceto-face systems. The perception toward the blended learning system could affect its © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 392–401, 2023. https://doi.org/10.1007/978-3-031-20429-6_36

Digital Support, Teacher Support, and Blended

393

performance. Understanding the determinants of the student’s perception can improve the effectiveness of the blended learning system’s implications in such a challenging context. Moreover, adopting semester-length blended learning to align with the university education system can result in undesired consequences [4]. These problems raise the importance of understating factors that could enhance the outcome of blended learning programs, especially in an in-need context such as Yemen [5]. Blended learning performance can be affected by the two main parts of this type of learning scheme, digital support, and teacher support. Teacher support represents the face-to-face part while digital support represents the online part. Teacher support is an essential part of blended learning. Blended learning enhances ability of the student to understand the teacher’s expectations [6]. Teachers can influence students learning by encouraging them to behave positively toward learning process, provide the optimal mix of learning resources, and have direct contact and impact on student and their learning behavior [7]. Teachers in blended learning are considered influential challenges in blended learning applications [8]. Teacher support motivates students as a real-life experience of the pedagogical process. Thus, the teacher should maintain a variety of pedagogical approaches to be more effective in the blended learning programs [9]. Others took another direction and studied the feeling, anxieties, attributes of teachers and their impact on Blended learning implementation [10, 11]. The pivot role of teacher support made previous literature recommend further investigation to assess the role of teachers in blended learning programs [8]. Digital support represents the second pivot part of blended learning. Digital support is about helping learners fulfill their learning needs through a supportive technological learning environment [1]. The online or virtual environment is usually associated with higher students’ satisfaction [6]. It is an important tool to shift learning attention from teacher-centered to learner-centered environments [12]. Technology can offer potential opportunities to offer a personalized learning experience [12]. Digital support is also found to bridge learners’ knowledge gap when they shift from one education level to the other [13]. This shows the importance of digital support in implementing successful blended learning programs. Literature tried to understand learner’s individual differences as predictors of learning performance. Some studied metacognitive regulation, metacognitive knowledge, efforts, and self-efficacy [14]. Others investigated the role of online activities and found positive effects [15]. However, no study tried to measure the effect of major differences such as gender on learning performance. Gender should not be neglected in any attempt to understand learning processes [7]. Male learners showed less learning engagement and perceived less support from their teachers [7]. Student’s learning performance is about improving the ability and skills of students to solve problems. Learning performance is also about improving critical thinking, and the overall course grade [16]. Learning performance is the most accurate measurement of blended courses’ effectiveness. Literature tried to establish a relationship between blended learning and student performance. Meta analysis found that blended learning demonstrated a slightly higher

394

M. A. Al-Awlaqi et al.

effect on student performance than traditional teaching schemes [17]. The effect of students’ engagement on learning performance was found to be significant, especially on web-based learning [18]. Using a quasi-experiment design individualized intervention showed significant impact on learning performance [19]. Other studies the impact of learners’ phycological characteristics such as emotion, motivation, and cognitive behaviors [20]. Others focused on the course’s characteristics to explain learning performance. Course difficulty, clear goals, quality of materials, content, and learning collaboration affected the level of learning performance [21, 22]. Literature also linked learning performance to perceived precision teaching [23]. Perceived precision teaching was found to have an indirect positive impact on learning performance through self-efficacy and learning motivation. But previous literature showed an unpreferable effect of blended learning on learning performance. Blended learning worsens the persistence of learners and has a small effect on pass rates [24]. The discussion above revealed a significant gap in previous blended learning literature. Previous literature did not give attention to the mutual effect of the face-to-face and online learning schemes as the main parts of blended learning programs. No previous study has attempted to understand the impact of digital support and teacher support on blended learning performance. Thus, this study filled this gap and investigated the direct relationship between digital and teacher support on blended learning performance. In addition, this study investigated the moderating effect of gender on this relationship. To achieve the objective of the current study, the research questions are formulated: RQ1: Is there a direct effect of digital support on blended learning performance? RQ2: Is there a direct effect of teacher support on blended learning performance? RQ3: Is there a moderating effect of gender on the relationship between digital and teacher support on blended learning performance?

2 Methodology Multigroup Partial least square structural equation modeling PLS-SEM is used in this study as the main technique to analyze the data. PLS-SEM is one of the most popular techniques in social and business studies [25]. PLS-SEM is a variance-based technique used as an exploratory analytical tool. Multigroup PLS-SEM is used to evaluate moderating effects by comparing the effect of the causal relationship between the targeted groups. 2.1 Measurements Digital support was measured using three sub-dimensions. These three sub dimensions were: relatedness support, autonomy support, and competence support [26]. Relatedness support was measured using 8 items. Autonomy support was measured using 7 items. Finally, competence support was measured using 6 items.

Digital Support, Teacher Support, and Blended

395

Teacher support was measured using their subdimensions which were recommended by previous literature. These three dimensions were: relatedness support, autonomy support, and competence support [27]. Relatedness support was measured using 5 items. Autonomy support was measured using 15 items, while competence support was measured using 4 items. Students’ learning performance was measured using three items. These items were used based on the measurement tool introduced by Law et al. [16]. The variables in this study were measured using the most common 5-points Likert scale in which 5 represents strongly agree while 1 represents strongly disagree. Gender was measured using a binary variable of 0 for males and 1 for females. 2.2 Sample and Data The data of this study came from two groups of students. The first group of 154 students from one of the Yemeni private universities. This group of students consisted of freshmen students who enrolled in blended classes on accounting 101 and accounting 102. The second group consisted of 127 freshmen and sophomore students who took another blended class on college algebra and business math courses. These blended classes were conducted over two academic semesters. The face-to-face part of these classes was conducted over three hours a week. The online parts of the blended classes were conduct using Google class platforms and Moodle platforms. The online parts were designed to give the student the freedom to be autonomous in learning and fully self-dependent. The questionnaire of the study was distributed manually to the targeted students. Students were given two days to answer the questions. The mean of the answering time was 2.4 days with at least 1 day and a maximum of 4 days. In total, the questionnaire was distributed to 281 students. 29 students didn’t return back their answers. Another 43 questionnaires were excluded due to significant missing data problems. The final sample size of the study was 209 respondents. The sample is considered enough to conduct a multigroup partial least square structural equation modeling. This sample size is adequate for the multigroup PLS-SEM as the recommended sample size should be between 100 to 200 respondents [28]. The sample description showed that 59% of the students were female, while 41% were males. The majority of the students were between 20–21 years old with a percentage of 55, 13% were less than 20 years old, and 32% were older than 21 years.

3 Data Analysis The analysis started with testing Cronbach’s Alpha, rho A, composite reliability, and average variance extracted. As shown in Table 1, the model did not suffer from any problem with these tests. In the second step, we tested the model fit using Standardized Root Mean Squared Residual (SRMR) as the most common test used in PLS-SEM. The model showed an extremely good fit for the data. Table 2 shows the SRMR test. To test the moderating effect of the gender variable, first, the effect of digital and teacher support were tested on the entire group (males and females). Digital support

396

M. A. Al-Awlaqi et al.

Table 1. Cronbach’s Alpha, rho A, composite reliability, and average variance extracted test. Cronbach’s alpha

rho_A

Composite reliability

Average Variance Extracted (AVE)

DS_autonomy 0.971

0.972

0.976

0.851

DS_ competence

0.956

0.97

0.965

0.822

DS_ relatedness

0.978

0.98

0.981

0.868

Digital_ support

0.982

0.984

0.983

0.604

Learning_ performance

0.938

0.939

0.96

0.89

TS_autonomy

0.988

0.988

0.989

0.856

TS_ competence

0.965

0.966

0.974

0.905

TS_ relatedness

0.981

0.982

0.985

0.931

Teacher_ support

0.988

0.988

0.989

0.787

Table 2. Root Mean Squared Residual (SRMR).

SRMR

Saturated model

Estimated model

0.091

0.096

shows significant yet negative impact on learning performance with β = −0.466, t(208) = 6.78, p-value < 0.001. Teacher support shows significant and positive effect on learning performance with β = 0.39, t(208) = 5.04, p-value < 0.001. The results are shown in Table 3. Conducting a multiple group PLS-SEM analysis shows different results and confirms the moderating effect of the gender variable Table 4. Multiple group analysis shows that digital support still negatively impacts learning performance. Digital support shows negative yet insignificant effect on learning performance of the males group with β = −0.03, t(208) = 0.279, p-value = 0.78. Digital support shows negative and significant effect on the learning performance of the female group with β = −0.89, t(208) = 25.98, p-value < 0.001. The gender variable does not show a moderating effect on the relationship between teacher support and learning performance. Multiple group analysis shows a significant impact of teacher support on learning performance of the male group with β = 0.58, t(208) = 6.97, p-value = 0.002. Moreover, Multiple group analysis shows a significant impact of teacher support on the

Digital Support, Teacher Support, and Blended

397

Table 3. Path analysis of one group. Original sample (O) Digital_support −0.466 -> Learning_ performance Teacher_ support -> Learning_ performance

0.39

Sample mean (M)

Standard deviation (STDEV)

T Statistics (|O/STDEV|)

P Values

−0.466

0.069

6.777

0.000

0.394

0.078

5.035

0.000

learning performance of the female group with β = 0.11, t(208) = 3.17, p-value < 0.001.

4 Conclusion and Discussion Using data from a group of 209 students, this study used PLS-SEM analysis to test the effect of digital support and teacher support on the learning performance of a blended learning program. Moreover, the study tested the moderating effect of gender on the previous relationships using the multigroup PLS-SEM technique. The study showed interesting results. Digital and teacher support significantly affected learning performance in the blended learning scheme. Teacher’s support showed positive while digital support showed negative effects. Gender showed a strong moderating effect on the relationship between digital support and learning performance. On the other hand, gender did not moderate the relationship between teacher support and learning performance. The findings of this study showed an interesting pattern of the blended learning implementation. Digital support was found to negatively impact learning performance. This indicates one weakness of blended learning implementation in Yemen. Learners do not use the online part of blended learning properly. They still stick to the traditional learning schemes and appreciated them more. Teacher support shows a positive impact on learning performance, as expected. Learners in Yemen still can’t get the full potential benefits of blended learning because they appreciate the traditional face-to-face part and do not use the online part properly. Yemeni learners still want to have traditional teacher control to perform well. Gender as a moderating variable also shows insightful results. Gender shows a moderating effect on the relationship between digital support and learning performance. The relationship was significant for the entire group. This result changed when gender was entered as a moderating variable on this relationship. Digital support turned to be insignificant in the male group while it remained significant in the female group. This showed particularly that male learners still do not deal properly with the idea of blended learning. Males still appreciate teacher support and do not appreciate digital support. Thus, males with setting still treated blended classes as traditional ones. This

−0.885

0.112

Digital_support − > Learning_ performance

Teacher_support − > Learning_ performance

Path coefficients original (GENDER_ FEMALE)

0.575

0.029

Path coefficients original (GENDER_MALE)

0.114

−0.882

Path coefficients mean (GENDER_ FEMALE)

0.573

0.032

Path Coefficients Mean (GENDER_ MALE)

0.035

0.034

STDEV (GENDER_ FEMALE)

0.082

0.104

STDEV (GENDER_ MALE)

Table 4. Path analysis of multiple groups.

3.166

25.981

t-Value (GENDER_ FEMALE)

6.967

0.279

t-Value (GENDER_ MALE)

0.002

0

p-Value (GENDER_ FEMALE)

0

0.78

p-Value (GENDER_ MALE)

398 M. A. Al-Awlaqi et al.

Digital Support, Teacher Support, and Blended

399

could indicate the necessity of the teacher part to impose control over male learners. On the other hand, females showed negative yet significant effects of digital support on their learning performance. This impact was strong. Digital support affected female learning performance badly, indicating misuse of digital support autonomy, relatedness, and competence. This study revealed some consistency with previous literature. Previous literature showed a positive relationship between blended learning and student engagement [17]. The effect of students’ engagement on learning performance was found to be significant, especially web-based learning [18]. Moreover, this study showed a significant impact of gender as a moderating variable on the relationship between blended learning and learning performance. This follows previous literature as it showed that individual characters could affect learning performance [20]. Others focused on the course’s characteristics to explain learning performance. Course difficulty, clear goals, quality of materials, content, and learning collaboration affected the level of learning performance [21, 22]. Teacher support showed a significant and positive impact on learning performance, which is in line with previous literature [23]. This study took an advanced step toward understanding the mechanism of blended learning in enhancing learning performance. 4.1 Practical Implications This study offers teachers and education institutions some practical suggestions to improve blended learning programs. Blended learning programs are a viable option for teaching in Yemen. It is practical and can improve learning performances and outcomes. Blended learning programs should be designed to promote digital support effectiveness. Blended learning programs should be designed to smoothly shift Yemeni learners from the traditional learning programs to more technology-supported programs such as blended learning. Digital support should contain more instructions and should promote a more controlled learning environment. The digital and technological environment should offer more explanations of how to be used and how it would help students to improve their learning performance. Blended learning programs should suit the need of both males and females. Males need blended learning programs to have less autonomous environments. Yemeni male learners found a more autonomous environment as a chance to escape their learning duties. The digital and technological environment should guide what skills must be learnt. The digital and technological environment should be given the same attention as the face-to-face learning environment.

5 Conclusion This study aimed to test the direct effect of digital support and teacher support on learning performance and the moderating effect of gender on this relationship. The study applied a multigroup partial least squares structural equation modeling on data from 209 students to answer the research questions. The digital support showed a negative impact on learning performance while teacher support showed a positive impact on the targeted students’ learning performance. Finally, gender showed a significant moderating effect on the relationship between digital, teacher support and learning performance.

400

M. A. Al-Awlaqi et al.

References 1. Chiu, T.K.F.: Digital support for student engagement in blended learning based on selfdetermination theory. Comput. Hum. Behav. 124, 106909 (2021) 2. Warren, L., Reilly, D., Herdan, A., Lin, Y.: Self-efficacy, performance and the role of blended learning. J. Appl. Res. Higher Educ. 13(1), 98–111 (2020) 3. Chao, H.-W., Wu, C.-C., Tsai, C.-W.: Exploring the effects of blended learning, flipped learning, and online remedial teaching on improving students’ learning performance and motivation. Int. J. Technol. Human Interact. 17(3), 98–114 (2021) 4. Kwak, D.W., Menezes, F.M., Sherwood, C.: Assessing the impact of blended learning on student performance. Econ. Record 91(292), 91–106 (2015) 5. Al-Awlaqi, M., Barahma, M., Basrda, T., Al-Tahitah, A.: The effect of B-learning adoption on the evolution of self-regulation skills: a longitudinal study on a group of private universities’ freshman students. In: Al-Emran, M., Al-Sharafi, M.A., Shaalan. K. (eds.), The International Conference on Information Systems and Intelligent Applications (ICISIA 2022). Lecture Notes in Networks and Systems, vol xx. Springer, Cham. (In press) 6. Grønlien, H.K., Christoffersen, T.E., Ringstad, Ø., Andreassen, M., Lugo, R.G.: A blended learning teaching strategy strengthens the nursing students’ performance and self-reported learning outcome achievement in an anatomy, physiology and biochemistry course – A quasiexperimental study. Nurse Educ. Pract. 52, 103046 (2021) 7. Lietaert, S., Roorda, D., Laevers, F., Verschueren, K., De Fraine, B.: The gender gap in student engagement: The role of teachers’ autonomy support, structure, and involvement. Br. J. Educ. Psychol. 85(4), 498–518 (2015) 8. Rasheed, R.A., Kamsin, A., Abdullah, N.A.: Challenges in the online component of blended learning: A systematic review. Comput. Educ. 144, 103701 (2020) 9. Jokinen, P., Mikkonen, I.: Teachers’ experiences of teaching in a blended learning environment. Nurse Educ. Pract. 13(6), 524–528 (2013) 10. Zhao, S., Song, J.: What kind of support do teachers really need in a blended learning context? Australas. J. Educ. Technol. 37(4), 116–129 (2021) 11. Bruggeman, B., Tondeur, J., Struyven, K., Pynoo, B., Garone, A., Vanslambrouck, S.: Experts speaking: Crucial teacher attributes for implementing blended learning in higher education. Internet High. Educ. 48, 100772 (2021) 12. Alamri, H.A., Watson, S., Watson, W.: Learning technology models that support personalization within blended learning environments in higher education. TechTrends 65(1), 62–78 (2020). https://doi.org/10.1007/s11528-020-00530-3 13. Drews, P., Martens, A.: Study properly: Digital support for the Pre-student phase. In: Ifenthaler, D., Isaías, P., Sampson, D.G. (eds.), Orchestration of Learning Environments in the Digital World, pp. 115–134. Springer, Cham (2022) 14. Cannon, J.P., Lohtia, R., Paulich, B.J.: Blended learning in principles of marketing: the effects of student differences on student performance. J. Mark. Educ. 02734753211058357 (2021) 15. Chango, W., Cerezo, R., Romero, C.: Multi-source and multimodal data fusion for predicting academic performance in blended learning university courses. Comput. Electr. Eng. 89, 106908 (2021) 16. Law, K.M.Y., Geng, S., Li, T.: Student enrollment, motivation and learning performance in a blended learning environment: the mediating effects of social, teaching, and cognitive presence. Comput. Educ. 136, 1–12 (2019) 17. Vo, H.M., Zhu, C., Diep, N.A.: The effect of blended learning on student performance at course-level in higher education: a meta-analysis. Stud. Educ. Eval. 53, 17–28 (2017) 18. Baragash, R.S., Al-Samarraie, H.: Blended learning: investigating the influence of engagement in multiple learning delivery modes on students’ performance. Telematics Inform. 35(7), 2082–2098 (2018)

Digital Support, Teacher Support, and Blended

401

19. Zhang, J.-H., Zou, L., Miao, J., Zhang, Y.-X., Hwang, G.-J., Zhu, Y.: An individualized intervention approach to improving university students’ learning performance and interactive behaviors in a blended learning environment. Interact. Learn. Environ. 28(2), 231–245 (2020) 20. Ramirez-Arellano, A., Bory-Reyes, J., Hernández-Simón, L.M.: Emotions, motivation, cognitive-metacognitive strategies, and behavior as predictors of learning performance in blended learning. J. Educ. Comput. Res. 57(2), 491–512 (2019) 21. Vo, M.H., Zhu, C., Diep, A.N.: Students’ performance in blended learning: disciplinary difference and instructional design factors. J. Comput. Educ. 7(4), 487–510 (2020). https://doi. org/10.1007/s40692-020-00164-7 22. Hill, T., Chidambaram, L., Summers, J.D.: Playing ‘catch up’ with blended learning: performance impacts of augmenting classroom instruction with online learning. Behav. Inf. Technol. 36(1), 54–62 (2017) 23. Yin, B., Yuan, C.-H.: Precision teaching and learning performance in a blended learning environment. Front. Psychol. 12, 631125 (2021) 24. Deschacht, N., Goeman, K.: The effect of blended learning on course persistence and performance of adult learners: A difference-in-differences analysis. Comput. Educ. 87, 83–89 (2015) 25. Karaboga, T., Erdal, N., Karaboga, H.A., Tatoglu, E.: Creativity as a mediator between personal accomplishment and task performance: a multigroup analysis based on gender during the COVID-19 pandemic. Curr. Psychol. 1–13 (2022) 26. Hew, T-S., Syed Abdul Kadir, S.L.: Understanding cloud-based VLE from the SDT and CET perspectives: Development and validation of a measurement instrument. Comput. & Educ. 101,132–149 (2016) 27. Standage, M., Duda, J.L., Ntoumanis, N.: A test of self-determination theory in school physical education. Br. J. Educ. Psychol. 75(3), 411–433 (2005) 28. Chin, W.W.: How to write up and report PLS analyses. In: Esposito Vinzi, V., Chin, W.W., Henseler, J., Wang, H., (eds.), Handbook of Partial Least Squares: Concepts, Methods and Applications, pp. 655–690. Springer, Heidelberg (2010)

The Research on Design and Application of Dynamic Mathematics Integrable Ware Design Model in Junior High School Jianlan Tang1(B) , Jerito Pereira1,2 , Shiwei Tan1 , and Tommy Tanu Wijaya2 1 Department of Mathematics and Statistics, Guangxi Normal University, Guilin 541004, China

[email protected], [email protected] 2 School of Mathematics Sciences, Beijing Normal University, Beijing, China

Abstract. In recent years, Internet + education has promoted the deep integration of mathematics education technology into mathematics teaching, and at the same time, information technology has played a very important role in teaching reform. The Ministry of Education of China released the Education Informatization Act 2.0, requiring the continued promotion of the deep integration of information technology and education. This research was conducted at Yanshan Middle School in Guilin City, The subjects of this study are all ninth grade students in the 2021– 2022 school year. Among them, there were 46 students in the experimental group and 47 students in the control group. The research design used in this study was non-equivalent group pretest-posttest design. This experiment is mainly divided into an experimental group and a control group to explore the teaching effect of Hawgent dynamic mathematics software in the teaching of graph and properties of quadratic functions. This research using a quantitative experiment so that the main tool for data processing, so the main using statistical software SPSS 25.0 The posttest questionnaire consists of 26 statements and is divided into two main parts, that is, the first part is using licket scale and contains statements number 1 to number 21. The research results show that: in the case that traditional classroom teaching cannot achieve, the use of dynamic mathematics integrable ware can significantly improve students’ learning effect, and has a positive impact on learning confidence, learning interest, and understanding. Keywords: Information technology · Dynamic mathematics · Integrable ware · Design model · Mathematics multiple representations

1 Introduction With the continuous development of "Internet Plus education (Wenxia 2020), The integration of information technology into mathematics education is not only a requirement of the times, but also a realistic demand. The "Education Informatization Act 2.0” issued by the Ministry of Education of China requires the continued promotion of the in-depth https://xueshu.baidu.com/scholarID/CN-B3HAFB1K and https://scholar.google.com/citations? user=gEccl60AAAAJ&hl=en © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 402–415, 2023. https://doi.org/10.1007/978-3-031-20429-6_37

The Research on Design and Application

403

integration of information technology and education, and encourages the two to promote each other; Further promote the application of new technologies such as big data, artificial intelligence, and virtual reality to real-life education and teaching. The “Key Points” also pointed out that the transformation and upgrading of education informatization and “Internet + education” are one of the important tasks of education now, and the ability to apply information technology is the core of teachers’ literacy in the new era (He 2021; Yan and Yang 2021). A large number of policy documents have been issued one after another, indicating that the integration of information technology into mathematics teaching is the trend of development in the new era, and it is also a new requirement put forward by the state for mathematics education. Information technology plays an important role in teaching reform, especially in mathematics teaching (Khalid Abdullah Bingimlas 2003). For some students, mathematics concepts are abstract and difficult to understand. In the traditional education model, teachers usually teach and students passively accept them, and the entire mathematics learning process is boring (Wijaya et al. 2020). The appropriate integration of information technology into mathematics teaching can not only help students deeply understand the content of mathematics learning, but also better grasp mathematics concepts. The understanding of concepts is the foundation of learning mathematics. Only on the basis of understanding concepts can students effectively apply knowledge and form correct cognitive schemes, so as to better solve practical problems. The emergence of dynamic mathematics software provides new tools and new teaching methods for teachers to carry out function teaching (Yates, Ellis, and Turk-Browne 2021). There are different types of dynamic mathematics software at home and abroad, such as network sketchpad, GGB, geometric sketchpad and Hawgent dynamic mathematics software (Gu, Huang, and Gu 2017; Martinovic and Manizade 2020). Among them, Hawgent is a dynamic mathematics software with a relatively complete and professional R&D and update team in China. It has a variety of icons, media, etc., which can make more exquisite teaching components, effectively attract students’ attention, and improve classroom efficiency. Teaching. At present, some research results have been produced on Hawgent dynamic mathematics software, but it is still insufficient for the development mode of Integrable ware. In the process of mathematics learning, functions are an indispensable key content. They are not only a key knowledge in junior high school mathematics learning, but also an important knowledge point in high school. Functions are abstract expressions of quantitative relationships in the real world. An important tool to help students develop abstract thinking. Students begin to systematically study the content of functions, such as linear functions, quadratic functions, etc., in the middle school stage, aiming to cultivate students’ symbolic awareness and reasoning ability. If only relying on the teacher’s dictation and drawing a rough quadratic function image on the blackboard, it is difficult to intuitively show the corresponding relationship between the accurate quadratic function expression and the image, and students will have difficulty in understanding, and even feel intimidated, and mathematics classroom will also It becomes boring and difficult to effectively help students master function-related knowledge. Therefore, how to enable students to effectively understand the properties and images of functions is the

404

J. Pereira et al.

“difficulty” and “pain point” for junior high school math teachers when teaching functions. Based on the above ideas, this study attempts to explore the deep integration of Hawgent dynamic mathematics software with quadratic functions. In order to improve the effectiveness of quadratic function teaching. The main purposes of this research are as follows: Firstly, to discuss the pattern of quadratic function graph and property design in junior high school based on state mathematics integrable ware parts; Thirdly, it discusses the effectiveness and application value of the quadratic function strategy based on the concept of multiple-representation learning and the design pattern of state mathematics integrable ware. The research questions are: 1. Based on the concept of mathematics multiple-representation learning, what are the design strategies for quadratic function graphs and properties integrable ware in junior high school? 2. What is the teaching effect of the quadratic function graph on mathematics achievement and properties integrable ware used in in junior high school classroom?

2 Research on Design Model of Dynamic Mathematical Integrable Ware In the 1960s, The British educator Dienes proposed the principle of multiple representation from the perspective of cognitive theory (Pantazi and Doukakis 2020). He believed that providing a variety of concrete physical figures representing a certain mathematical object could meet the cognitive styles of different students and help abstract mathematical structures from it. Briana L et al. proposed and proved that multiple representation is an important part of students’ mathematics learning. In particular, students’ ability to transform between symbols, figures, tables or oral functional forms is closely related to improving students’ ability to recognize each element. Different researchers have defined it from different perspectives. In this paper, the mathematical multiple representation is the representation of a mathematical object in different forms. Mathematical characterization of multiple representation form to include the characterization and visual words represent two different kind of essential attribute of characterization, among them, the words change characterized mainly includes life speaking, formal writing, and mathematics mathematical symbols, visual representation of a static presentation material object, model, and dynamic graphics and show animation, etc. (Çikla and In 2004). The forms of mathematical multiple representation are diverse and can provide verbal or visual representation for an abstract mathematical object through mathematical multiple representation. From the perspective of students’ cognition, on the one hand, multiple representations may increase students’ cognitive load; on the other hand, it is difficult for students to translate between different representations. Therefore, from the perspective of mathematical multiple representation is not enough to represent mathematical objects need multiple representation into mathematics learning theory, this paper argues that maths multivariate property index of learning is to learn multiple learning highly abstract and rigorous logic is based on the mathematical characteristics of the human brain information processing, as well as the characteristics of learning objects

The Research on Design and Application

405

for multiple representation so as to deeply analyze the nature of mathematical objects. Learn by multiple representation of mathematical objects, experience the richness of the multiple representation of mathematical object and relevance, thus to build good mathematics learning ecosystem, strengthen the transformation between the different characterization and translation ability, is likely to make the students deep mathematics study, so as to realize the overall promotion of students’ mathematics accomplishment. The integrable ware is the teaching material which is designed for a certain knowledge point and has certain teaching significance (Zhao, Zhang, and Yu 2021). Each section at the beginning of the classroom teaching in the teaching design, set up the teaching emphasis and difficulties, in the limited teaching key points and difficulties and time efficient study and teaching, can according to the characteristics of the product a short auxiliary teaching, the teaching difficult point for product design, using accumulates a dynamic, intuitive features, assist teachers crack teaching difficulty, to help students understand the knowledge. At present, scholar Yan Caiping uses Linglong drawing board to design eight senior high school mathematics integrations and applies them in teaching practice. He finds that the application of integrations in senior high school teaching has a positive impact on students’ cognitive beliefs, emotional attitudes and mathematical achievements. Thus, multiple representation based on mathematics learning theory design and optimization of integrable ware is feasible, when accumulate a design object as the mathematical knowledge, multiple representation based on mathematics learning theory to design and optimize product parts, can provide the abstract mathematical knowledge with conform to the diversity of students’ cognitive regularity of visual representation, not only intuitive present abstract knowledge structure from various angles, Furthermore, it highlights the essential characteristics of mathematical knowledge, dynamically displays the formation process of abstract knowledge and highlights the understanding of learning objects. Integrable ware is a kind of mathematics education software based on computer or CAI, which can be used by teachers and students in mathematics class, that is, multimedia teaching of mathematics (Caiying et al. 2012; Yu et al. 2019). Teachers can use math education software to teach math concepts to students and help students understand basic math concepts. The idea of integrable ware is a new development of computer-aided instruction, and also the theory and practice of multimedia instruction in information coding, retrieval, design, compilation, use, management and evaluation. Integrals have the following characteristics: primitives, that is, integrals resources are basic elements composed of each individual point, and each point is a primitive module with functions such as classification, retrieval and construction. Integrable ware, knowledge primitives in an integrable ware can be reorganized, accumulated, and alternated as needed. The common primitive information can be flexibly applied by different teachers in actual teaching activities, no matter how the curriculum system or text version changes. Open, through the way of primitive language storage can be integrated instrument material resources and teaching strategy resources, for teachers to recombine. Therefore, teachers or students can add and use the latest information or ideas to existing storage anytime and anywhere.

406

J. Pereira et al.

3 Multiple Representation Strategy Knowledge stores information in a schema and constitutes the learner’s cognitive structure. After the schema is constructed, the learned knowledge can be migrated and reorganized after a lot of practice, and then internalized into knowledge that the learner can apply. In this process Teachers should also try their best to reduce the memory load of students’ brain work through human intervention. From the point of view of the existing design ideas of teaching integrators, solutions around practical problems, sorting and collection of cognitive resources, and development of multiple-functional platforms are the main paths for the current research on the design ideas of integrators. Some subjects, such as physics, ideology and politics, and Chinese, have achieved certain research results, demonstrating the powerful role of integrable ware parts in practical classroom teaching. However, the design of the integrator resource library based on multiple representations is still in a blank state, and in terms of integrators for mathematics teaching, although many representation concepts have been applied to the design of integrators and micro-lecture teaching, but the A comprehensive description of the development and how to realize the design of Integrable ware from 0 to 1, how to progress from 1 to n, and how to grasp the sustainable development and application of Integrable ware Repository under the Multiple-Representation Theory are rarely relevant. The literature gives sufficient explanation. Based on the above considerations, this paper takes multiple representations as the theoretical basis, and on the basis of existing experience, proposes a new design model-APDDOI model for the resource package collection of integrators, as shown in Fig. 1. In terms of the hierarchy structure of design framework, the multi-representation theory includes three categories: service layer, path layer and task layer. The first is the service layer, which represents the main audience of the design resource base for designers, knowledge and users. The second is the path layer, which represents the complete process of product resource database design, and consists of “analysis- preparationdesign-development-optimization-integration”. Finally, there is the task layer, which explains the specific tasks that need to be performed under each path. Of multiple representation concept through from all process of the path to the task layer, the main definition, find, design, development, application, assessment and examination, not only ensure multiple representation theory guidance to the design of accumulates a repository, the embedded design ideas at the same time, to the effective implementation of the overall design process have the effect of self-monitoring, That is, whether each path or task is achieved is judged on the basis of whether multiple representation plays the corresponding six roles. The following is a further description of the design framework of the product repository under the multiple representation theory. The design idea of integrable ware parts is composed of six parts: analysispreparation-design-development-optimization-integration. Among them, the analysis stage includes the analysis of the teaching habits of the instructors, the learning characteristics of the trainees and the components of the teaching content, so as to ensure the pertinence of the early information. In the preparation stage, materials that may be used in active design should be collected on the basis of analysis, including learning behaviors in classroom teaching, and other auxiliary materials such as videos, pictures and sounds used to enrich the product. The design stage is the structure design, type design

The Research on Design and Application

407

Fig. 1. APDDOI integrable ware design model.

and frame design of integrable ware parts which belong to the teaching activity design, which is determined by the design sequence of “teaching → courseware → integrable ware parts”. The development phase begins with the creation of integrals, including integration of collected materials, active interface design, creation of interaction design for classroom activities, and storage of integrals. The next step is the optimization of integrated components. In the completion of any integrated component development, it is necessary to evaluate the feasibility of integrated components and test the application effect of integrated components in practice. Finally, it is the integration of integrated pieces. All integrated part that mach teaching needs are classified and sorted to form a resource package consisting of single integrated pieces, integrated pieces group (unit knowledge) and integrated pieces library (subject knowledge). In the normal educational elements, teachers (educators) are generally the leader, responsible for the development of class curriculum plan and coordination of the implementation of class teaching tasks; Take students (educates) as the main body of learning and undertake the responsibility of learning in learning activities; With teaching content and teaching tools as the intermediary system (educational influence), it undertakes the task of connecting teachers’ teaching with students’ learning.

4 Method and Research Framework The main assumptions of this experiment are as follows: (1) The use of dynamic mathematics software to teach the graph and properties of quadratic functions in junior high school can improve the teaching effect; (2) The use of dynamic mathematics software to study the graph and properties of quadratic functions in junior high school can stimulate students’ interest and motivation in learning. This research was conducted at Yanshan Middle School in Guilin City from November 15th, 2021 to December 15th, 2021. The subjects of this study are all ninth grade

408

J. Pereira et al.

students in the 2021–2022 school year. Among them, there were 46 students in the experimental group and 47 students in the control group. Independent variable: whether integrated Hawgent dynamic mathematics software integrable ware into the teaching of “graph and properties of quadratic functions”; Dependent variables: (learning interest, learning style, problem solving, thinking level, etc.) and result variables (study performance) that are generated in the teaching process. The research design used in this study was non-equivalent group pretest-posttest design. This experiment is mainly divided into an experimental group and a control group to explore the teaching effect of Hawgent dynamic mathematics software in the teaching of graph and properties of quadratic functions (Table 1). Table 1. Non-equivalent group pre test and post test design. Class

Pre-test

Experiment

Post-test

Control class

O1

–

O2

Experiment class

O3

X

O4

This research using a quantitative experiment so that the main tool for data processing, so the main using statistical software SPSS 25.0 The post-test questionnaire consists of 26 statements and is divided into two main parts, that is, the first part is using like art scale and contains statements number 1 to number 21 (Fig. 2).

Fig. 2. Research Design framework.

The Research on Design and Application

409

5 Result and Discussion 5.1 The Results and Analysis of the Pre-test on Student Achievement The instrument used in this experimental study was a test containing questions (pre-test and post-test). The research instrument was designed to measure student performance before and after the experiment. Table 2. Pre-test scores S-W test Kolmogorov-Smirnova

Class

Statistic

Shapiro-Wilk

df

Sig

Statistic

df

Sig

0.987

47

0.864

.969

46

0.246

Control class

0.085

47

0.200*

Experiment class

0.120

46

0.095

*. This is a lower bound of the true significance a. Lilliefors Significance Correction

Can be seen from Table 2 that after the scores of the pre-test of the experimental class are counted in the S-W test, the statistic is 0.969, the df value is 46, and the significant probability value is Sig = 0.246 (> 0.05). Similarly, the test statistic of the control class is 0.987 > 0.05, the df value is 47, and the significant probability value Sig = 0.846 (> 0.05), and the null hypothesis is accepted, there is no significant difference, and it can be considered that the experimental class and the control class are the first The test scores are all subject to normal distribution, and the independent sample T test can be carried out on the above data to verify whether the two classes are parallel classes. Further results are shown in Tables 3 and 4. Table 3. Comparison of pre-test scores between the two classes Class

N

Mean

Std. Deviation

Std.Error Mean

Control class

47

98.13

8.917

1.301

Experiment class

46

98.00

8.914

1.314

According to the above output Table 3 “Group Statistics”, the learning outcome data of the experimental class is 46 students, and that of the control class is 47 students. The average or mean of the learning results of the experimental class is 98.13, while that of the control class is 98.00. Therefore, it can be concluded statistically that the average learning results of the experimental class and students are different. Next, to prove whether the difference is significant (real), we need to interpret the output of the independent sample T-test. According to the above results, you can know the value of Sig. Levene’s Test for Equality of Variances is 0.653 > 0.05, which indicates that the variance of data between

410

J. Pereira et al. Table 4. Independent sample test

t-test for equality of means F

Sig

Equal variances assumed

0.203

t

0.653

Equal variances not assumed

df

Sig. (2-tailed)

Mean differ

Std. Error differ

95% Confidence interval of the difference Lower

Upper

0.069

91

0.945

0.128

1.849

– 3.545 3.801

0.069

90.958

0.945

0.128

1.849

– 3.545 3.801

the experimental class and the control class is homogeneous (Sujarweni 2014:99). According to the output table of “Independent sample test” in “Assumed equal variance”, Sig value is 0.653 > 0.05, and the conclusion can be drawn that the original hypothesis H0 is rejected and H1 is accepted. Therefore, it can be concluded that there is no significant difference between the average academic performance of students in the experimental class and the control class. Furthermore, from the output table above, the value of “Mean Difference” is 0.128 or 0.13. This value represents the difference between the average academic performance of students in the experimental class and that of students in the control class, or 98.13–98.00 = 0.13, and the difference is −3.545 –3.801 (95% confidence interval of the lower limit of the difference). 5.2 The Results and Analysis of the Post-test The tool instrument used in this experimental study are tests involving questions (pretest and posttest). The research tool was designed to measure student achievement after the experiment. Therefore, analysis prerequisites are tested prior to data analysis. Table 5. S-W test of the first post-test scores of the control class and the experimental class Kolmogorov-smirnova

Shapiro-wilk

Statistic

Df

Sig

Statistic

df

Sig

Control class

0.154

47

0.007

0.943

47

0.023

Experiment class

0.151

46

0.010

0.941

46

0.021

a. Lilliefors Significance Correction

As can be seen from Table 5, Sig = 0.023 ( .50), values that exceed the variance due to the error measure for that construct (Table 2). Finally, discriminant validity is also supported as the square root of the average variance extracted (AVE) for a construct is higher than any correlation with another construct (Table 3). Thus, all criteria for internal consistency and convergent and discriminant validity are met. 4.2 Analysis of the Structural Equation Model Examination of the results allows us to note that EE, FC, HM, PE, PV, and SI account for 61.6% of the variation in AR-learning BI (R2 = 0.616).

496

M. Benrahal et al. Table 2. Composite reliability, loadings, and convergent validity.

Constructs

Items

Loadings

Cronbach’s Alpha

AVE

CR

BI

BI1

0,858

0,863

0,709

0,907

BI2

0,818

BI3

0,834

BI4

0,857

EE1

0,896

0,873

0,797

0,922

EE2

0,891

EE3

0,891

FC1

0,776

0,792

0,705

0,877

FC2

0,883

FC3

0,857

HM1

0,929

0,884

0,812

0,928

HM2

0,931 0,918

0,859

0,948

0,831

0,740

0,895

0,9

0,834

0,938

EE

FC

HM

PE

PV

SI

HM3

0,841

PE1

0,897

PE2

0,951

PE3

0,932

PV1

0,786

PV2

0,919

PV3

0,871

SI1

0,864

SI2

0,945

SI3

0,929

Table 3. Discriminant validity. Constructs

1

2

3

4

5

6

BI

0,84

EE

0,7

0,893

FC

0,55

0,596

0,84

HM

0,61

0,777

0,626

0,901

PE

0,65

0,682

0,513

0,707

0,927

PV

0,44

0,45

0,452

0,403

0,33

0,86

SI

0,63

0,572

0,479

0,45

0,482

0,385

7

0,913

UTAUT2 Model to Explain the Adoption of Augmented Reality Technology

497

From Table 4, we can see that EE positively and significantly affects AR-learning BI (β = 0.274 and p = 0.012). PE positively and significantly affects the BI to adopt AR in education among Moroccan students (β = 0.251 and p = 0.019). SI positively and significantly affects the BI to adopt AR in education (β = 0.274 and p = 0.000). This led us to accept hypotheses H1, H2, and H3. However, the other three proposed hypotheses were rejected. Indeed, no link was found between PV, HM, FC, and the BI to adopt AR in education. Table 4. Hypothesis testing. Hypotheses

Std.beta

Std.error

t-value

P values

Decision

H1

0,251

0,107

2,356

0,019

Supported

H2

0,274

0,109

2,504

0,012

Supported

H3

0,274

0,072

3,797

0.000

Supported

H4

0,077

0,103

0,748

0,454

Not Supported

H5

0,02

0,13

0,151

0,88

Not Supported

H6

0,08

0,084

0,948

0,343

Not Supported

BI

R2

Q2

0,616

0,392

Implementing a cross-redundancy blindfolding [22] method allows us to obtain, all positive Stone-Geisser coefficient values. This is true for AR-learning BI (Q2 = 0.392). This finding allows us to argue that each of the structural equations in our model is satisfactory in terms of predictive relevance as shown in Table 4.

5 Discussion Regarding EE, our results establish an influence of EE on learners’ BI of AR in education. They are consistent with previous research findings [23]. Indeed, the learners’ PE positively affects learners’ BI to use AR in education. In the same vein, the work of [23] concluded that the BI to use AR applications is positively affected by the perceived usefulness of AR. It should be noted that SI has a stronger influence on learners’ BI to use AR in education, which is consistent with previous research [12]. Nevertheless, FC, HM, and PV were not found to be influential factors in learners’ BI to use AR in education. To some extent, there is consistency in our results with earlier research on the adoption of new technologies such as AR in education [12, 24], the adoption of mobile AR in Tourism [25], as well as the user behavior of AR technology [26].

6 Conclusion The main purpose of the present study was to study the acceptance of AR in education. By adopting the UTAUT 2, an adjusted questionnaire from prior studies was administered

498

M. Benrahal et al.

among a total of 100 participants consisting of current students in one of the public universities in Morocco. This paper is timely in meeting the necessity of adopting the 21st-century teaching and learning framework that was initiated recently by the Minister of Education. The intent of the current investigation was to examine the likely drivers behind the students’ acceptance of AR in education. In particular, we hypothesized that learners’ AI to use AR in education would be affected by EE, PE, and SI. Our work adds to the long list of works that have confirmed the UTAUT 2 model. Our study, in this context of the pandemic that imposes educational technologies on almost all educational actors, shows that AR technology is required to enhance the apprenticeship experience as it can serve as interactive learning cases. Using advanced methods such as fuzzy logic [27, 28] to study students’ behavior would be of great interest in future work leading to more comprehensive results. Funding The authors gratefully acknowledge the financial support and technical assistance provided by the Moroccan Association of Innovation and Scientific Research in Artificial Intelligence and Extended Reality, BP.154, Settat, Morocco. Acknowledgements. The authors gratefully acknowledge the financial support and technical assistance provided by the Moroccan Association of Innovation and Scientific Research in Artificial Intelligence and Extended Reality, BP.154, Settat, Morocco. Without its generous support, this publication would not have been possible.

References 1. Chien, Y.-C., Su, Y.-N., Wu, T.-T., Huang, Y.-M.: Enhancing students’ botanical learning by using augmented reality Univ. Access Inf. Soc. 18 (2017). https://doi.org/10.1007/s10209017-0590-4 2. Akçayır, M., Akçayır, G.: Advantages and challenges associated with augmented reality for education: a systematic review of the literature. Educ. Res. Rev. 20, 1–11 (2017) 3. Dunleavy, M., Dede, C., Mitchell, R.: Affordances and limitations of immersive participatory augmented reality simulations for teaching and learning. J. Sci. Educ. Technol. 18, 7–22 (2009) 4. Garzón, J., Pavón, J., Baldiris, S.: Systematic review and meta-analysis of augmented reality in educational settings. Virtual Real. 23(4), 447–459 (2019). https://doi.org/10.1007/s10055019-00379-9 5. Hincapie, M., Diaz, C., Valencia, A., Contero, M., Güemes-Castorena, D.: Educational applications of augmented reality: a bibliometric study. Comput. Electr. Eng. 93, 107289 (2021) 6. Arici, F., Yıldırım, P., Caliklar, S., Yilmaz, R.: Research trends in the use of augmented reality in science education: content and bibliometric mapping analysis. Comput. Educ. 142, 103647 (2019) 7. Venkatesh, V., Thong, J.Y.L., Xu, X.: Consumer acceptance and use of information technology: extending the unified theory of acceptance and use of technology. MIS Q. 36, 157–178 (2012) 8. Labti, O., Belkadi, E.: Factors affecting the online travel purchasing decision: an ıntegration of fuzzy logic theory. ˙In: Shakya, S., Balas, V.E., Haoxiang, W., Baig, Z. (eds.) Proceedings of International Conference on Sustainable Expert Systems, pp. 77–93. Springer (2021). https:// doi.org/10.1007/978-981-33-4355-9_7

UTAUT2 Model to Explain the Adoption of Augmented Reality Technology

499

9. Labti, O., Belkadi, E.: Modeling travelers behavior using FSQCA. ˙In: Abraham, A. et al., (eds.) Intelligent Systems Design and Applications, pp. 657–666. Springer International Publishing (2022). https://doi.org/10.1007/978-3-030-96308-8_61 10. Tamilmani, K., Rana, N.P., Dwivedi, Y.K.: Use of ‘Habit’ ıs not a habit in understanding ındividual technology adoption: a review of UTAUT2 based empirical studies. ˙In: Elbanna, A., Dwivedi, Y.K., Bunker, D., Wastell, D. (eds.) Smart Working, Living and Organising, pp. 277–294. Springer International Publishing (2019). https://doi.org/10.1007/978-3-03004315-5_19 11. Venkatesh, V., Morris, M.G., Davis, G.B., Davis, F.D.: User acceptance of information technology: toward a unified view. MIS Q. Manag. Inf. Syst. (2003). https://doi.org/10.2307/300 36540 12. Faqih, K.M.S., Jaradat, M.-I.R.M.: Integrating TTF and UTAUT2 theories to investigate the adoption of augmented reality technology in education: perspective from a developing country. Technol. Soc. 67, 101787 (2021) 13. Wang, Y.-S., Wu, M.-C., Wang, H.-Y.: Investigating the determinants and age and gender differences in the acceptance of mobile learning. Br. J. Educ. Technol. 40, 92–118 (2009) 14. Wu, Y.-L., Tao, Y.-H., Yang, P.-C.: The use of unified theory of acceptance and use of technology to confer the behavioral model of 3G mobile telecommunication users. J. Stat. Manag. Syst. 11 (2008) 15. Rauschnabel, P.A., Rossmann, A., Tom Dieck, M.C.: An adoption framework for mobile augmented reality games: the case of Pokémon Go. Comput. Hum. Behav. 76 276–286 (2017) 16. Sharif, A., Afshan, S., Qureshi, M.A.: Acceptance of learning management system in university students: an integrating framework of modified UTAUT2 and TTF theories. Int. J. Technol. Enhanc. Learn. 11, 201–229 (2019) 17. Shang, L.W., Siang, T.G., Zakaria, M.H. Bin & Emran, M.H.: Mobile augmented reality applications for heritage preservation in UNESCO world heritage sites through adopting the UTAUT model. In: AIP Conference Proceedigns, vol. 1830, pp. 030003 (2017) 18. Moorthy, K., Yee, T.T., T’ing, L.C., Kumaran, V.V.: Habit and hedonic motivation are the strongest influences in mobile learning behaviours among higher education students in Malaysia. Australas. J. Educ. Technol. 35 174–191 (2019) 19. Palau-Saumell, R., Forgas-Coll, S., Sánchez-García, J., Robres, E.: User acceptance of mobile apps for restaurants: an expanded and extended UTAUT-2. Sustainability 11, 1210 (2019) 20. Brislin, R.W.: Translation and content analysis of oral and written materials. In: Triandis, H.C., Berry, J.W. (eds.) Handbook of cross-cultural psychology (1980) 21. Chinn, W.W.: The partial least squares approach to structural equation modelling. Mod. Methods Bus. Res. (1998) 22. Hair, J., Hollingsworth, C.L., Randolph, A.B., Chong, A.Y.L.: An updated and expanded assessment of PLS-SEM in information systems research. Ind. Manag. Data Syst. 117, 442– 458 (2017) 23. Hadi, S.H., et al.: Developing augmented reality-based learning media and users’ intention to use it for teaching accounting ethics. Educ. Inf. Technol. 27, 643–670 (2021). https://doi. org/10.1007/s10639-021-10531-1 24. Chauhan, S., Jaiswal, M.: Determinants of acceptance of ERP software training in business schools: empirical investigation using UTAUT model. Int. J. Manag. Educ. 14, 248–262 (2016) 25. Pinto, A.S., Abreu, A., Costa, E., Paiva, J.: Augmented reality for a new reality: using UTAUT3 to assess the adoption of mobile augmented reality in tourism (MART). J. Inf. Syst. Eng. Manag. 7, 14550 (2022) 26. Mohd Nizar, N.N., Rahmat, M.K., Maaruf, S.Z., Damio, S.M.: Examining the use behaviour of augmented reality technology through MARLCardio: adapting the UTAUT model. Asian J. Univ. Educ. 15, 198 (2019)

500

M. Benrahal et al.

27. Bourhim, E.M., Cherkaoui, A.: Efficacy of virtual reality for studying people’s pre-evacuation behavior under fire. Int. J. Hum. Comput. Stud. 142, 102484 (2020) 28. Bourhim, E.M., Cherkaoui, A.: Usability evaluation of virtual reality-based fire training simulator using a combined AHP and fuzzy comprehensive evaluation approach. ˙In: Jeena Jacob, I., Kolandapalayam Shanmugam, S., Piramuthu, S., alkowski-Gilski, P. (eds.) Data Intelligence and Cognitive Informatics, pp. 923–931. Springer (2021). https://doi.org/10.1007/978981-15-8530-2_73

Question Guru: An Automated Multiple-Choice Question Generation System Abdul Rehman Gilal1,2(B) , Ahmad Waqas3 , Bandeh Ali Talpur4 , Rizwan Ali Abro2 , Jafreezal Jaafar1 , and Zaira Hassan Amur1 1 Department of Computer and Information Sciences, Universiti Teknologi PETRONAS, 31750

Seri Iskandar, Malaysia [email protected] 2 Department of Computer Science, Sukkur IBA University, Sukkur 65200, Pakistan 3 Knight Foundation School of Computing and Information Sciences, Florida International University, Miami, FL, USA 4 School of Computer Science and Statistics, Trinity College Dublin, Dublin, Ireland

Abstract. During the last two decades, natural language processing (NLP) puts a tremendous impact on automated text generation. There are various important libraries in NLP that aid in the development of advanced applications in a variety of sectors, most notably education, with a focus on learning and assessment. In the learning environment, objective evaluation is a common approach to assessing student performance. Multiple-choice questions (MCQs) are a popular form of evaluation and self-assessment in both traditional and electronic learning contexts. A system that generates multiple-choice questions automatically would be extremely beneficial to teachers. The objective of this study is to develop an NLP based system, Quru (Question Guru), to produce questions automatically from text content. The Quru is broken into three basic steps to construct an automated MCQs generation system: Stem Extraction (Important Sentences Selection), Keyword Extraction, and Distractor Generation. Furthermore, the system’s performance is validated by university lecturers. As per the findings, the MCQs generated are more than 80% accurate. Keywords: Keyword extraction · MCQs · NLP · Question generation · Automated questions

1 Introduction Recent advancements in e-learning have seen tremendous expansion around the world by allowing learners and institutions to customize the learning environment. Many organizations have their own online auto-assessing tools that help academics evaluate students’ performance. The purpose of the assessment is to determine the level of understanding that students gain from the given material, Students can benefit from the assessment since it helps them expand their knowledge and improve their cognitive development [1]. The assessment procedure is complicated in general, so it’s been divided into subjective and © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 501–514, 2023. https://doi.org/10.1007/978-3-031-20429-6_46

502

A. R. Gilal et al.

objective categories to make it more transparent. Subjective evaluation is a thinking process in which there are numerous ways to convey the proper response to a specific topic, However, objective assessment is the most preferable method when it comes to evaluating the learner’s performance. Moreover, in objective assessment, multiple-choice questions (MCQs) are a common kind of evaluation [2]. Meanwhile, the automated grading assessment can help learners with frequent responses and immediate feedback [3, 4, 22]. Multiple-choice questions (MCQs) are made up of a series of questions called items. Each item has a stem, which is a short paragraph that describes the question, these MCQs provide the number of alternative choices, and the choices can be three to four. In single response of MCQs, one of the choices is the correct answer and the wrong alternatives are called distractors [3, 23, 26]. For a question to give a valid evaluation, the number of components that make up the question must typically be large enough. As a result, creating MCQs is a time-consuming process, this process needs an automated system that helps to extract the frequent choices for the question. With the use of NLP, MCQ creation from text is a challenging but not impossible task variety of natural language processing libraries are available that can be beneficial for the MCQ creation system [3, 24]. The goal of this study is to develop a web-based application for the auto-generated MCQs that helps the instructor for the evaluation of students’ performance. This study will be achieved by implementing the concepts and techniques of natural language processing.

2 Automated Multiple-Choice Questions (MCQs) Generation Systems The "QURU" is a Web-based system that primarily assists teachers in automatically creating multiple-choice questions (MCQs) from a given article or paragraph. Users can upload material by copying it or saving it as a text file with the.txt extension. By passing through each stage, the system will provide multiple choice questions in response to the submitted content. Stem extraction, keyword extraction, and distractor extraction are the steps (Fig. 1). The stem extraction approach is a method of analyzing a user’s raw text to find suitable MCQ question phrases. One research employed a text summarizing strategy that employs the page rank algorithm to rank phrases [5, 25, 27]. We have utilized the top-ranked sentences as stems for our MCQ questions. The top ten sentences will be transferred to this module after the sentences have been rated. Each sentence will be given a keyword at this step. This keyword will become the answer to the MultipleChoice Question. The result of this step will be a correct answer for each ranking sentence (stem) that has been passed to it. For each keyword termed as “answer” to a single MCQ in the previous phase, we need a set of three distractors that will be used as alternatives in MCQ. There has been a lot of effort done on generating MCQs using domain vocabularies. Bongir et al. [6], Discuss how to generate MCQs by utilizing classbased, property-based, and terminology-based procedures to pick distractors. Moreover, Yahiya et al. [7], are using the methods and use fill-in-the-blank questions as the stem.

Question Guru: An Automated Multiple-Choice

503

Fig. 1. Showing the process of the system

The system used in the study is divided into three main phases, the first phase is to extract the term from a source text and parse the text by a parser. Each word of the source text is labeled with the parts of speech POS and its syntactic category by the parser. The nouns are sorted based on the frequency after the POS identification. The system works on certain rules and thresholds of frequency for each noun, the noun is regarded as a key term if it exceeds the threshold level. The second phase is to generate the stems from the source text by seeing the clauses of sentences [1, 6, 8]. The eligibility of a clause is considered if it follows the structure, and it must have subject-Verb and Object (SVO) or Subject Verb (SV). There are several rules defined in the system to generate the stems. The WordNet library is widely used by the system in the third phase of distractors selection to construct realistic distractors, and get hypernyms, and keyword coordinates from WordNet [6, 7]. Another study[9] provided a different strategy, based on rule mining and grammatical pattern matching that employed named entity recognition for sentence selection. One of the findings of the study [10] was the development of an effective method for selecting informative phrases and creating MCQs from them. They chose instructive sentences based on the importance of terms in defining the subject or issue, as well as parsing structural similarities. Their approach was based on manually designed patterns which were further used to find authentic sentences from the Web and were then transformed into grammatical test items. Distractors were also obtained from the Web with some modifications in manually designed patterns e.g., changing part of speech, adding, deleting, replacing, or reordering of words. The experimental results of this approach revealed that 77% of the generated MCQs were regarded as worthy (i.e., can be used directly). The disadvantage of this approach is that it requires a substantial amount of effort and knowledge to manually design patterns which can later be employed by the system to generate grammatical test

504

A. R. Gilal et al.

items. Furthermore another study [11] uses bloom taxonomy to extract the multiplechoice questions. They have determined the minimum degree of expertise necessary to answer the question. They performed an examination of the demands placed on student question-setters while designing MCQs in the second round of assessments, based on the competences that the question-setters needed to create their MCQs.

3 System Design and Development We developed the mechanism for the system that generates multiple choice questions automatically. The process is broken down into three phases: planning and design, implementation, and testing. Figure 2 presents the summary of the overall study design. Following sections discuss each phase in detail. The requirements for the systems are gathered throughout the planning and designing phase, which is referred to as requirement gathering. To continue with this procedure, we use the strategies listed below for acquiring requirements. In requirement collecting phase, we used requirement modelling to adapt and model the gathered needs [16–19]. This stage improved the quality of our work and clarified the specifications. QURU is built on a two-tier architecture (see Fig. 3), in which the client sends a request to the server through the internet, and the server responds to the request. In the below architecture client has variety of functions, like it can download the MCQs, Preview MCQs and can print the MCQs. However, Server is responsible to generate MCQs.

4 Implementation The system’s implementation is split into two stages: backend development and frontend web interface development. The QURU backend is the central resource on which the system generates MCQs from user-provided content. The task of creating MCQs is broken into three major parts. Stem Extraction (Important Sentences Selection), Keyword Extraction, and Distractor Generation. 4.1 Stem Extraction User data will be provided to this module which helps us to remove punctuations, citations, and other special characters. The main purpose of this module is to process user’s raw text to identify what are the potential sentences which can be used as a MCQs. We have used a similar approach as of text summarization which uses page rank algorithm technique to rank sentences. The outcome of this module will be ranked sentences by which we can use top ranked sentences as stem of our multiple-choice questions. For stem extraction, we have followed the preprocessing pipeline to remove the stop words. Stop words is a set of commonly used words in any language. Stop words can be distracting and non-informative or non-discriminative.

Question Guru: An Automated Multiple-Choice

505

Fig. 2. Methodology chart

After applying the stop words techniques, we have applied the word tokenizer. It is a process of splitting sentences into words. We have used vectors and Glove word embedding to convert the sentences into vectors representation. Later we have taken average of all word vectors in a sentence to form sentence vector (Figs. 4 and 5). Similarity matrix for sentence vectors will be generated based on Cosine Similarity. Cosine similarity is a metric used to measure how similar the sentences are irrespective of their size. The similarity scales between 0 and 1 (maximum). This matrix will have

506

A. R. Gilal et al.

Fig. 3. Architecture diagram

Fig. 4. Stop words removal

Fig. 5. Word vectors

similarity count for each sentence against every other sentence. This matrix will be used as input to next phase where graph will be generated and then ranked sentences will be popped out based on page rank algorithm. 4.2 Ranked Sentences Generation Matrix generated in the previous phase will be used to create graph. Sentences in matrix are taken as nodes of graph. Similarity count among sentences represents link between these nodes. This graph will then be used as input to page rank algorithm which gives ranked sentences as its output. 4.3 Keyword Extraction After getting ranked sentences, top ten sentences will be passed to this module. In this phase a keyword will be extracted against each sentence. This keyword will become answer to the Multiple-Choice Question. The outcome of this phase will be a correct answer against each ranked sentence (stem) passed to this phase.

Question Guru: An Automated Multiple-Choice

507

4.4 Key Identifiers Extraction In MCQs generally the answers are either number, proper noun, adjective, or noun. Key Identifiers are extracted from the text using POS-Tagger and stored in dictionary of candidate keywords. 4.5 Answer Generation for Each Stem For each ranked sentences system will distinctively choose a keyword from key identifier dictionary created in previous technique. Among the key identifier dictionary, noun is given the least priority to be chosen as keyword [12–15]. Since every sentence have maximum numbers of noun as compared to adjective, number, or proper noun. So, we have programmatically reduced the probability of choosing the noun as keyword. If there is no adjective, number, or proper noun then noun will be selected from that sentence as keyword. The chosen keyword for each ranked sentence will become the correct answer (Fig. 6).

Fig. 6. Highlighted sentences

4.6 Distractor Generation For each keyword termed as “answer” to a single MCQ in previous phase we need a set of three distractors that will be used as alternatives in MCQ (Fig. 7). 4.7 Word Vector for Answer As distractors of MCQs should be like answer in terms of meaning or understanding. We again need computation to generate misleading/confusing distractors. We used Glove Word Embedding to find 100d vector for each extracted keyword.

508

A. R. Gilal et al.

Fig. 7. Question preparation

4.8 Generating Distractors Based on the word vector extracted for each keyword. We have computed three nearest vectors from glove by using Euclidean Distance. These three vectors represent the options. By finding corresponding word to these vectors from glove. We get the options of our MCQS (Fig. 8).

Fig. 8. Distractor generation

Figure 9 shows the MCQ as an outcome of whole process. These types of questions will be generated by the system. We have manually tested our system on various articles taken from Wikipedia and paragraphs extracted from books to test our system on various domains. The results are quite satisfactory. Results have concluded that generated MCQs, stems are concrete meaningful sentences, alternatives are mostly number, proper noun and adjectives.

Question Guru: An Automated Multiple-Choice

509

Fig. 9. Output MCQ

5 Results and Discussion We personally tested our approach on a variety of topics collected from various sources. We picked any text at random and requested our faculty members to create MCQs from it. After that, we transmitted the same text to our system and compared the number of identical MCQs generated by the algorithm vs MCQs created by faculty members. The accuracy is calculated as matched number of MCQs divided by total number of MCQs generated by system. Moreover 10 faculty members participated into the survey and prepared the number of MCQs. The results in Table 1 have pointed out that quality of results generated by system are dependent on complexity of input text provided to system. Complex sentences and long sentences in paragraph produce poor results, while system performs quite well on paragraphs (Football and Computer) having simpler sentences. System is generating more than 50% results in most of the above cases. Results have also highlighted an interesting fact that the time taken to process a paragraph is directly proportional to the length of paragraph and varied with the amount of vocabulary used in it. Greater the number of distinct and uncommon words in paragraph more time system will take to process it. Furthermore, we have performed comparison with past studies, and results proved that our model provides state of the art results. Moreover, the survey has concluded that system is producing above average results in all measured aspects. The satisfactory level for Understandability and accuracy of the content produced by system is 60% as per survey results, this concludes that system can be useful for automatically making MCQs. Table 2 presents the comparison with past studies. Knot et al. [20], uses the sequential inference model for multiple-choice questions. The overall accuracy is 77.3%, Moreover, Moholkar et al. [21], selects the best possible multiple choice answer using long short term memory. The overall accuracy results are 77.4%. On the other hand, however, the experiments result of our study has gained the overall accuracy 80%. Table 3 shows the results of measured accuracy on different paragraphs examined by different teachers. Our assessment also includes a post survey based on questions of Interface Layout, Functionality of the system, Understandability of system, Efficiency and Maintainability. MCQs generated by the system have performed well on the various topics. The overall results show the acceptance of the system by the experts.

510

A. R. Gilal et al. Table 1. Quru results on various subject categories

No

Topic

Number questions highlighted by the experts

Total number of questions generated by QURU

Number of same questions identified experts and QURU

Percentage (%)

1

The road

6

5

3

60

2

Computer

8

5

4

80

3

Football

8

5

4

80

4

Football

8

5

3

60

5

The Atom

8

5

3

60

6

The Road

6

5

3

60

7

The Atom

6

5

2

40

8

The Atom

7

5

3

60

9

Computer

7

5

4

80

10

Computer

8

5

4

80

Table 2. QURU’s comparison with studies References

Topic

Type of Questions Model

Accuracy (%)

Khot et al. [20]

SCITAIL: a Multiple choice textual entailment questions dataset from science question answering

Sequential inference model

77.3

Moholkar et al. [21]

Multiple choice question answer system using ensemble deep neural network

Multiple choice questions

Ensemble Deep Neural Network (LSTM)

77.4

Our study

Question guru

Multiple choice question answers

Stem extraction Distractor generation

80

Question Guru: An Automated Multiple-Choice

511

Table 3. Overall results of multiple-choice questions generation system No Name

Cumulative results Very Dissatisfactory Average Satisfactory Very dissatisfactory (%) (%) satisfactory (%)

1

Interface Layout/Design

0

0

40

40

20

0

0

20

50

30

0

0

30

40

30

0

0

20

50

30

0

0

30

40

30

Understandability 0

0

20

60

20

The system provides simple, well organized, and readable interface 2

Navigation and Browsing Feature The system is easy to use and follow page to page

3

Attractiveness The system has excellent use of color, fonts, graphics, and effects

4

Functionality The system provides completeness of functions required

5

Creativeness The system has innovation of extra features

6

The system contents are relevant and accurate (continued)

512

A. R. Gilal et al. Table 3. (continued)

No Name

Cumulative results Very Dissatisfactory Average Satisfactory Very dissatisfactory (%) (%) satisfactory (%)

7

Robustness

0

0

20

50

30

0

0

20

50

20

0

0

40

50

10

0

0

30

40

20

Effort Expectancy 0

0

20

50

30

0

20

50

30%

0

10

60

30

The system does not crash and provides error recovery 8

Efficiency The system has good response time and page/graphics generation

9

Maintainability The system is easy to be extended and adapted for new requirements

10

Performance Expectancy The system is useful in job, and it can increase productivity

11

Interaction with the system is clear and understandable 12

Attitude Towards 0 Using Technology Interaction with this system becomes interesting

13

Overall of the system

0

Question Guru: An Automated Multiple-Choice

513

6 Conclusion and Future Work In this paper, we have presented an approach to generate Automated Multiple-choice questions (MCQs) using the Natural language processing (NLP) techniques. This automated generation system named as QURU. Our approach consisted of three main components, in the first component; stem extraction which is achieved through the techniques used in text summarization and page ranking algorithm. Stems are concrete meaningful sentence. The second component is keyword extraction which is achieved through POS-Tagger to identify the adjectives, nouns, proper nouns, and number of the MCQs, then the answer is generated for each stem. The third component is distractor generation through glove word embedding to find 100d vector then the three nearest vectors generated through Euclidean distance. The overall results show that the system satisfy the needs of users. Moreover, in the current system, we have trained the model based on the textual data only. Therefore, we aim to extend the model by considering the data from charts and tables. Currently, QURU only makes MCQs based questions using NLP techniques. In the future, we also plan to extend the QURU to perform descriptive assessments.

References 1. Das, S., Deb, N., Cortesi, A., Chaki, N.J.S.C.S.: Sentence embedding models for similarity detection of software requirements, vol. 2, no. 2, pp. 1–11 (2021) 2. Homma, Y., Sy, S., Yeh, C.: Detecting duplicate questions with deep learning. In: Proceedings of the International Conference on Neural Information Processing Systems NIPS (2016) 3. Sadhuram, M.V., Soni, A.: Natural language processing based new approach to design factoid question answering system. In: 2020 Second International Conference on Inventive Research in Computing Applications (ICIRCA), pp. 276–281: IEEE (2020) 4. Zhang, Y., Tuo, M., Yin, Q., Qi, L., Wang, X., Liu, T.J.N.: Keywords extraction with deep neural network model, vol. 383, pp. 113–121 (2020) 5. Rajpurkar, P., Jia, R., Liang, P.J.A.P.A.: Know what you don’t know: Unanswerable questions for SQuAD (2018) 6. Bongir, A., Attar, V., Janardhanan, R.: Automated quiz generator. In: The International Symposium on Intelligent Systems Technologies and Applications, pp. 174–188. Springer (2017) 7. Al-Yahya, M.: OntoQue: a question generation engine for educational assesment based on domain ontologies. In: 2011 IEEE 11th International Conference on Advanced Learning Technologies, pp. 393–395: IEEE (2011) 8. Godbole, A., Dalmia, A.., Sahu, S.K.J.A.P.A.: Siamese Neural Networks with Random Forest for detecting duplicate question pairs (2018) 9. Majumder, M., Saha, S.K.J K M, Journal, E.-L.A.I.: Automatic selection of informative sentences: the sentences that can generate multiple choice questions, vol. 6, no. 4, pp. 377–391 (2014) 10. Majumder, M., Saha„ S.K.: A system for generating multiple choice questions: with a novel approach for sentence selection. In: Proceedings of the 2nd workshop on natural language processing techniques for educational applications, pp. 64–72 (2015) 11. Yeong, F.M., Chin, C.F., Tan, A.L.J.P.A.I.J.: Use of a competency framework to explore the benefits of student-generated multiple-choice questions (MCQs) on student engagement, vol. 15, no. 2, pp. 83–105 (2020)

514

A. R. Gilal et al.

12. Hamed, S.K., Ab Aziz, M.J.J.J.C.S.: A question answering system on holy Quran translation based on question expansion technique and neural network classification, vol. 12, no. 3, pp. 169–177 (2016) 13. Huang, W., Qu, Q., Yang, M.J.N.C., Applications: Interactive knowledge-enhanced attention network for answer selection, vol. 32, no. 15, pp. 11343–11359 (2020) 14. Mohammad, A.S., Jaradat, Z., Mahmoud, A.-A., Jararweh, Y. J.I.P., Management.: Paraphrase identification and semantic text similarity analysis in Arabic news tweets using lexical, syntactic, and semantic features, vol. 53, no. 3, pp. 640–652 (2017) 15. Song X., Min, Y.J., Da-Xiong, L., Feng, W.Z., Shu, C.J.P.C.S.: Research on text error detection and repair method based on online learning community, vol. 154, pp. 13–19 (2019) 16. Gilal, A.R., Jaafar, J., Omar, M., Basri, S., Waqas, A.: A rule-based model for software development team composition: team leader role with personality types and gender classification. Inf. Softw. Technol. 74, 105–113 (2016) 17. Amin, A., Basri, S., Rehman, M., Capretz, L.F., Akbar, R., Gilal, A.R., Shabbir, M.F.: The impact of personality traits and knowledge collection behavior on programmer creativity. Inf. Softw. Technol. 128 106405 (2020) 18. Gilal, A.R., Jaafar, J., Capretz, L.F., Omar, M., Basri, S., Aziz, I.A.: Finding an effective classification technique to develop a software team composition model. J. Softw. Evol. Process 30(1), e1920 (2018) 19. Gilal, A.R., Jaafar, J., Omar, M., Basri, S., Aziz, I.A.: Balancing the personality of programmer: Software development team composition. Malays. J. Comput. Sci. 29(2), 145–155 (2016) 20. Khot, T., Sabharwal, A., Clark, P.: Scitail: a textual entailment dataset from science question answering. In: Proceedings of the AAAI Conference on Artificial Intelligence 2018, vol. 32, no. 1 21. Moholkar, K., Patil, S.H.: Multiple choice question answer system using ensemble deep neural network. In: 2nd International Conference on Innovative Mechanisms for Industry Applications (ICIMIA), vol. 5, pp. 762–766. IEEE (2020 Mar) 22. Almomani, M.A., Basri, S., Gilal, A.R.: Empirical study of software process improvement in Malaysian small and medium enterprises: the human aspects. J. Softw. Evol. Process 30(10), e1953 (2018) 23. Basri, S., Almomani, M.A., Imam, A.A., Thangiah, M., Gilal, A.R., Balogun, A.O.: The organisational factors of software process improvement in small software industry: comparative study. In: International Conference of Reliable Information and Communication Technology, pp. 1132–1143. Springer, Cham (2019) 24. Jameel, S.M., Gilal, A.R., Rizvi, S.S.H., Rehman, M., Hashmani, M.A.: Practical implications and challenges of multispectral image analysis. In: 2020 3rd International Conference on Computing, Mathematics and Engineering Technologies (iCoMET), pp. 1–5 IEEE (2020) 25. Gilal, A.R., Jaafar, J., Abro, A., Omar, M., Basri, S., Saleem, M.Q.: Effective personality preferences of software programmer: a systematic review. J. Inf. Sci. Eng. 33(6), 1399–1416 (2017) 26. Katper, S.H., Gilal, A.R., Alshanqiti, A., Waqas, A., Alsughayyir, A., Jaafar, J.: Deep neural networks combined with STN for multi-oriented text detection and recognition. Int. J. Adv. Comput. Sci. Appl. 11(4) (2020) 27. Gilal, A.R., Omar, M., Mohd Sharif, K.I.: Discovering personality types and diversity based on software team roles. In: International Conference on Computing and Informatics (ICOCI) (2013), pp. 259–264 (2013)

Importance and Implications of Theory of Bloom’s Taxonomy in Different Fields of Education Abdul Momen, Mansoureh Ebrahimi(B) , and Ahmad Muhyuddin Hassan Faculty of Social Sciences and Humanities, Academy of Islamic Civilization, Universiti Teknologi Malaysia, 81310 UTM Johor Bahru, Malaysia [email protected]

Abstract. The theory of Bloom’s taxonomy has offered a set of three hierarchical models for cognitive, affective, and psychomotor domains that are used for the classification of educational learning objectives into levels of complexity and specificity. Its cognitive domain as a classification system has been developed to categorize intellectual skills and behavior important to learning. Through its six stages, namely; remembering, understanding, applying, analyzing, evaluating, and creating, this theory emerges different applications from a lower degree to a higher degree. Scholars have examined the taxonomy’s main premise, which is that teachers want their students to better understand what they want them to know in an order from simple to sophisticated. The stages are supposed to be hierarchical, with mastery of one level necessitating progression to the next. Many researchers have identified learning objectives that involve higher levels of cognitive capabilities employing Bloom’s taxonomy, which focuses on the application of knowledge and abilities to a greater variety of activities and contexts. The data collection technique employed is qualitative method to investigate, and to draw conclusions based on analyzing and synthesizing sources of books and articles on Bloom’s Taxonomy. This study examines contribution of Bloom’s Taxonomy to evaluation, the teaching-learning process, classroom management, curriculum development, and its various applications in various fields of education. Improving the teaching-learning process and the assessment system at the level of secondary and higher education will give an idea of the application of this taxonomy in different educational aspects. Keywords: Bloom’s taxonomy · Importance · Applications · Education

1 Introduction Benjamin Bloom and his colleagues released The Taxonomy of Educational Objectives in 1956 as a tool for analyzing learning goals. Bloom’s Taxonomy is very well supervised learning model used by elementary and secondary school instructors, university and college learners, and scholars for many years. It provides a unified framework as well as a standard lexicon that educators and researchers can employ. Although some © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 515–525, 2023. https://doi.org/10.1007/978-3-031-20429-6_47

516

A. Momen et al.

have incorporated other categories, these taxonomies mainly concentrate on analytical thinking that belongs within the cognitive domain. It can be used in a variety of ways [1]. It’s used to determine whether stages of thinking skills are covered by existing instructional content such as learning objectives, curriculum plans, lessons, and evaluations. It can also be used as a guideline for creating educational materials to guarantee that the appropriate cognitive skills are met. It can also be used as a sort of evaluation. This taxonomy can be used in a wide range of educational settings, with the taxonomy being tailored to the particular subject of inquiry. Faculty members should have little issue creating instructional outcomes, customizing technology, and planning classrooms for each course provided they have a basic understanding of Bloom’s taxonomy and how to implement it to higher education. It can also be used by educators to break through the clutter of attention by assisting students in setting goals for themselves. This study examines its importance and application in different fields of learning.

2 The Learning Domains of the Bloom’s Taxonomy Bloom’s Taxonomy divides academic learning into three categories: cognitive, affective, and psychomotor. The cognitive domain contains mental abilities for knowledge production; the affective domain involves continuous emotional development of mindset; and the psychomotor domain involves physical skills. In 2001, a subsequent edition of the taxonomy adjusts the original version’s vocabulary and order of cognitive processes in light of domain findings. According to this reorganization, the ability to synthesize, rather than evaluate, has climbed to the forefront of the hierarchy [2]. Additionally, this alteration imparts a new perspective on all six cognitive abilities. In the diagram [3], the higher levels, the more sophisticated mental processes are allegedly necessary. Higher levels are not always preferable to lower levels, because higher levels cannot be attained without the capacity to use lower levels.

3 Importance and Applications in Different Fields of Education of This Theory No theory is absolute. There are some limitations in Bloom’s theory of taxonomy but the taxonomy of cognitive domain is applied in various segments of learning throughout the decades. One of the most valuable tools accessible to instructors is Bloom’s taxonomy. It provides learners with the understanding needed to construct effective evaluation procedures by delivering the answer key to how students learn. It came to a multi-layered answer to the question, which the writers added some explanations to as a teaching tool. • In a pedagogical exchange, it is important to set learning goals so that both teachers and students know what the exchange is all about. • The use of templates to define objectives can be beneficial to teachers. • Planning objectives assist in the clarification of goals for both teachers and pupils.

Importance and Implications of Theory

517

• Teachers can: “plan and provide suitable lessons”; “create legitimate assessment tasks and techniques”; and “confirm that lesson and evaluation are linked to the objectives” with the assistance of a well-organized set of objectives [4]. Bloom’s taxonomy of learning objectives is used to determine the extent to which an ability or performance has been learnt or mastered. The domain applications linked with each level are mentioned below. • A learner can define things at their knowledge level of learning. • A learner can work on specific issues and show off what they’ve learned at the level where they can understand things. • At the application level of learning, a learner identifies which approaches to employ and then applies those techniques to issues. • A learner can describe why the solving method works at the analysis level of learning. • At the synthesis level of learning, a learner can integrate parts of a procedure in new and valuable ways. • At the evaluation level of learning, a learner can come up with a lot of different ways to deal with the problem. Then, based on established criteria, the learner can choose the best way to deal with the problem. Bloom’s taxonomy is most typically used to define cognitive learning skills rather than psychomotor or affective capabilities, two categories that are crucial for the achievement of health professionals. It can be used by research scholars who teach or guide others to set learning objectives that define the talents and capabilities they expect their pupils to develop and demonstrate. Bloom’s taxonomy encourages the formulation of learning objectives that concentrate on the abilities that a student must acquire, which has implications for teaching and learning. According to Adams, this system allows for the integration of learning goals that require higher levels of cognitive skills into the system [5]. According to Yildirim and Baur, enhanced class structure, the use of a variety of techniques to lead all learners, increased awareness of learning philosophies, and better supervision of the outputs of instructional strategies are all part of a framework for constructing learning objectives. Problem-based coursework, team projects, and case study are examples of assessment methods that will enhance the learning process [6]. Moreover, according to Dubey, the domains in the original Bloom’s taxonomy can be used for educational reasons, such as developing quality question papers to confirm that students’ cognitive levels are categorized, quality control of a question paper to check up on an evaluation, and question designation for the advancement of automatic question-answering systems [7]. Using the six stages of Bloom’s original taxonomy, the learning procedures can be structured to meet the instructional objectives. Nkhoma et al. discovered that using the amended Bloom’s taxonomy in combination with case-based educational process boosted higher-order thinking. A higher practical assessment could result in more understanding, which could stimulate and motivate study [8]. Bloom’s taxonomy has acquired broad recognition within the assessment community due to its use at United Nations Educational, Scientific, and Cultural Organization (UNESCO) and Organization for Economic Cooperation and Development (OECD) sessions early after

518

A. Momen et al.

its publication. It’s also been used as a basis for curriculum review, exam development, and data analysis. Teachers and students must understand the cognitive domain’s hierarchy of processes and capabilities in order to identify learning skills and how these skills must be adjusted to understand more complex portions of the discipline-specific concept inventory. The enhancement of effective learning should never be assumed when teaching or studying new knowledge. Foundation courses should introduce learners to lower-level process skills, which should then be further improved in intermediate-level courses. Skills associated with higher-level techniques should be effectively implemented and retained in upper-division courses. Another technique to add depth to course exercises while also improving lifetime learning skills is to utilize effective learning skills from different process domains and clusters throughout the cognitive domain. Like the social domain, this aspect serves to remind individuals that their cognitive domain has advanced. Faculty can be much more particular and concentrate on the cognitive talents students must display in a certain course activity and on an evaluation using Bloom’s taxonomy. It also helps teachers figure out how other parts of their course, like subjects, lectures, assignments, and classroom activities, can be better to help students move from lower to higher levels of learning [9]. Bloom’s taxonomy supports not only the creation of highly explicit learning objectives and teaching materials in terms of language and definition, but also the assessment of such objectives. In the approach to programming, the process of organizing, educating, learning, and evaluating, there are various inquiries that look into the application of Bloom’s taxonomy to teaching and learning. This taxonomy is extremely useful for teachers in correctly implementing the curriculum and assessing students. For example: • Assessing a unit’s or syllabus’ objectives gives a clear, simple, visual depiction of the unit or syllabus that may be used to analyze the relative importance assigned to each goal. The Taxonomy’s incorporation of labeled, authentic instructional scenarios is one of its strongest features. It gathered accounts of ordinary teaching from teachers to help understand the Taxonomy’s divisions and categories, as well as how it may be used to examine objectives. • Assisting teachers in distinguishing between activities and objectives. The Taxonomy can be used to categorize the learning and teaching activities that were utilized to attain the objectives, as well as the evaluations that were used to indicate how effectively the students learned the objectives [10]. A noteworthy finding emerged from a review of the identified authentic teaching scenarios. Teachers listed actions rather than objectives when citing their unit’s outcomes. • Supporting instructors in understanding the link between evaluation and teaching activities. Airasian and Miranda present an excellent illustration of how teaching affects evaluation [11].

Importance and Implications of Theory

519

• Examining Curriculum orientation is the procedure of structuring three important parts in a classroom so that they are closely aligned or linked. These are the three aspects: (a) teaching and materials, (b) goals or objectives, and (c) assessments. The idea is that when classroom time and materials are aligned with objectives or guidelines, and these are aligned with examinations, students will learn more efficiently and effectively [12]. Curriculum arrangement is a procedure that enhances the coherence between the written, taught, and assessed curriculum. According to Anderson, the Taxonomy table can be a helpful method for determining curriculum concordance in all subject areas at almost any class or school level. The Taxonomy table can be used with any particular subject by combining themes with categories of knowledge [13]. • Implications for educators and the taxonomy serves to give a framework within which aspiring teachers and instructors can reflect not only how they teach but also how they evaluate and critique their own teaching. They should understand that the only way to assess the effectiveness of their lesson is to look at what students actually understand. As a result, the taxonomy steers aspiring instructors away from "best practice" teaching methods [14]. Within a competence-based curriculum, the Taxonomy functions well as a shareable content in syllabus discussions to reflect various sorts of cognitive capacity. Such arguments address both the curriculum and the students’ comprehension of it [15]. In these kinds of classrooms, teachers can use traditional methods to measure how well students have learned, or they can use student self-assessments, or even both [16]. Bloom’s taxonomy is especially fascinating in the context of student self-assessment because it relates to double-loop learning, [17] student self-responsibility and self-management, and the student-centered classroom [18]. Bloom’s taxonomy has also been used to look at things like effective learning, appropriate faculty interaction, and the long-term design and implementation of global experience-based learning in college teaching. McKeachie based his claim that students learn more when they talk than when they listen to lectures on his own studies [19]. This assertion inspired them to investigate how Bloom’s notions may be applied to establish a student-centered educational environment. Self-assessment and increased student participation, which courses are meant to encourage, are important for understanding (Table 1). Bloom’s Taxonomy can be used to determine a student’s level of performance in a class. A need to foster further improvement of critical thinking abilities, to ensure understanding with students about their levels of success in classes, and to increase student accountability in courses led us to the taxonomy, moving toward student selfmanagement. The taxonomy is useful for assisting students in taking greater responsibility for their learning and better understanding their critical thinking habits. Scaffolding is teaching that offers support to help the student learn for themselves, and it is used in this way to support metacognition. According to Hogan and Pressley, scaffolding is an instructional component that enables individual pupils with intellectual assistance so that they can operate at the forefront of their cognitive growth. In order to use Bloom’s taxonomy as a scaffolding tool, the student must first define the level of his or her effort.

520

A. Momen et al. Table 1. Practical tool to use of Bloom’s taxonomy for successful teaching.

Levels

Taxonomy

Outcomes

Level 1 Remembering Students recalling key information

Words (Action Verbs)

Questions

Arrange, Define, Describe, List, Match, Name, Order, Recall, Reproduce

(What, Who, Which) What can you remember about the event?

Level 2 Understanding Students demonstrate Classify, Discuss, knowledge of the Explain, Identify, issues acquired while Report, Summarize doing so

(Are you able to write in your own words…? Can you make a quick overview…? What are your thoughts on this….?) Summarize the event in your own words

Level 3 Applying

Focused with pupils’ ability to implement their skills and information to a variety of circumstances

Apply, Calculate, Demonstrate, Interpret, Show, Solve, Suggest

(Do you have any further examples of…? Is it possible that this happened in…? Can you adopt the approach to a personal experience…?) Suggest how the main lessons in this event could help other young people

Level 4 Analysing

Pupils’ ability to make links between ideas and to analyze critically

Analyse, Appraise, Compare, Contrast, Distinguish, Explore, Infer, Investigate

(What could have happened if…? How did this compare to…? What other results do you perceive as being possible?) Why did the different characters in the event behave the way that they did? (continued)

Students can then utilize the taxonomy to promote their own higher-level thinking as a result of this self-analysis. Many comprehensive courses, particularly those with critical thinking as a purpose, would seem to benefit from such a method [20].

Importance and Implications of Theory

521

Table 1. (continued) Levels

Taxonomy

Outcomes

Words (Action Verbs)

Questions

Level 5 Evaluating

Students can use their experience and wisdom to draw inferences, identify creative solutions to issues, and defend findings

Argue, Assess, Critique, Defend, Evaluate, Judge, Justify

(Is there a smarter method to…? Assess the worth of… ? Are you able to defend your viewpoint on…? How efficient are…?) Evaluate the strength of the main character’s decision to leave

Level 6 Creating

Students show what they’ve learned by making something new, whether it’s physical or abstract

Compose, Construct, Create, Devise, Generate, Organise, Plan, Produce

(Do you think you can come up with a solution to…? How would you manage with… if you had unlimited resources? Can you think of unique and unexpected ways to use…?) Rewrite the ending of this event, to show a different outcome

Bloom’s Taxonomy, when used in higher education, provides a consistent approach to teaching methods while also boosting students’ acquisition of subject-specific knowledge. In this exercise, Bloom’s Taxonomy is combined with adult education techniques, which enhances students’ learning and involvement in the classroom. This tool has been used in a number of college curricula. Positive outcomes have been achieved in each case. The assignment appears to guide students’ studying, organizing, and comprehension of course contents, based on their qualitative responses [21]. The use of Bloom’s taxonomy in the area of business information systems is wellknown around the world [22]. Students’ critical thinking abilities develop, and their understanding and application of theory to work improve as a result of the discussion sessions. Likewise, Ching and da Silva found that applying Bloom’s taxonomy to a business undergraduate program in a higher education institution was beneficial to both students and teachers [23]. Moreover, Britto and Usman demonstrated the utility of Bloom’s taxonomy through an investigation of several studies relevant to software engineering [24]. Bloom’s taxonomy was used in the course development and assessment, with four papers publicly acknowledging Bloom’s taxonomy’s value in software engineering education. Studies on the application of Bloom’s taxonomy in the subject of Electrical Engineering at a University of Technology in South Africa, comparable to the

522

A. Momen et al.

one in this assessment, were conducted during the preparation of curriculum and learner aids in the domain of Electrical Engineering. Academics should attend seminars analyzing learning goals on a regular basis, according to Meda and Swart, ‘to verify that the verbs employed in Bloom’s taxonomy are used effectively, resulting in increased student learning’ [25]. Swart and Daneti looked at the learning results of an electrical fundamentals curriculum that was presented in Romania and South Africa [26]. The usage of Bloom’s taxonomy model revealed that the learning results were dominated by the two bottom levels of the taxonomy. The remaining levels were found in the majority of the research, with the authors recommending that other colleges alter learner outcomes to align with Bloom’s taxonomy to encourage greater cognitive development. Bloom’s taxonomy is applied to a variety of courses in computer science education. When Michael V. Doran and David D. Langan talk about how to teach Computing I and II, they talk about a method that is based on how people think. This method includes clearly defining and planning course micro-objectives, assigning each micro-objective to a Bloom knowledge level, and helping students achieve and measure their goals [27]. According to Philip Machanick, Bloom’s taxonomy is used to build three separate computer science courses: Data and Data Structures, Algorithms and Artificial Intelligence, and Computer Architecture [28]. The Bloom taxonomy was tested by Johnson and Fuller to see if it was suited for computing [29]. Essi Lahtinen provides the findings of research on 254 undergraduate students in a basic programming course whose achievement was assessed on various levels of Bloom’s Taxonomy, as well as the outcomes of numerical cluster analysis, which recommends that the groups of students acquired align with Bloom’s taxonomy: underperforming students at lower levels may still perform well at higher levels of taxonomy. As a result, Lahtinen proposes that while structuring basic programming courses, the instructor should distinguish six kinds of pupils: capable students, practical students, poorly prepared students, theoretical students, memorizing students, and indifferent students. Theoretical students, for example, did well at the highest level (evaluation), but had a hard time with application and synthesis [30]. In terms of Bloom’s Taxonomy, Gardner and colleagues discovered the following Merge and Quick Sorts Learning Goals: (1) The learner will understand how to perform a fusion process. (2) The student will understand how to use the operations in order. (3) The student will understand the fusion process and quick sort Big O analysis. (4) The student will assess each of Kruse’s five sorting strategies [31]. Andrew Churches proposed a "digital" taxonomy for 21st-century schools and students. He rewrote the main verbs that indicate cognitive aims, combining creative and classic words. Because digital technology has become a vital component of education and other aspects of life, A. Churches’ modification allows Bloom’s pedagogical taxonomy to be extended by bringing it nearer to the present realities of human life [32]. Diana Cukierman and Donna McGee Thompson present the results of their research into the impact of integrating learning strategies teaching into standard laboratory time courses to aid student learning. Learning techniques and students were exposed to Bloom’s learning stages taxonomy and taught how this approach applies to course content in the first session.

Importance and Implications of Theory

523

4 Conclusion Finally, Bloom’s taxonomy for educational objectives can be regarded as having made a significant contribution to teaching and learning. Bloom’s Taxonomy is a useful framework for instructors to employ in order to concentrate on higher-order thinking and gives a hierarchy of levels. This taxonomy can help teachers create performance assignments, prepare questions for student consultations, and provide assessment. Educational psychologist Benjamin Bloom classified what and how humans learn into three distinct domains of learning in the mid-twentieth century. Content knowledge and the enhancement of human talents are included in the Cognitive Domain. Using the taxonomy at all levels of education, from kindergarten to college, and across different fields shows that it can be used in a wide range of different fields [33]. When critical thinking is used in higher levels of learning, the process of learning is both sequential and fluid. In other words, rather than being purely hierarchical, the cognitive action of assessing new knowledge, synthesizing valuable learning with other information, and then assessing the pieces of that and acquiring new knowledge is continual and interconnected. Bloom’s taxonomy is widely used in the field of education. In order to better comprehend this learning, each field must use and apply it. Bloom’s taxonomy has been used in a variety of fields, including biology, [34] management, [35] and music [36]. Bloom’s taxonomy is widely regarded as a well-defined strategy for instructors to use since it improves students’ understanding of and capacity to analyze, perform, and understand. It’s a way to organize educational goals in a way that makes it easier to improve educational instruction and come up with ways to measure educational progress. It has been used to learn in the fields of science, engineering, business, humanities, and social science. Many scholars use the technique to explicitly establish learning objectives, while others use it to evaluate students’ understanding and create learning materials and evaluations. Bloom’s taxonomy is fundamental knowledge for all educational and cognitive psychology programs. Though it was originally created as a simple evaluation tool, it has evolved into a symbol for curriculum planning, being used to define learning objectives and plan activities in the classroom. It has been developed for use in primary and secondary school classes, as well as in every academic area.

References 1. Marzano, R.J.: Designing a New Taxonomy of Educational Objectives, p. 95. Corwin Press, Inc, California, USA (2001) 2. Anderson, L.W., Krathwohl, D.R.: A taxonomy for learning, teaching, and assessing: a revision of Bloom’s taxonomy of educational objectives. Longmans, New York, NY (2001) 3. Santos, M.J., et al.: Compartmental learning versus joint learning in engineering education. Mathematics 9, 662 (2021). https://doi.org/10.3390/math9060662 4. Anderson, L.W., Krathwohl, D.R.: A Taxonomy for Learning, Teaching, and Assessing: A Revision of Bloom’s Taxonomy of Educational Objectives. Longman (2001) 5. Adams, N.E., MLIS,: Bloom’s taxonomy of cognitive learning objectives. J. Med. Lib. Assoc. 103(3), 152 (2015) 6. Yildirim, S.G., Baur, S.W.: Development of learning taxonomy for an undergraduate course in architectural engineering program. Am. Soc. Eng. Educ. 1, 1–10 (2016)

524

A. Momen et al.

7. Dubey, M.: Classifying stack overflow questions based on Bloom’s Taxonomy. Thesis. Indraprastha Institute of Information Technology Delhi (2016). https://repository.iiitd.edu. in/jspui/handle/123456789/431 8. Nkhoma, M., Lam, T., Richardson, J., Kam, K., Lau, K.H.: Developing case-based learning activities based on the revised Bloom’s taxonomy. In: Proceedings of Informing Science & IT Education Conference (InSITE), vol. 86 (2016) 9. Preville, P.: The professor’s guide to using Bloom’s taxonomy how to put America’s most influential pedagogical model to work in your college classroom, tophat.com, pp. 14– 15. file:///F:/Article%20Doc.%20File/Article%20on%20Bloom’s%20Taxonomy/sources/ Bloom’s/20.pdf 10. Krathwohl, D.: A revision of Bloom’s taxonomy: an overview. Theory Into Practice 41(4), 217 (2002) 11. Airasian, W., Miranda, H.: The role of assessment in the Revised Taxonomy. Theory Into Practice 41(4) 250 (2002) 12. Gorin, J., Blanchard, J.: The Effect of Curriculum Alignment on Elementary Mathematics and Reading. Paper presented at the 2004 Annual Meeting of the American Educational Research Association in San Diego, CA, p. 2. (2004) 13. Anderson, L.: Curricular realignment: A re-examination. Theory Into Practice 41(4), 258 (2002) 14. Byrd, P.: The revised taxonomy and prospective teachers. Theory Into Practice 41(4), 248 (2002) 15. Brownell, J., Chung, B.: The management development program: a competency based model for preparing hospitality leaders. J. Manag. Educ. 25(2), 124–145 (2001) 16. Spee, J., Tompkins, T.: Designing a competency-based master of arts in management program for mid-career adults. J. Manag. Educ. 25(2), 191–208 (2001) 17. Brookhart, S.: Successful students’ formative and summative uses of assessment information. Assess. Educ. Princ. Policy Pract. 8(2), 153–169 (2001) 18. Harvey, C.: Putting self-management into the classroom: one person’s journey. J. Manag. Educ. 22(4), 408–415 (1998) 19. McKeachie, W.: Teaching Tips: A Guidebook for the Beginning College Teacher, 8th edn. D. C. Heath, Lexington, MA (1984) 20. Hogan, K., Pressley, M.: Scaffolding Student Learning: Instructional Approaches & Issues (Advances in Teaching and Learning Series). Brookline Books, Cambridge, MA (1997) 21. Williams, A.E.: Promoting meaningfulness by coupling Bloom’s taxonomy with adult education theory: introducing an applied and interdisciplinary student writing exercise. Transform. Dialogues Teach. Learn. J. 10(3), 8 (2017) 22. Nkhoma, M., Lam, T., Richardson, J., Kam, K., Lau, K.H.: Developing case-based learning activities based on the revised Bloom’s taxonomy. In: Proceedings of Informing Science & IT Education Conference (InSITE), vol. 2016, pp. 86–87 (2016) 23. Ching, H.Y., da Silva, E.C.: The use of Bloom’s taxonomy to develop competences in students of a business undergraduate course. J. Int. Bus. Educ. 12, 107–126 (2017) 24. Britto, R., Usman, M.: Bloom’s taxonomy in software engineering education: a systematic mapping study. In: IEEE Frontiers in Education Conference (FIE) El Paso, TX, vol. 2015, pp. 1–8 (2015) 25. Meda, L., Swart, A.J.: Analysing learning outcomes in an Electrical Engineering curriculum using illustrative verbs derived from Bloom’s Taxonomy. Eur. J. Eng. Educ. 1–14 (2017) 26. Swart, A.J., Daneti, M.: Analyzing learning outcomes for electronic fundamentals using Bloom’s taxonomy. In: IEEE Global Engineering Education Conference (EDUCON), 9–11 April, vol. 2019, pp. 39–44 (2019)

Importance and Implications of Theory

525

27. Doran, M.V., Langan, D.D.: A cognitive-based approach to introductory computer science courses: lesson learned, presented at SIGCSE Technical Symposium on Computer Science Education (1995) 28. Machanick, P.: Experience of applying bloom’s taxonomy in three courses, presented at Southern African Computer Lecturers’ Association Conference (2000) 29. Johnson, C., Fuller, U.: Is Bloom’s taxonomy appropriate for computer science? presented at Baltic Sea Conference on Computing Education Research: Koli Calling (2006) 30. Lahtinen, E.: A categorization of novice programmers: A cluster analysis study, presented at Workshop of the Psychology of Programming Interest Group (2007) 31. Howard, et al., R.: Felder’s learning styles, Bloom’s taxonomy, and the Kolb learning cycle: tying it all together in the CS2 course, presented at SIGCSE Technical Symposium on Computer Science Education (1996) 32. Churches, A.: Bloom’s Digital Taxonomy (2008). Accessed 15 May 2020. https://www.res earchgate.net/publication/228381038_Bloom’s_Digital_Taxonomy 33. Bissell, A.N., Lemons, P.P.: A new method for assessing critical thinking in the classroom. Bioscience 56, 66–72 (2006). (Buxkemper, A., Hartfiel, D.J.: Understanding. Int. J. Math. Educ. Sci. Technol. 34, 801–812, 2003) 34. Crowe, A., Dirks, C., Wenderoth, M.: Biology in bloom: implementing Bloom’s taxonomy to enhance student learning in biology. CBE Life Sci. Educ. 7(4), 368–381 (2008) 35. Athanassiou, N., Mcnett, J., Harvey, C.: Critical thinking in the management classroom: Bloom’s taxonomy as a learning tool. J. Manag. Educ. 27(5), 555–575 (2003) 36. Hanna, W.: The new Bloom’s taxonomy: implications for music education. Arts Educ. Policy Rev. 108(4), 7–16 (2007)

Learning Chemistry with Interactive Simulations: Augmented Reality as Teaching Aid Mohamed Benrahal1 , El Mostafa Bourhim2,4(B) , Ali Dahane1 , Oumayma Labti3,4 , and Aziz Akhiate1 1 Artificial Intelligence and Complex Systems Engineering (AICSE), Hassan II University of

Casablanca, Ecole Nationale Supérieure Des Arts Et Des Métiers, ENSAM, Casablanca, Morocco 2 Industrial Engineering Department, EMISYS: Energetic, Mechanic and Industrial Systems, Engineering 3S Research Center, Mohammadia School of Engineers, Mohammed V University, Rabat, Morocco [email protected], [email protected] 3 Laboratory of Research in Management, Information and Governance, Faculty of Juridical Economic and Social Sciences Ain-Sebaa, Hassan II University of Casablanca, Route Des Chaux Et Ciments Beausite, BP 2634, Casablanca, Morocco 4 Moroccan Association of Innovation and Scientific Research in Artificial Intelligence and Extended Reality, BP.154, Settat, Morocco

Abstract. Augmented Reality (AR) has been identified by educational scientists as a technology with significant potential to improve emotional and cognitive learning outcomes. However, very few papers highlighted the technical process of creating AR applications reserved for education. The following paper proposes a method and framework for how to set up an AR application to teach primary school children the basic forms and shapes of atoms, molecules, and DNA. This framework uses the Unity 3D game engine (GE) with Vuforia SDK (Software Development Kit) packages combined with phone devices or tablets to create an interactive App for AR environments, to enhance the student’s vision and understanding of basic chemistry models. We also point out some difficulties in practice. As for those difficulties mentioned, a series of solutions plus further development orientation are put forth. Keywords: Augmented reality · Primary school children · 3D models · Unity 3D game engine · Vuforia SDK · Interactive simulations

1 Introduction In recent years, research on technology-enhanced learning (TEL) has been more focused on emerging technologies. AR, mobile learning (m-learning), serious gaming, and other technologies use learning analytics to improve user engagement and experiences in enhanced multimodal learning environments [1]. These studies make use of technological advancements in mobile device hardware and software, as well as their growing popularity among people, the considerable growth of user modeling, and personalization © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 526–535, 2023. https://doi.org/10.1007/978-3-031-20429-6_48

Learning Chemistry with Interactive Simulations

527

procedures that position the student at the center of the learning process. AR research, in particular, has advanced to the point that its applications can now be found in both mobile and non-mobile devices. AR studies have also shown themselves to be quite beneficial for enhancing student motivation in the educational process [2–6]. An AR system enables the combination of "supplementation" of the concrete world with virtual objects or overlaid information. As a result, virtual items appear to cohabit alongside real-world objects in the same place [7]. AR enables the seamless integration of virtual and real-world content [8]. This is opposed to the concept of a Virtual Environment (VE), in which the user is entirely immersed within a synthetic environment. Developing an AR application dedicated to education required a special process and software reserved for this kind of application. Tongprasom, K.et al., concentrated on an AR system for forensic medicine training in Thailand, where drowning is a leading cause of death [9]. The AR SDK type suited for use in constructing an AR system for teaching Forensic Medicine is chosen with the goal of comparative analysis and selection. The major goal of this study was to design and construct a prototype of an AR application in an educational context to demonstrate the capability of AR in education, particularly at the university level [10]. Many more papers present a comparative study about the different AR SDKs such as [11]. However, the overall papers discuss the use of AR in Education without highlighting the complete guide on how can we develop our AR applications starting from Application design, and 3D objects till the end of the process. The goal of this study was to respond to the aforementioned difficulties and how each step of the design works and how can we implement all the components and come up with usable applications. The rest of this work is structured as follows: In Sect. 2, we first outline related works of AR in education. Section 3 selects the suitable SDK as well as presents its different characteristics. Section 4 presents an overview of the application’s comprehensive structural architecture. Section 5 presents the case study we created and discusses some of the technical issues and potential solutions. Finally, in Sect. 6, we end our paper with a conclusion and future intentions.

2 Related Works AR is a prominent technology that has lately gained prominence in the educational field. A large number of research studies have demonstrated that AR may improve learning practice and teaching strategies. Ibanez, M. et al. Developed an AR application for teaching the fundamental principles of electromagnetism as an example of current AR uses in education [12]. Students can investigate the effects of a magnetic field in this application. The experiment’s components (cable, magnets, battery, etc.) may be detected using the camera of a mobile device, such as a tablet. As a response, students may use the tablet to view overlay information such as electromagnetic forces or circuit performance. According to the findings of this study, AR boosted academic attainment and gave fast feedback. Chien, Y et al. Have discovered that AR technology may be used to improve learning activities and students’ motivation and achievement [13]. University students can use AR technology to improve their laboratory abilities and establish positive attitudes about physics lab work [14]. Moreover Dunleavy, M. et al. states that AR’s most

528

M. Benrahal et al.

significant benefit is its ability to create immersive hybrid learning environments that blend physical and digital aspects, hence enhancing critical thinking, problem-solving, and communication growth [15]. A great number of systematic reviews in the literature have focused on the use, benefits, limitations, and challenges of AR technology in a variety of educational situations, providing useful insights into how AR technology improves student learning. For example Garzón, J. et al. present a meta-analysis of 46 research articles that use AR in education to define how pedagogical techniques affect AR’s impact on students’ learning goals and examine the impact of different moderating variables such as the education system, the learning method, and the intervention duration on the learning outcomes of students in AR interventions [16]. According to recent research Hincapie et al. demonstrate via a bibliometric study concentrating on the application of AR in the sector of education According to the findings of the survey, engineering education is the most prevalent educational research tendency, accompanied by simulation, gamification, tracking, and human-computer interface [17]. Likewise, no particular framework or design plan for the creation of AR content has been identified. Arici, F. et al. Suggest a content and bibliometric study of AR in scientific education scientific publications between 2013 and 2018 [18]. The papers investigated variables, data collection tools, method trends, sampling techniques, sample sizes, data processing methodologies, and sample populations. According to the content analysis results, the most studied factors in the research were “motivation,” “learning/academic achievement,” and “attitude”. Akçayır, M. et al. Consider doing a thorough study of the literature on AR’s educational uses [19]. The authors analyze numerous aspects in this review, including learner type, most often used AR technology, publication year, and the benefits and problems imposed by AR in educational settings. The review findings imply that AR has various benefits, including improving students’ learning motivation and performance, increasing satisfaction and positive attitude, and minimizing cognitive overload in learners. In terms of AR problems, the survey states that AR may be difficult to use, require more time, have technical faults, and have a negative attitude.

3 Vuforia Package for Unity The process we’ve been through to create the AR application in a mobile device uses the Vuforia software development kit. To make it works the steps are simple, the user must create an account for the registration and download the license key from Vuforia Website. Many reasons why developers choose to work with Vuforia among all the rest of the SDKs, First it’s free pricing as we mentioned, supports 2D and 3D recognition, tracks planar images in real-time, and recognition stability of immovable markers, maximum distance capture, cloud recognition, and many other functionalities. In general, Vuforia has two major categories: Objects, and images and Vumarks as the new generation of bar code [20, 21]. 3.1 Image Vuforia Engine can recognize and track content overlay over 2D pictures called image targets. By comparing the natural traits observed in the 2D photos to a known resource

Learning Chemistry with Interactive Simulations

529

database, the Vuforia Engine will recognize and track them. It will show material layered into 2D pictures once the image target is recognized. A multi-target is made up of many picture targets arranged in a certain geometric pattern. This target allows you to track all of the image targets on a multi-target object at the same time. Finally, Cylinder Targets allow pictures rolled into cylindrical and conical forms to be detected and tracked. The Vuforia target manager may employ JPG or PNG pictures in RGB or grayscale to build image targets, multi-targets, and cylinder targets. The submitted photos must be under 2 MB in size. 3.2 Object Object Recognition may be used to create rich and engaging 3D object experiences. It was created to operate with toys and other consumer items that can fit on a tabletop and are found indoors, such as enhancing 3D content to bring toys to life. Model targets enable the detection and tracking of certain objects in the real world based on their form by utilizing pre-existing 3D models. Model targets may be made from a broad range of things, including household gadgets and toys, automobiles, large-scale industrial equipment, and even architectural landmarks. Objects used as targets for object identification and model targets must not be flexible or malleable, and they must not have any shining surfaces. 3.3 Vumarks VuMark is the latest generation of bar code, which allows the user to customize unique designs and at the same time encode Data that will take place as trackable AR targets. VuMarks are capable to: • Present more than millions of unique instances. • Encode different data formats. • Track the Vumarks continuously. Vumarks are designed with Adobe Illustrator, once it is created in an SVG file, it can be directly uploaded to Target Manager in Vuforia as Vumark Template.

4 Unity Game Engine Among the numerous available technologies for 3D game production, Unity3D was chosen for our project because of its free development cost, ability to adapt a game to several platforms, extensive community, and the enormous amount of models and objects in the assets store. Two widely used Object-oriented programming languages (JavaScript and C#) are supported. According to Bourhim, E. M. et al., Unity GE lies as the best adaptable, cost-effective, and long-term solution for developing VR/AR applications [22]. As seen in Fig. 1, Unity3D is the fundamental system that links all other components and delivers applications to users.

530

M. Benrahal et al.

4.1 Overall Design In this part, we provide a practical, all-inclusive structural framework for developing an AR techniques application, and we describe how each component of the framework is integrated into Unity3D. As mentioned in Fig. 1, our framework contains four major components: 2D/3D Arts, the Vuforia package for Unity3D, the Unity3D GE, and the AR Application. Starting with 3D models, each AR or VR application needs 3D or 2D content, even designing it or free downloading it from 3D Contents websites or Unity asset store. Then move to the Vuforia developer portal, create an account, get a BASIC license manager, add a database with targets that we will be using in Unity 3D, import our FBX Models, and then follow the rest of the workflow shown in Fig. 1. 4.2 Material Contents Material contents are the most crucial components in our application for engaging and motivating students in the learning process. Creating it takes a long time to go through all the things, thus we came up to obtain freely available content without the requirement to reinvent the wheel. Some Molecules, RNA, are examples of models we import from the internet. Our resources are mostly derived from two sources: free 3D models and manually made models. There are several free 3D models accessible, such as on Sketchfab.com or in Unity’s assets store.

5 Case Study: XR Chemistry LAB The application is called XR Chemistry LAB, and it was developed within the XR the Moroccan association. The main purpose of our application is to provide small children with a geometric study of different chemical elements, as well as a briefing on Nucleic acids and Antibodies and their roles. In addition, to enable elementary kids to engage with virtual objects placed in the real-world environment around them for a better learning experience anywhere and at any time. Our application will be used in the context of real work manipulation in primary education. 5.1 Interface and Structure An activity diagram is created for each use case to illustrate the workflow performed by the user and application objects. In reaction to an event, each item does a separate action. The activity caused by the actions of other items: depicts a sequence of procedures for AR case in point Beginning with the user action of launching the application, the application will reply by displaying the main menu When the user chooses a mode, the program The mobile device camera is then launched. The customer then points the camera towards the marker to trace its location at the pointer. Following the successful tracking of the marker, the app will identify the marking and show the virtual item corresponding to the marker if the marker is not present, if the application is not recognized, then no virtual objects. In addition, a class diagram is created in line with the activity diagram concept.

Learning Chemistry with Interactive Simulations

531

Fig. 1. AR application production pipeline

After selecting the discovery mode, the AR camera is turned on then the system begins to receive frames from the actual scene the program looks for feature points in picture targets that are previously saved in the database for each frame. When the feature points are discovered, the program begins monitoring the picture target and superimposes a 3D object or model on top of it (Fig. 2). 5.2 Image Target When the user launches the program, the camera begins to capture frames of the realworld situation while looking for picture targets. Image targets are printed pictures that provide information about a chemical element; each target image contains the following info (see Fig. 3). In the middle of the card, we added a QR Code holding the name of the chemical to the image target to increase the number of feature points for improved tracking and detection. The cloud database of Vuforia allows us to observe feature points on each picture target (Fig. 4). 3D models of nucleic acids and the remaining molecules were free downloaded or built with unity editor or Blender and superposed to picture targets so that they appeared on top when the relevant image was identified (see Fig. 8). As long as the picture is visible, the 3D object remains tied to it. Totally or partially visible from the camera’s perspective (see Fig. 5).

532

M. Benrahal et al.

Fig. 2. XR Chemistry LAB interfaces

Fig. 3. Image Target

6 Conclusion In this article, a comprehensive framework for developing an AR application for educational purposes, as well as any other AR application in general, is described. And how to incorporate each phase of the process into the Unity 3D GE. We choose the optimal AR SDK as well as the suitable GE for use. We reviewed the framework’s problems, and

Learning Chemistry with Interactive Simulations

533

Fig. 4. Feature points in the example of our image target

Fig. 5. 3D Objects on the top of the image target (AR Camera)

several remedies were presented. This effort will leave a footprint to assist educators and anybody else interested in developing comparable apps. In the next work, we’ll utilize this case study to assess Moroccan users’ adoption of this technology in education, to resolve this big question. Should chemistry be taught in primary school? Funding The authors gratefully acknowledge the financial support and technical assistance provided by the Moroccan Association of Innovation and Scientific Research in Artificial Intelligence and Extended Reality, BP.154, Settat, Morocco. Acknowledgements. The authors gratefully acknowledge the financial support and technical assistance provided by the Moroccan Association of Innovation and Scientific Research in Artificial Intelligence and Extended Reality, BP.154, Settat, Morocco. Without its generous support, this publication would not have been possible.

References 1. Johnson, L F., Krueger, K.: New Media Consortium, & Consortium for School Networking. NMC horizon report (2015) 2. Liu, T.-Y., Chu, Y.-L.: Using ubiquitous games in an English listening and speaking course: Impact on learning outcomes and motivation. Comput. Educ. 55, 630–643 (2010) 3. Di Serio, Á., Ibáñez, M.B., Kloos, C.D.: Impact of an augmented reality system on students’ motivation for a visual art course. Comput. Educ. 68, 586–596 (2013)

534

M. Benrahal et al.

4. Jara, C.A., Candelas, F.A., Puente, S.T., Torres, F.: Hands-on experiences of undergraduate students in Automatics and Robotics using a virtual and remote laboratory. Comput. Educ. 57, 2451–2461 (2011) 5. Bujak, K.R., et al.: A psychological perspective on augmented reality in the mathematics classroom. Comput. Educ. 68, 536–544 (2013) 6. Chang, K.-E., et al.: Development and behavioral pattern analysis of a mobile guide system with augmented reality for painting appreciation instruction in an art museum. Comput. Educ. 71, 185–197 (2014) 7. Bourhim, E.M.: Augmented reality for fire evacuation research: an A’WOT analysis. In: Abraham, A., et al. (eds.) Intelligent Systems Design and Applications, pp. 277–285. Springer International Publishing (2022). https://doi.org/10.1007/978-3-030-96308-8_25 8. Azuma, R., Billinghurst, M., Klinker, G.: Special section on mobile augmented reality. Comput. Graph. 35, vii–viii (2011) 9. Tongprasom, K., Boongsood, W., Boongsood, W., Pipatchotitham, T.: Comparative study of an augmented reality software development kit suitable for forensic medicine education. IJIET 11, 10–15 (2021). (The School of Manufacturing Engineering, Suranaree University of Technology, Nakhon Ratchasima, 30000, Thailand) 10. Hanafi, A., Elaachak, L., Bouhorma, M.: A comparative study of augmented reality SDKs to develop an educational application in chemical field. In: Proceedings of the 2nd International Conference on Networking, Information Systems & Security, pp. 1–8. Association for Computing Machinery (2019). https://doi.org/10.1145/3320326.3320386 11. Bourhim, E.M., Akhiate, A.: Augmented reality SDK’s: a comparative study. In: Abraham, A. et al. (eds.) Intelligent Systems Design and Applications, pp. 559–566. Springer International Publishing (2022). https://doi.org/10.1007/978-3-030-96308-8_52 12. Ibanez, M.-B., Di-Serio, A., Villaran-Molina, D., Delgado-Kloos, C.: Augmented realitybased simulators as discovery learning tools: an empirical study. IEEE Trans. Educ. 58, 208–213 (2015) 13. Chien, Y.-C., Su, Y.-N., Wu, T.-T., Huang, Y.-M.: Enhancing students’ botanical learning by using augmented reality. Univ. Access. Inf. Soc. 18, 231–241 (2019) 14. Akçayır, M., Akçayır, G., Pekta¸s, H.M., Ocak, M.A.: Augmented reality in science laboratories: the effects of augmented reality on university students’ laboratory skills and attitudes toward science laboratories. Comput. Hum. Behav. 57, 334–342 (2016) 15. Dunleavy, M., Dede, C., Mitchell, R.: Affordances and limitations of immersive participatory augmented reality simulations for teaching and learning. J. Sci. Educ. Technol. 18, 7–22 (2009) 16. Garzón, J., Kinshuk, Baldiris, S., Gutiérrez, J., Pavón, J.: How do pedagogical approaches affect the impact of augmented reality on education? A meta-analysis and research synthesis. Educ. Res. Rev. 31, 100334 (2020) 17. Hincapie, M., Diaz, C., Valencia, A., Contero, M., Güemes-Castorena, D.: Educational applications of augmented reality: a bibliometric study. Comput. Electr. Eng. 93, 107289 (2021) 18. Arici, F., Yildirim, P., Caliklar, S, ¸ Yilmaz, R.M.: Research trends in the use of augmented reality in science education: content and bibliometric mapping analysis. Comput. Educ. 142, 103647 (2019) 19. Akçayır, M., Akçayır, G.: Advantages and challenges associated with augmented reality for education: a systematic review of the literature. Educ. Res. Rev. 20, 1–11 (2017) 20. Getting Started|VuforiaLibrary. https://library.vuforia.com/ 21. Best Augmented Reality SDK for AR development in 2018–2021. Thinkmobiles (2017). https://thinkmobiles.com/blog/best-ar-sdk-review/

Learning Chemistry with Interactive Simulations

535

22. Bourhim, E.M., Cherkaoui, A.: Selection of optimal game engine by using AHP approach for virtual reality fire safety training. In: Abraham, A., Cherukuri, A.K., Melin, P., Gandhi, N. (eds.) Intelligent Systems Design and Applications, vol. 940, pp. 955–966. Springer International Publishing (2020)

Upskilling Educators for Pandemic Teaching: Using Video Technology in Higher Education Chee Heong Lee1(B) , Pek Hoon Er2 , Tiny Chiu Yuen Tey3 Priscilla Moses4 , Phaik Kin Cheah5 , and Tat-Huei Cham6

,

1 Centre for Foundation Studies (Kampar Campus), Universiti Tunku Abdul Rahman, Perak,

Malaysia [email protected] 2 Faculty of Accountancy and Management, Universiti Tunku Abdul Rahman, Selangor, Malaysia [email protected] 3 Higher Education Research Institute, University of Sanya, Sanya, China 4 Faculty of Creative Industries, Universiti Tunku Abdul Rahman, Selangor, Malaysia [email protected] 5 Faculty of Arts and Social Science, Universiti Tunku Abdul Rahman, Perak, Malaysia [email protected] 6 UCSI Graduate Business School, UCSI University, Kuala Lumpur, Malaysia

Abstract. The popularization of technology in the higher education environment has raised awareness about the importance of educators’ technological knowledge for teaching and learning. The Technological Pedagogy Content Knowledge (TPACK) framework explains the set of knowledge that educators need to be equipped with for enhanced pedagogies. This will help them to adapt to the expectations of their digital native students. From a training needs analysis conducted in a higher education institution in Malaysia, a two-day training was conducted for the institution’s lecturers to cultivate their use of video technology based on the TPACK framework. The training facilitators evaluated the lecturers’ performance, and the results of the evaluation were presented in this study. Results show that the training successfully closed the identified training gap, and enhanced the lecturers’ capabilities in using video technology for teaching and learning purposes across various disciplines. Keywords: TPACK · Higher education · Training · Technology use · Teaching and learning

1 Introduction The use of technology has become a vital skill for educators in 21st-century teaching and learning. Technology has also been recognized as a powerful solution to urgent educational challenges such as the global pandemic [1], which has further accelerated educational advancement [2, 3]. However, this was burdensome to educators who perceive technology as opaque and protean [4, 5]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 536–545, 2023. https://doi.org/10.1007/978-3-031-20429-6_49

Upskilling Educators for Pandemic Teaching

537

While technology in education is devised to facilitate knowledge transactions in academic settings, educators and students might not process knowledge and information similarly [6, 7]. According to a previous study [8], digital transformation has greatly impacted the educational environments in which students (also known as digital natives) learn differently from their educators (digital immigrants). It was explained that educators are indeed experts in their content areas, but are not well-trained on how to incorporate technology for teaching purposes [2]. This is consistent with past research [8] that many university lecturers are knowledgeable and skilled in their subject fields, yet professional development activities are needed to help them augment their experience with technology. It had been pointed out previously that educators were not provided with sufficient training and their experience with digital technologies for teaching purposes was dearth [9]. It is also not an unusual issue in developed higher education contexts like Sweden, where educators also demand professional training on using digital tools for pedagogical enhancement [10]. In the context of this study, a training need analysis was conducted in a higher education institution to identify training gaps and needs of the lecturers. The analysis found that the lecturers required training on technology use and teaching digital natives. The lecturers had also indicated their desire to learn how to create and present contents using videos to help them further develop and adapt e-learning materials. The lecturers also showed interests in exploring more features and techniques on Microsoft PowerPoint to enrich lesson presentations. To remedy this gap, a two-day training workshop was conducted to guide lecturers in identifying suitable strategies to teach digital natives using videos and advanced features of Microsoft PowerPoint for lesson delivery. Following the workshop, the researchers were interested to evaluate to what extent the technology knowledge and skills that were taught could be used effectively by the lecturers for teaching purposes. Hence, facilitators of the workshop evaluated the lecturers’ lessons which incorporated technology components taught during the workshop (videos and PowerPoint features) using the Technological Pedagogical Content Knowledge (TPACK) framework.

2 Literature Review The TPACK framework was introduced by Koehler and Mishra [4] as a powerful framework that conceptualizes how educators use technology to support teaching and learning. It was proposed that there was a gap in the field of educational technology as technology was discussed independently from teaching and learning [11]. Hence, the TPACK framework extended Shulman’s idea of Pedagogical Content Knowledge by emphasizing technology integration in educators’ teaching. The TPACK framework addresses the gap by emphasizing technology components and how educators incorporate technological knowledge into their pedagogical and content knowledge [4, 11]. The framework has since exuded great influence not only in educational research and practice but also in professional development for educators [11]. It is also known as a practical guide to the development of research and program for its emphasis on the integration of technology into pedagogy and technology [12].

538

C. H. Lee et al.

As illustrated in Fig. 1, the TPACK framework consists of three primary domains and four intersected domains of the primary forms [4]. The primary domains are Technological knowledge (TK), Content Knowledge (CK), and Pedagogical Knowledge (PK). TK refers to educators’ knowledge about how they can use technology, tools and resources productively and continually on a regular basis [13]. CK is defined as educators’ knowledge about the contents to be taught and learnt, whereas PK is referred to as educators’ profound knowledge about the processes related to teaching and learning [13]. As highlighted in the TPACK framework, knowledge types that overlap between three primary domains are Technological Content Knowledge (TCK), Pedagogical Content Knowledge (PCK), Technological Pedagogical Knowledge (TPK), and Technological Pedagogical Content Knowledge (TPACK). TCK describes the knowledge of how technology can create changes or new representations for the subject matter or vice versa [11, 13]. It was also explained in [13] that PCK indicates educators’ knowledge in terms of how pedagogical approaches can be adapted appropriately in teaching. Besides, TPK denotes how teaching and learning may change when various technologies are used in teaching based on pedagogical designs and strategies [11, 13]. TPACK is the central intersection of the three primary domains in the framework. Hence, it is the combination of TK, CK, and PK that addresses educators’ integrated knowledge of technology, content, and pedagogy into their teaching in any content area [11, 13].

Fig. 1. Technological, Pedagogical Content Knowledge (TPACK) [14].

Upskilling Educators for Pandemic Teaching

539

Due to the uniqueness of the TPACK framework, it has been widely used in various educational contexts, especially in training for educators from secondary to tertiary levels [3, 5, 15]. In a previous study, selected domains of TPACK were used to provide technology training for university lecturers to help them better relate the technology knowledge to actual contexts and practice how they can use the technology skills in their respective content areas [16]. This study focuses on the technology domains of the TPACK framework to evaluate lecturers’ skills in using technology namely videos and PowerPoint, in designing and presenting their lessons. Therefore, TK in this study refers to lecturers’ knowledge in preparing and editing video lesson, while TCK indicates lecturers’ knowledge in preparing and editing video lesson which fits the lecture content. TPK is referred to as the lecturers’ knowledge in preparing and editing video lesson that matches their respective effective pedagogy, whereas TPACK denotes lecturers’ knowledge in preparing and editing video lesson which fits the lecture content and matches with effective pedagogy.

3 Methods This study involved 21 participants from a higher education institution in Malaysia, of which 19 were males and two were females. Among the participants, 18 were identified from a training needs analysis conducted earlier, whereby these lecturers indicated the need to cultivate the use of technology for effective teaching. The remaining three lecturers had volunteered themselves to attend the training. The training involved two experienced facilitators from the research team who also evaluated the lecturers’ performance using the TPACK framework at the end of the training. On the first day of the training, the participants were briefed on the digital natives and how the digital natives learn. They were also taught about approaches to teach the digital natives and the challenges they may face. Subsequently, the participants were trained in creating and editing videos such as trimming the video, inserting and formatting the photos and captions inside the video, animating the objects inside the video, applying effects on the video and exporting the video into different file formats and qualities using Easy Movie Maker freeware available from Microsoft Apps store. In addition, the participants were also trained on creating narrated PowerPoint lecture slides, embedding and trimming embedded videos, and screen recording to create video lectures using features available on Microsoft PowerPoint. At the end of the first day, the participants were assigned to prepare a teaching plan based on the technology domains of the TPACK framework. They were given the freedom to decide on how they wanted to create the educational videos such as to adapt existing videos with some editing and formatting, to self-author or self-record video based on the video creating and editing skills learnt in training, or to do screen recording using PowerPoint slides. The two training facilitators evaluated the teaching plans and mock teachings making use of the educational videos created. They also provided feedback right after each presentation. The facilitators’ evaluations were done using the TPACK framework on the second day of the workshop.

540

C. H. Lee et al.

4 Results and Findings Results presented in Table 1, 2, 3, 4 and 5 are based on the evaluation of the teaching plans and the videos presented by the lecturers on a five-point-Likert scale evaluation ranging from strongly disagree (1) to strongly agree (5). Results are supported by facilitators’ feedback based on their observations of the lecturers’ performance during the mock teaching. Table 1. Descriptive statistics of the measured TPACK dimensions. TPACK domain

Min

Max

M

SD

Technological Knowledge (TK)

1.83

4.33

3.468

0.499

Technological Content Knowledge (TCK)

2.88

4.38

3.643

0.426

Technological Pedagogical Knowledge (TPK)

2.50

4.38

3.482

0.493

Technological Pedagogical Content Knowledge (TPACK)

2.00

4.50

3.286

0.604

Table 1 shows the four technological dimensions of the TPACK framework that were evaluated by the two facilitators based on the presented videos. Based on the results, it was noticed that the training facilitators scored highest for the lecturers’ technological content knowledge (M = 3.643, SD = 0.426). This is followed by their TPK (M = 3.482, SD = 0.493), TK (M = 3.468. SD = 0.499), and TPACK (M = 3.286, SD = 0.604). Overall, the lecturers were competent at identifying the topics or contents appropriate to be presented using video and have the ideas on the best way to utilize the videos in their lessons. However, they were less competent to materialize their ideas and plans as they did not have sufficient knowledge and skills to identify and use the suitable technology such as utilizing video to craft an effective teaching and learning session. In order to precisely determine the areas of knowledge and skills in creating an educational video that the lecturers had mastered, each of the technological dimensions of the TPACK framework was analyzed in detail. Table 2. Descriptive statistics of technological knowledge (TK). Item

Description

M

SD

TK1

The lecturer has good video making skills

3.619

0.610

TK2

The lecturer has good video editing skills

3.333

0.577

TK3

The lecturer is able to add in additional elements (e.g., annotations, narrative texts, quiz, simulation) into the video instead of narrating the content of PPT slides

3.452

0.498

Upskilling Educators for Pandemic Teaching

541

Table 2 shows the results of lecturers’ TK as evaluated by the facilitators. The training facilitators somewhat agreed that, on average, the lecturers demonstrated good video making skills (M = 3.619, SD = 0.610). The lecturers acquired fairly good skills in adding elements and features such as captions, transition effects and background music instead of just adopted the video or just narrated the slides (M = 3.452, SD = 0.498). Meanwhile, the lecturers’ video editing and creating skills need to be sharpened further (M = 3.333, SD = 0.577), whereby some of the videos did not start or end timely or the effects applied were not ideal. Table 3. Descriptive statistics of technological content knowledge (TCK). Item

Description

M

SD

TCK1

The subject content is suitable to be presented by way of video lesson 4.024

0.295

TCK2

The video lesson has effectively presented the subject content in a clear manner

3.548

0.568

TCK3

The selected topic is appropriate to be presented within the allotted time

3.833

0.555

TCK4

The additional elements used (e.g., narrative texts, quiz, simulation) enhance the presentation of the subject matter

3.167

0.599

In terms of the lecturers’ TCK, it is shown in Table 3 that the facilitators agreed (M = 4.024, SD = 0.295) that the lecturers were good at selecting topics that were suitable to be presented through video and within the allocated time limit (M = 3.833, SD = 0.555). Besides, it was agreed that the video lessons created could present the selected topics effectively (M = 3.548, SD = 0.568) to a certain extent. However, the lecturers needed to pay attention to the additional elements or effects used in their videos (M = 3.167, SD = 0.599). For instance, some lecturers had added strong background music that interfered with the presentation of the topic in the videos, while many of the videos did not add captions or indicators to highlight the area of focus while narrating the content, which usually causes difficulty for students to follow. Table 4. Descriptive statistics of technological pedagogical knowledge (TPK). Item

Description

M

SD

TPK1 The way of presenting the subject content through the video lesson enhances student’s learning

3.476 0.642

TPK2 The subject content is presented in an organized manner

3.929 0.576

TPK3 Appropriate elements (e.g., graphical representation, narrative texts, 3.214 0.561 simulation) have been employed to enhance the teaching of the subject content TPK4 The presentation of the subject content is able to engage the audience

3.310 0.512

542

C. H. Lee et al.

Based on the facilitators’ evaluations shown in Table 4, facilitators scored the lecturers highest in TPK2 for TPK. The result suggests that the facilitators agreed (M = 3.929, SD = 0.576) that the lecturers were competent to present the subject content in an organized manner either through the screen-recorded video lesson or adapted videos. This is followed by TPK1 (M = 3.476, SD = 0.642) and TPK4 (M = 3.310, SD = 0.512), in which the facilitators fairly agreed that the lectures were able to present subject content using video lessons to enhance student learning and engage the audience respectively. Besides, the facilitators opined that there was room for improvement in terms of the use of appropriate elements (M = 3.214, SD = 0.561) such as background music, graphical representation and narrative texts that either makes teaching through video more engaging or annoying. Table 5. Descriptive statistics of technological pedagogical content knowledge (TPACK). Item

Description

M

SD

TPACK1 The video lesson shows that the chosen subject content has been 3.286 0.604 presented effectively that will enhance the student’s learning of the subject content

Table 5 illustrates the evaluation score for the lecturers’ TPACK. Based on the results, the facilitators considered the lecturers were able to produce video lessons that enhanced students’ learning to a certain degree (M = 3.310, SD = 0.512). There were areas, as highlighted above, that required more attention namely the technological knowledge and skills dimensions. This is because the limitation of knowledge and skills restrict the options of better educational technology and diminish lecturers’ TPACK performance in teaching and learning.

5 Discussion and Conclusion This study aimed to evaluate the knowledge and skills of trained lecturers in integrating video technology (videos and screen-recorded PowerPoint presentation) into teaching and learning using the TPACK framework. Based on the findings of this study, it can be concluded that the workshop reported in this study closed the training gap of 21 lecturers in a higher education institution in Malaysia. Despite short training on video editing and screen recording, all the lecturers managed to produce their own educational videos and presented them on the second day of the training. There were adapted third-party videos with little editing embedded and played during the PowerPoint presentations, screen-recorded PowerPoint presentations, and a few fully self-created video lectures. The videos were mostly meant for in-class teaching, supplementary learning material of upcoming lectures, overview summary of the immediate past lecture, and one exception where the video was meant as a cue for class discussion.

Upskilling Educators for Pandemic Teaching

543

Overall, throughout the hands-on training on editing videos using Easy Movie Maker software and screen recording using Microsoft PowerPoint, it was found that most of the lecturers were new to these skills, and they were generally weak in these areas. The common practice of the lecturers was searching for videos on desired topics and adopting the videos in teaching without editing or customization to better suit their context. The deficiency in the technological knowledge in customizing videos to teach among the lecturers had restricted or weakened their technological pedagogical content knowledge and skills. Therefore, the training conducted in this study was considered essential to enhance the lecturers’ TPACK knowledge and skills to create and use educational videos in teaching. This study has upskilled the lecturers’ capability in using technology which is vital in higher education as it helps them to accommodate digital native students’ needs and optimize learning across various fields of study [7, 10]. Though the global pandemic has popularized the TPACK concept and further increased awareness of educational technology, it should also be aware that technology integration in teaching is voluntary and continual, rather than an enforced alternative due to the pandemic [17]. Hence, educators’ development of TPACK through training, as reported in this study, is essential not only to improve the quality of teaching [13] but also to inform lecturers on how they can enhance pedagogical approaches to better deliver their lessons and engage their learners, who are the digital natives [18, 19]. Nonetheless, this study has a few limitations that should be addressed for future improvements. It was noticed during the two-day workshop that the lecturers had to prepare a video lecture overnight after learning the relevant technological skills on the first day of the workshop. Given the short period, the task was considered demanding, especially for the lecturers who were unfamiliar with producing and editing videos. This could have affected the lecturers’ performance on the second day when they demonstrated their lesson and were evaluated by the facilitators. In future, workshop facilitators can assign lecturers from the same department or who teach the same subject to one group for lesson preparation and demonstration. This will encourage peer support while completing tasks and reduce lecturers’ burden to practice new knowledge overnight. Besides, some of the lecturers proclaimed that it was difficult to incorporate all the four technological domains of TPACK framework together during the instructional process, especially when they needed to produce their own teaching aids using technology. It required more practice to hone their TPACK knowledge and skills. Acknowledgments. We would also like to thank all lecturers, research assistants and students who offered their assistance in the fieldwork.

544

C. H. Lee et al.

References 1. Alfiras, M., Bojiah, J., Yassin, A.A.: COVID-19 pandemic and the changing paradigms of higher education: a gulf university perspective. Asian EFL J. 27(5.1), 339–347 (2020) 2. Shafie, H., Abd Majid, F., Ismail, I.S.: Technological pedagogical content knowledge (TPACK) in teaching 21st century skills in the 21st century classroom. Asian J. Univ. Educ. 15(3), 24–33 (2019). https://doi.org/10.24191/ajue.v15i3.7818 3. Sahrir, M.S.B., Zaini, A.R., Hassan, Y., Hamat, Z., Ismail, T.: Employing technological pedagogical content knowledge (TPACK) skill among teachers in preparing online school assessment for home-based learning. Ijaz Arabi J. Arabic Learn. 4(2) 2021. https://doi.org/10.18860/ ijazarabi.v4i2.11493 4. Mishra, P., Koehler, M.J.: Technological pedagogical content knowledge: a framework for teacher knowledge. Teach. Coll. Rec. 108(6), 1017–1054 (2006). https://doi.org/10.1111/j. 1467-9620.2006.00684.x 5. Mutanga, P., Nezandonyi, J., Bhukuvhani, C.: Enhancing engineering education through technological pedagogical and content knowledge (TPACK): a case study. Int. J. Educ. Dev. Using Inf. Commun. Technol. 14(3), 38–49 (2018) 6. Santos, J.M., Castro, R.D.R.: Technological Pedagogical content knowledge (TPACK) in action: application of learning in the classroom by pre-service teachers (PST). Soc. Sci. Hum. Open 3(1), 100110 (2021). https://doi.org/10.1016/j.ssaho.2021.100110 7. Mohamed Yusof, Y.H., Abdullah, N.K., Jantan, H.: Benchmarking the potential educational technology competency standard based on TPCK in Malaysia for local higher education institutions. In: Proceedings of the Second International Conference on the Future of ASEAN (ICoFA) 2017, vol. 1, pp. 343–356 (2019). https://doi.org/10.1007/978-981-10-8730-1_35 8. Faizan, N., Gottlieb, M., Loffler, A., Utesch, M.C., Krcmar, H.: State-of-the-art to measure the TPACK level of trainees in higher education to increase the learnability of the train-thetrainer (TTT) sessions. In: 2019 IEEE Global Engineering Education Conference (EDUCON), (2019). https://doi.org/10.1109/educon.2019.8725056 9. Koehler, M.J., Mishra, P., Cain, W.: What is technological pedagogical content knowledge (TPACK)? J. Educ. 193(3), 13–19 (2013). https://doi.org/10.1177/002205741319300303 10. Amhag, L., Hellström, L., Stigmar, M.: Teacher educators’ use of digital tools and needs for digital competence in higher education. J. Digit. Learn. Teach. Educ. 35(4), 203–220 (2019). https://doi.org/10.1080/21532974.2019.1646169 11. Foulger, T.S., Jin, Y., Mourlam, D.J., Ebersole, L.: Equitable access to TPACK research: tensions about using social media to disseminate scholarship. Comput. Educ. Open 3, 100067 (2022). https://doi.org/10.1016/j.caeo.2021.100067 12. Brinkley-Etzkorn, K.E.: Learning to teach online: measuring the influence of faculty development training on teaching effectiveness through a TPACK lens. Internet Higher Educ. 38 28–35 (2018). https://doi.org/10.1016/j.iheduc.2018.04.004 13. Koehler, M.J., Mishra, P.: What is technological pedagogical content knowledge? In: Contemporary Issues in Technology and Teacher Education, vol. 9, no. 1 (2009) 14. Koehler, M.J.: The TPACK Image (2011). Accessed 03 Jun 2022. http://www.tpack.org/ 15. Akun, J.C.A., Mohamad, F.S.: Technological pedagogical content knowledge (TPACK) and the teaching of science: determiners for professional development. Stud. Appl. Econ. 39(1), 1–12 (2021). https://doi.org/10.25115/eea.v39i1.4272 16. Fabian, K., Clayes, E., Kelly, L.: Putting design into practice: an investigation of TPACK scores of lecturers in a networked institution. Res. Learn. Technol. 27 (2019). https://doi.org/ 10.25304/rlt.v27.2296 17. Radmehr, F., Goodchild, S.: Switching to fully online teaching and learning of mathematics: the case of Norwegian mathematics lecturers and university students during the Covid-19

Upskilling Educators for Pandemic Teaching

545

pandemic. In: International Journal of Research in Undergraduate Mathematics Education, pp. 1–31 (2022). https://doi.org/10.1007/s40753-021-00162-9 18. Adipat, S.: Developing technological pedagogical content knowledge (TPACK) through technology-enhanced content and language-integrated learning (T-CLIL) instruction. Educ. Inf. Technol. 26(5), 6461–6477 (2021). https://doi.org/10.1007/s10639-021-10648-3 19. Hosen, M., Ogbeibu, S., Giridharan, B., Cham, T.-H., Lim, W.M., Paul, J.: Individual motivation and social media influence on student knowledge sharing and learning performance: Evidence from an emerging economy. Comput. Educ. 172, 104262 (2021). https://doi.org/ 10.1016/j.compedu.2021.104262

Analysing English for Science and Technology Reading Texts Using Flesch Reading Ease Online Formula: The Preparation for Academic Reading Lubna Ali Mohammed1(B) , Musheer Abdulwahid Aljaberi2 , Antony Sheela Anmary1 , and Mohammed Abdulkhaleq3 1 Department of TESL, Faculty of Social Sciences, Arts, and Humanities, Lincoln University

College (LUC), Petaling Jaya, Selangor, Malaysia [email protected], [email protected] 2 Faculty of Nursing & Applied Sciences, Lincoln University College (LUC), Petaling Jaya, Selangor, Malaysia 3 Department of English Language and Literature, Xiamen University Malaysia, Sepang, Malaysia

Abstract. This study aims to determine the effectiveness of the Yemeni Senior Secondary School curriculum in preparing students for academic reading at the tertiary level. In this qualitative study, the data used comprised all reading comprehension texts in the English for Science and Technology (EST) senior secondary school textbook, and the reading instructional design was analyzed in terms of the types, readability level, and grade level of the texts. The types of reading texts were analyzed by calculating the percentages of narrative versus expository texts, Flesch Reading Ease (FRES) readability test and the Flesch-Kincaid Grade Level test were used to analyze the readability level, and the length of reading texts was calculated based on Leslie and Caldwell’s Qualitative Reading Inventory (QRI 3 & 4). The findings showed a clear gap between the academic level of reading texts and the reading texts at the senior secondary school level. Approximately, all of the reading texts were found to be far below the grade level in terms of length and ability level. Moreover, not enough emphasis was given to the expository texts. The findings suggest that the Yemeni EST Senior Secondary Reading curriculum is one of the possible causes of the reading problems faced by Yemeni learners at the tertiary level. The researchers suggested a revision of the EST Senior Secondary reading instructional design. Keywords: Readability · Length of reading texts · Expository and narrative texts · Academic reading · Reading instructional design · Flesch reading ease and Flesch-kincaid grade level

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 546–561, 2023. https://doi.org/10.1007/978-3-031-20429-6_50

Analysing English for Science and Technology Reading Texts Using …

547

1 Introduction Many non-English-speaking countries (e.g. India, Algeria, Singapore, Malaysia, Turkey, Korea, Middle East countries, and Japan,) have announced English as the most important foreign language that must be taught in schools regardless of all the challenges they face [1, 2]. Achieving a good level of competence in English will facilitate interaction between people of different nationalities and ease their acculturation and adjustment all over the world [3–8]. It can also enhance the acquisition and processing of Baiardi information from the print and electronic media [9, 10]. Although all language skills are important to successfully pursuing higher education, in ESP courses, reading proficiency is the keystone of students’ academic success [11, 12]. At the university level, reading receives incomparable importance among all other language skills [13]. Considering the importance of reading skills, students need to be equipped to handle reading tasks for future higher educational settings [14]. Despite the increased interest in English reading [11, 15, 16], students still face substantial challenges in reading the vast academic texts at the tertiary level; to them, reading is a complex skill which, despite its importance, is not easily acquired [12, 17– 20]. In countries where English is a foreign language, such as Yemen, textbooks designed for native speakers of English are used by college professors; hence, students must be proficient in both the English language and their subject areas to achieve the desired success [21]. However, numerous challenges are faced by English language learners and teachers in all levels of education [19, 22–24]. These challenges were found to result from the weakness of the educational system in Yemen [18]. To avoid these problems, students need to be trained at the secondary level with reading comprehension tasks that resemble the reading demands at the tertiary level. For example, exposing the students to grade-level reading texts in terms of difficulty level and length will facilitate their comprehension more than narrative texts would and will familiarise them with suitable strategies to successfully process information from this text type in the same content area [13, 25]. Past studies on EFL reading in Yemen have indicated that students find reading in English difficult at the tertiary level [26–31]. Considering that secondary school is the gateway to the tertiary level, students’ ability to successfully comprehend grade-level expository texts at secondary school is essential for their academic performance at the tertiary level. However, no previous study has researched the effectiveness of the secondary school curriculum in preparing students with reading materials similar to academic-level materials. The current study; therefore, aims to investigate how well the Yemeni EST senior secondary school curriculum grooms students for academic reading at the tertiary level by analyzing their reading texts in terms of type, readability, and grade-level. The following research questions guided the aim of the current study: 1. What reading comprehension text types are used in the Yemeni EST senior secondary school textbook? 2. What readability level is reflected in the reading comprehension texts used in the Yemeni EST Senior Secondary school textbook? 3. How well do the reading texts used in the Yemeni EST senior secondary school textbook represent the grade level of senior secondary school?

548

L. A. Mohammed et al.

2 Literature Review Textbooks are regarded as the most common type of teaching material in language teaching discourse [32]. However, inappropriate textbook use can de-skill students and teachers [33]. As a result, textbook analysis is critical in determining the worth of instructional design resources [34]. It is worth noting that the importance of reviewing and revising the curriculum has been overlooked by the Yemeni MoE as the Crescent English Course for Yemen (CECY) textbooks that were introduced in 1999 are still being used today without any updates. This curriculum has been criticized by researchers in terms of different variables such as its theoretical emphasis, roles of the teacher and the learner, and emphasis on reading skills [26, 27, 30, 35, 36]. Hence, this study aims to analyze the MoE-recommended textbooks that teachers use in planning their instructional design. The duration of secondary education in Yemen is three years (grades 10–12). Based on their academic performance at the end of grade ten, students can choose to continue their studies in either of science or humanities track. At the end of senior secondary school (grade 12), students sit for the national exam which admits them to higher education. At the higher education level, more faculties are available for students that are enrolled in the science track than for the humanities. In the science track, English is mostly used as the medium of instruction and the academic resources are also in English in faculties such as dentistry, medicine, and pharmacy. Therefore, the senior secondary English reading texts must be at the same grade level as higher education texts in terms of length, difficulty, and type. 2.1 Types of Texts Generally, there are two main types; narrative and expository. These two differ in their structure and the reading strategies needed for comprehension [37, 38]. Narrative texts such as poems, short stories, and novels aim to entertain the readers, while expository texts, which are also called informative texts, aim to inform and provide new scientific information [39]. Examples of expository texts are problem-solution, cause-effect, and compare-contrast [40]. The Yemeni EST Senior Secondary learners should be trained on the structure of the expository text more than any other structure because this is the text type that academic materials are based on [41, 42]. Past studies (e.g. [11, 27, 37, 38]) have declared the positive effect of text structure instruction on students’ reading comprehension performance level. Abdualameer [11] recommended that teaching students reading comprehension and reading strategies for different types of reading texts should be emphasized in the syllabus of the English language curriculum. Therefore, training students on these strategies will enhance their understanding when reading comprehension texts. Hebert et al. [38] found that text structure instruction improves the students’ comprehension level while reading expository texts. In a recent meta-analysis study conducted by Pyle et al. [37], 21 studies that used expository texts as an intervention to enhance students’ reading comprehension in all levels of education (kindergarten–Grade 12) between 1970–2013 were analyzed. The researchers suggested that expository text structures should be explicitly described and taught in the classroom. As a result, to enhance students’ reading comprehension

Analysing English for Science and Technology Reading Texts Using …

549

skills at the tertiary level, they must be trained on expository text reading strategies as previously emphasized by researchers. 2.2 Readability Readability refers to “how easily written materials can be read and understood” [43]. The readability of reading texts depends on many factors, including the length of sentences and the number and difficulty level of vocabulary [44]. According to Izgi, Seker [44], the readability level of textbooks is considered a significant tool in the teaching-learning process. Hence, the lack of readability in textbooks seems to be a noticeable deficiency. For this reason, textbooks included in the current study were examined in terms of the readability of their texts. In addition, the length of passages was predicted to be one of the factors that added to the difficulty level of the texts, as it has been established that the most difficult to read texts are the lengthiest [45]. Previous studies emphasized the importance of exposing students to readability and grade level reading texts at secondary school to prepare them to read academic texts efficiently as they will be more confident and skilled (e.g., [15, 26, 27, 40, 42, 46, 47]. Azizi [15], examined the readability level of 8 texts by second-grade junior high school students using the Flesch Reading Ease analysis and showed that the reading texts are lower than the intended students’ level. Similarly, Rohmatillah [40] examined the course book used to teach senior high school (Grade 10) students in terms of the reading text types and their readability level. To analyze the readability level of the reading texts, the researcher used the Flesch readability formula. The findings reflected five different types of text. The majority of the texts (11 out of 16) were below the grade level of senior high school students. Later in 2016, Maryansyah [47] analyzed the readability level of the Sixty-three reading texts used in teaching ninth-grade students. Out of the 63 texts, only 9% were in the right grade level, while 54% were easy for Grade 9 students, 27% were difficult, and 10% were invalid. Consequently, the researcher suggested that English language teachers and curriculum designers to be aware of the importance of the readability level of reading texts and their suitability for the intended level of students. They have to conduct a readability analysis on reading texts before implementation. Recently, Zantoni [46] examined the readability level and student perception of 16 reading texts used in English teaching for Grade 8 students at Junior High School using the Flesch Reading Ease formula and the Flesch Kincaid Grade level. It was found that the reading texts were inappropriate for eighth-grade students. Ten (62.5%) out of sixteen reading texts were found to be very easy for the students. However, few studies found that the examined curriculum and textbooks presented appropriate materials for teaching EFL at the grade levels. For example, Budiarti [3] examined the readability level of English reading texts for Grade Eight students using the Fry Readability Formula (FRF) and Fog Index (FI) and found that the selected reading texts are readable and suitable for the level of intended students. Likewise, Hidayat [48] used the Flesch Reading Ease Formula only to analyze the readability level of five reading texts in an English textbook for senior high school grades and found that the reading texts under study were appropriate for the student’s level.

550

L. A. Mohammed et al.

3 Research Methodology 3.1 Research Design This study adopted a qualitative content analysis (CA) approach [3, 15, 26, 27, 48]. The data comprised all the reading texts used in the senior secondary school textbook. This textbook consists of six core units that are divided into 12 sub-units. An additional unit for science readers constituted the seventh unit of the book. A total of 22 reading texts were collected. 3.2 Theoretical Framework The theoretical framework of the current study is based on Sidek’s model of language teaching [49]. According to the model, language teaching can be analyzed in terms of its approach and design. Given that the current study focuses on analyzing the textbooks, the theoretical framework of the design is relevant. The analysis of the types, length, and readability level of reading texts is represented in the theoretical framework of the reading instructional design developed by the MoE in Yemen. 3.3 Data Analysis To answer question one, the reading texts found in the EST Senior Secondary School textbook were labeled as either expository or narrative. As this study examined how well the Senior Secondary school curriculum prepares the student for academic reading at the tertiary level, it was anticipated that the majority of reading comprehension texts would be expository. The percentage of each type was calculated based on their frequency in the textbook (Table 1). To answer the second question, the Flesch Reading Ease (FRES) readability test and the Flesch-Kincaid Grade Level test were used. The Flesch-Kincaid Grade Level (GL) formula was used to analyze the readability grade level of texts in terms of average sentence length (syntactic complexity) and the average word length in syllables (semantic complexity). Specifically, this readability formula was selected because it is one of the best formulas to predict the complexity of expository texts and it is also the most regularly tested and reliable formula [50]. The Flesch Reading Ease Score test measures the difficulty level of reading texts and predicts the typical grade level of students. The Flesch Reading Ease formula was selected because it is the most reliable method [51]. Flesch-Kincaid Grade Level is an index that gives the required years of education to comprehend a document. However, Flesch’s tests were developed for measuring the readability of texts for native English speakers. Their validity in measuring EFL reading difficulty was proved by Greenfield (1999), as cited in Greenfield [52], who found that readability formulas for native readers are also valuable tools for measuring the readability level of texts for EFL learners. The combination of these two formulas was selected to analyze the readability of EST Senior Secondary School reading texts because they are the best formulas for readability analysis.

Analysing English for Science and Technology Reading Texts Using …

551

Table 1. Titles of EST Senior Secondary School Reading Texts Unit

Titles of core units

Sub units reading texts

Science reader texts

1

Describing things

1. A Drive in the Country side 2. A View from the Window

1. Acids and alkalis 2. State of Matter

2

Reporting events

3. Today’s News: In the Daily Post Today 4. Hurricane Hits Central America. Thousands Dead

3. Light 4. Sound

3

Looking for a job

5. Thinking about the Future 6. Applying for a Job

5. Arabic Scientists 6. Vaccination

4

Tables, flow charts and diagrams

7. Agriculture in Yemen 8. Frozen Peas

7. Experimental Procedures 8. Internal Combustion Engine

5

Working things out

9. Puzzles and Riddles 9. The Moon 10. The Mystery of the Mary 10. Radio Activity Celeste

6

Looking back

11. Emergencies in the News 12. A Long Life in Medicine

Total

22 EST reading texts

Flesch-Kincaid Grade Level Formula: (0.39 × ASL) + (11.8 × ASW ) − 15.59 where: ASL = average sentence length = ASW = average number of

number of words number of sentences . of syllables syllables per word = number number of words .

Flesch Reading Ease Formula: FRES = 206.835 − (1.015 × ASL) − (84.6 × ASW ) where: ASW = average number of syllables per word =

number of syllables number of words .

Flesch Reading Ease Score uses a scale of 0 to 100. Table 2 provides the interpretation of the Flesch Reading Ease Score [53] as extracted from Heydari [54]. According to Flesch’s Reading Ease Scale, a text with a reading ease score of 100 should be very easy for students who have finished Grade Four, while a reading ease score of 0 denotes that a text is difficult for secondary school students. Table 2 also shows

552

L. A. Mohammed et al. Table 2. Flesch’s reading ease scale [53]

Reading ease score

Description

Predicted reading grade

0–30

Very difficult

College graduate

30–40

Difficult

College level 13–16th

50–60

Fairly difficult

10–12th grade

60–70

Standard

8–9th grade

70–80

Fairly easy

7th grade

80–90

Easy

6th grade

90–100

Very easy

5th grade

(Source Heydari [54], p. 424)

that the range 40–50 was left out in DuBay’s Table for interpreting the Flesch Reading Ease Score [53]. The reason for neglecting this range is not clear [54]. To answer the third research question, Leslie and Caldwell’s Qualitative Reading Inventory (QRI 3 & 4) Leslie, Caldwell [55] was used to analyze the grade level of texts. The selection of these inventories, as justified by Sidek [49], is based on the unavailability of other published inventories measuring the texts’ grade-level in terms of length for FL reading context. Based on these inventories, the grade-level length of texts for the Senior Secondary school level should be between 470–550 words. The word length for both types of texts was calculated using Microsoft Word and was then interpreted as follows: (1) Texts with 470 words and above are grade-level texts. (2) Texts with less than 470 words are under the grade-level.

4 Findings 4.1 Types of Reading Texts The types of reading texts in the textbook were analyzed in terms of two categories: narrative and expository. The findings are presented in Table 3: Table 3. Analysis of reading text types of reading comprehension in the Yemeni senior secondary textbook Type of texts

Number

Percentages

Expository

12

55%

Narrative

10

45%

The findings show that the reading instructional design of the Senior Secondary school exposes students to both narrative and expository text types. Out of 22 reading texts, 12 of them were expository and 10 were narrative.

Analysing English for Science and Technology Reading Texts Using …

553

4.2 Readability Level The readability level of the senior secondary school reading texts was analysed using the Flesch Reading Ease Index and the Flesch-Kincaid Grade Level (GL) Score. The results are presented in Tables 4 and 5. Table 4. Analysis of readability level of 22 reading texts in the Yemeni senior secondary textbook No. reading texts

Reading ease

Grade level

1

92

Very easy to read

3.3

Grade 3

2

89

Easy to read

4.4

Grade 4

3

67.2

Standard

8.1

Grade8

4

72.1

Fairly easy to read

6.4

Grade6

5

69

Standard

5.9

Grade 6

6

69

Standard

6.8

Grade 7

7

73

Fairly easy to read

6.7

Grade 7

8

80

Easy to read

5.7

Grade6

9

91

Very easy to read

2.7

Grade3

10

74.4

Fairly easy to read

6.3

Grade 6

11

66.1

Standard

7.9

Grade 8

12

67.8

Standard

6.9

Grade 7

13

70.3

Fairly easy to read

6.4

Grade 6

14

77.8

Fairly easy to read

5.9

Grade6

15

81.2

Easy to read

5

Grade5

16

79.4

Easy to read

6.1

Grade 6

17

64.9

Standard

7.8

Grade 8

18

61.7

Standard

8.4

Grade 8

19

46.6

Fairly difficult to read

11

Grade11

20

68.6

Standard

6.9

Grade7

21

79.9

Easy to read

6.4

Grade 6

22

52.4

Fairly difficult to read

10

Grade10

Mean

72.4

Fairly easy to read

6.6

Grade 6–7

As shown in Table 4, almost all reading texts in the EST Senior Secondary School were below grade level. The mean score of overall text readability in terms of reading ease was 72.4 (fairly easy to read), while the mean level of the reading texts in terms of grade level was 6.6 (grade level 6–7). According to the Flesch Reading Ease score, reading texts at the university level are in the difficult category (30–40); therefore, for EST 3rd grade students to be able to comprehend authentic English texts in content-based

554

L. A. Mohammed et al.

Table 5. Percentages of the analysis of the 22 reading texts based on Flesch reading ease scale Reading ease score

Description

Predicted reading grade

0–30

Very difficult

College graduate

30–40

Difficult

College grade

00

50–60

Fairly difficult

10–12th grade

9%

60–70

Standard

8–9th grade

36%

70–80

Fairly easy

7th grade

23%

80–90

Easy

6th grade

23%

90–100

Very easy

5th grade

9%

00

areas at the university level, they need to be trained to process fairly difficult reading texts with a reading ease score of between 50–60 at the secondary level. Nevertheless, as shown in Table 5, only two passages, accounting for 9% of the overall texts, were designed with a fairly difficult level at 46.6 and 52.4 for grades 10 and 11. Moreover, other texts ranged from very easy to standard. 4.3 Length of Reading Texts The results of analyzing the length of the reading texts are presented below in Table 5. In terms of the two genres of reading texts in the Yemeni Secondary School Textbook, the findings show that there is no significant difference between the mean length of narrative texts and expository texts. The mean length of narrative texts was 324 words, while that of expository texts was 316. Neither the mean length of the expository texts nor that of the narrative texts conformed to the grade level suggested in Leslie and Caldwell’s Qualitative Reading Inventory 3 and 4, except for one narrative text, which recorded 517 words. However, it was very easy in terms of the readability level, while the other 21 passages were far below the grade level, with the longest text having 416 words (Table 6).

5 Discussion At the tertiary level, students are expected to effectively read various sorts of texts, from textbooks, journal articles, web pages, magazines to newspapers. This is confirmed by Sani, Chik [56], who indicated that “diploma students and undergraduates need to read a lot of academic texts, journals, websites, and magazines regularly, which requires them to use high levels of reading comprehension skills and therefore make reading an effortful activity” (p. 34). Thus, paying attention to the construction of text elements in the EST curriculum is of great importance in order to prepare EST learners well for reading at a tertiary level. Since the textbook of interest in the current study was designed for EST students, it was expected that the reading texts to be expository, as they would encounter in the

Analysing English for Science and Technology Reading Texts Using …

555

Table 6. Length of reading Texts in EST Senior secondary school textbook No

Narrative texts

Length (In Words)

No

Expository texts

Length (In Words)

1

A Drive in the Country side

325

1

Agriculture in Yemen

323

2

A View from the Window

259

2

Frozen Peas

217

3

Today’s News: In the Daily Post Today

237

3

Acids and alkalis

307

4

Hurricane Hits Central America. Thousands Dead

258

4

State of Matter

293

5

Thinking about the Future

390

5

Light

274

6

Applying for a Job

335

6

Sound

397

7

Puzzles and Riddles

360

7

Arabic Scientists

391

8

The Mystery of the Mary Celeste

517

8

Vaccination

328

9

Emergencies in the News

204

9

Experimental Procedures

252

10

A Long Life in Medicine

354

10

Internal Combustion 207 Engine

11

The Moon

416

12

Radio Activity

385

Mean length

316

Mean length

324

relevant content area at the tertiary level. This expectation was based on the consensus of many researchers that reading texts in academic settings are expository texts [41, 42]. Hence, if EST Senior Secondary reading instructional design trains students to process narrative genre texts more than expository, the students will be more proficient in processing the former. This practice will create difficulty for Yemeni students in processing content area texts that they are not regularly trained upon i.e. the expository genre texts. The finding on text types in this study is to some extent, in line with the findings of Sidek [49] who found that the Malaysian Upper Secondary school English language reading curriculum emphasizes the training of narrative reading texts more than expository texts. Although the EST senior secondary school curriculum contains one section at the back of the textbook for the science reader, containing 10 expository texts, the inclusion of general texts at the beginning of the textbook made the narrative texts 2 fewer than the

556

L. A. Mohammed et al.

expository ones in the reading instructional design. In Yemeni settings, reading teachers are not given the flexibility to use texts other than those included in the textbook. The findings show that the EFL curriculum designers in Yemen seemed to be unaware of the primary objective of the EST program, which is to familiarize and prepare students to read and comprehend academic texts at the tertiary level. Restricting English teachers to use the reading texts that are in the EST textbook goes against the notion that the Yemeni EFL curriculum is a communicative-based curriculum. Must be used to achieve learning outcomes based on the communicative-based curriculum. Such over-reliance on reading texts in the EST textbook should be discouraged in order to prepare the learners to read academic expository texts successfully. In line with the Yemeni educational philosophy, which prioritises the needs of the learner in designing the curriculum, EST students need to be sufficiently trained to process information from different structures of expository texts [11, 27, 37, 38]. In terms of the readability level of reading texts in the EST Senior Secondary school curriculum, the findings showed that all the reading texts in the EST textbook were below the grade level in terms of reading ease (100%), while in terms of the grade level, no single reading text reflected the grade level of the Senior Secondary school grade level (level 12). The readability level of texts is one of the textbook features that affect the students’ reading comprehension [57, 58]. However, this was not given adequate consideration while developing the EST secondary school reading instructional design. The readability level of EST reading texts in the EST course book was not appropriately addressed—only 2 texts out of 22 were at the borderline of grade-level. However, those 2 texts entitled “Radio Activity” and “Experimental Procedure” did not match the exact level of senior secondary school learners. Rather, they fell into the “fairly difficult category” at levels 10 and 11, respectively. Nonetheless, the findings showed that all examined EST reading texts conformed to non-grade-level, grade 6 (36.3%), grade 7 (18.1%), grade 8 (18.1%), grade 10 (4.5%), and grade 11 (4.5%). For EST senior secondary school learners to succeed in their academic areas, they need to be prepared to process texts whose difficulty levels are equivalent to those used at the tertiary level. It is obvious that a significant gap exists between the readability level of texts that the students read in Senior Secondary school and the level of difficulty of academic texts used at higher levels. These findings are in line with previous studies that found a clear gap in the complexity level of texts in high school and those at the university level [15, 47, 50]. Such a gap may indicate the reading challenges that the students will be faced with while studying at the university level. As such, this deficiency in the EST textbook seems to contribute to the reading comprehension difficulties that Yemeni secondary school graduates face at the university level. Moreover, academic texts require the purposeful and critical reading of a range of lengthy and complex texts [37, 59]. Thus, training students on grade level texts in terms of their length is essential for preparing them for academic reading. However, the findings on text length in this study showed that the majority of reading texts in the EST textbook did not follow the senior secondary school level, while only one text, “The Mystery of the Mary Celeste,” conformed to the grade-level. However, it is a narrative text. These findings suggest that the reading instructional design of senior secondary schools was

Analysing English for Science and Technology Reading Texts Using …

557

designed without proper planning and consideration for students’ needs. It is thus clear that Yemeni EST senior secondary school students are not sufficiently trained to read grade level texts. It thus follows logically that the students be challenged in processing long and complex reading texts at the tertiary level. Past studies suggest that many students register a high level of failure in academic reading due to text misinterpretation (e.g. [15, 27, 46]), which could limit their ability to process grade-level academic texts. [49] reported a similar finding in the Malaysian context, in which the majority of reading texts in the national Malaysian upper secondary school were found to be below grade level.

6 Conclusion and Recommendation The current study aimed to find out the effectiveness of the Yemeni EST Senior Secondary school curriculum in preparing students for academic reading at the tertiary level. A content analysis approach was used in analyzing the EST Senior Secondary reading instructional design. A total of 22 reading comprehension texts found in the EST senior secondary school textbook were analyzed in terms of text types, readability, and grade level. The findings showed a clear gap between the academic level reading texts and the reading texts at the senior secondary school level. The findings showed that the EST Senior Secondary reading curriculum does not prepare Yemeni learners for reading in English at the tertiary level in terms of the designed reading comprehension texts. These findings provide evidence to suggest that the Yemeni EST Senior Secondary reading curriculum is one of the possible causes of reading problems faced by Yemeni learners at the tertiary level. This study is the first of its kind that focused on the EST reading curriculum in the Yemeni context for preparing EST Yemeni learners for academic reading at the university level. The findings of the senior secondary reading curriculum analysis can be extended for future research. However, this study only examined the EST reading curriculum at the senior secondary school level in Yemen. Since the secondary school level lasts 3 years, it might be beneficial to analyze reading curriculum for the junior high school reading curriculum (Grade 10 and 11). By examining and revising the EFL reading curriculum at the secondary level, the Yemeni MoE may best prepare its students for academic reading in English at the university level. Based on the findings of this study, some recommendations are proposed to ensure that the curriculum fully prepares the EST senior secondary school students for successful reading in English in their content-based areas at the tertiary level. Firstly, Curriculum designers should have a clear understanding of the main objective of the National Secondary Educational Philosophy, which is to prepare Yemeni secondary school students to transit to higher education successfully. Ensuring the accomplishment of the curriculum objectives is vital. According to Obanya (2002), a perfect match must be found between the designed curriculum, the implemented curriculum, and the achieved curriculum. Secondly, in terms of the types, readability level, and length of texts, it is recommended that in the revised version of the EST Senior Secondary curriculum, the reading texts in the textbook should be selected with great attention. The reading texts in the proposed revised curriculum should be appropriate in terms of the type and grade level

558

L. A. Mohammed et al.

of students as well as their readability level and length. Specifically, the selected reading texts should be expository grade-level passages because students frequently encounter such complex and lengthy texts at the university level [37, 59]. To effectively prepare the learners to be able to process expository texts at the university level, they should first be trained in the organization and structure of such texts at the secondary school level. Thirdly, to address the gaps between curriculum goals and outcomes, instructional design in the curriculum should be carefully analyzed and improved such that textbooks are prepared to proper standards. This is because textbooks can help in achieving the aims of the curriculum as they constitute an important element of the educational process. The revisions that are recommended in this study for the EST Senior Secondary reading curriculum could ensure that EST Senior Secondary students become successful readers even when they encounter complex expository texts at the tertiary level. The recommended revisions in the EST Senior Secondary reading curriculum are illustrated in Fig. 1.

Fig. 1. The proposed revisions for EST senior secondary school reading instructional design

Analysing English for Science and Technology Reading Texts Using …

559

References 1. Harishini, S., Lubna Ali, M., Asra, A.: Challenges associated with E- Learning among ESL undergraduates in Malaysia: a conceptual framework. Int. J. Manag. Hum. Sci. (IJMHS) 4(4), 30–38 (2020) 2. Shalini, A., Lubna Ali, M., Nithiyaroobi, K.: Code switching phenomenon in english language classrooms at the tertiary level. Int. J. Manag. Hum. Sci. (IJMHS) 3(1), 35–42 (2019) 3. Budiarti, N.I.: The readability level of english reading texts for grade VIII students of SMP Negeri 1 Jetis Bantul in the academic year of 2014/2015. Department of English Language Education Faculty of Languages and Arts State University of Yogyakarta (2014) 4. Saeed, M.A., Ghazali, K., Aljaberi, M.A.: A review of previous studies on ESL/EFL learners’ interactional feedback exchanges in face-to-face and computer-assisted peer review of writing. Int. J. Educ. Technol. High. Educ. 15(1), 1–25 (2018). https://doi.org/10.1186/s41239-0170084-8 5. Aljaberi, M.A., Alsalahi, A., Juni, M.H., Noman, S., Al-Tammemi, A.A.B., Hamat, R.A.: Efficacy of interventional programs in reducing acculturative stress and enhancing adjustment of international students to the new host educational environment: a systematic review and meta-analysis. Int. J. Environ. Res. Public Health 18(15), (2021). https://doi.org/10.3390/ije rph18157765 6. Mohammed, M.A.S., Al-Jaberi, M.A.: Google docs or microsoft word? Master’s students’ engagement with instructor written feedback on academic writing in a cross-cultural setting. Comput. Compos. 62, 102672 (2021). https://doi.org/10.1016/j.compcom.2021.102672 7. Al-Jaberi, M.A., Juni, M.H., Kadir Shahar, H., Ismail, S.I.F., Saeed, M.A., Ying, L.P.: Effectiveness of an educational intervention in reducing new international postgraduates’ acculturative stress in Malaysian Public Universities: protocol for a cluster randomized controlled trial. JMIR Res Protoc. 9(2), e12950 (2020). https://doi.org/10.2196/12950 8. Musheer, A.-J., Juni, M.H., Shahar, H.K., Ismail, S.I.: Acculturative stress and intention to dropout from the university among new postgraduate international student in public universities, Malaysia. Malays. J. Med. Health Sci. 15(104) (2019) 9. Azizi, A.R.: An analysis of the readability level of reading texts in passport to the world 2 textbook for second grade of junior high school. UIN Syarif Hidayatullah Jakarta, Indonesia (2015) 10. Sivanisswary, K., Lubna, A.M.: Predictive factors associated with online learning during COVID-19 pandemic in Malaysia: a conceptual framework. Int. J. Manag. Hum. Sci. (IJMHS) 4(4), 19–29 (2020) 11. Abdualameer, R.T.: Effects of Text Types and Reading Strategies on Reading Comprehension Performance. University Sains Malaysia, Penang Malaysia (2016) 12. Kavaliauskien˙e, G., Anusien˙e, L.: Online reading and writing in English for specific purposes classes. Kalb˛u Studijos 17, 99–104 (2010) 13. Carrell, P.L.: Facilitating ESL reading by teaching text structure. TESOL Q 19(4), 727–752 (1985) 14. Grabe, W.: Reading-writing relations: theoretical perspectives and instructional practices. Linking literacies: Perspectives on L2 reading-writing connections, pp. 15–47 (2001) 15. Azizi, A.R.: An analysis of the readability level of reading texts in passport to the world 2 textbook for second grade of junior high school. FITK UIN Syarif Hidayatullah Jakarta, Jakarta (2015) 16. Levine, A., Ferenz, O., Reves, T.: EFL academic reading and modern technology: how can we turn our students into independent critical readers. TESL-EJ 4(4), 1–9 (2000) 17. Rahim, P.R.M.A.: Fostering reading comprehension skill among ESL tertiary level students through discourse engagement. Int. J. Adv. Appl. Sci. 4(12), 263–272 (2017)

560

L. A. Mohammed et al.

18. Abdullah, N., Patil, V.: English-language teaching in Yemen: importance and challenges. Int. J. Soc. Sci. Tomorrow 1(5), 1–8 (2012) 19. Mourtaga, K.R.: Some reading problems of Arab EFL students. Al-Aqsa Univ. J. (Humanities Ser.) 10(2), 75–91 (2006) 20. Hazrimah, B.A., Harison, M.S., Lubna A.M.: An analysis of reading instructional approach at the Malaysian lower secondary school level. In: Suzanah, S., et al. (eds.) Language and Education Trends, pp. 151–73. University Sains Islam Malaysia Press, Malaysia (2015) 21. Huang, S.: Reading English for academic purposes–what situational factors may motivate learners to read? System 34(3), 371–383 (2006). https://doi.org/10.1016/j.system.2006. 04.006 22. Fareh, S.: Challenges of teaching English in the Arab world: Why can’t EFL programs deliver as expected? Procedia. Soc. Behav. Sci. 2(2), 3600–3604 (2010). https://doi.org/10.1016/j. sbspro.2010.03.559 23. Balfakeh, S.A.D.: Problems in reading comprehension skills among secondary school students in Yemen. Lang. India 9(10) (2009) 24. Almahedi, I.: Reading comprehension skills among Yemeni students: a case study. Ph.D. dissertation, Pune University, India 25. Pugh, S.L., Pawan, F., Antommarchi, C.: Academic literacy and the new college learner. Handbook of college reading and study strategy research, pp. 25–42. Lawrence Erlbaum Associates Publishers, Mahwah, NJ, US (2000) 26. Mohammed, L.A., Sidek, H.M.: EST secondary school reading curriculum: the preparation for reading at tertiary level. Asian J. English Lang. Pedag. 4, 40–49 (2016) 27. Mohd, S.H.: EFL textbook analysis: a case study. Lang. Lit. 14(3), 27–45 (2012). https://doi. org/10.20360/G2HP4J 28. Bhooth, A., Azman, H., Ismail, K.: The role of the L1 as a scaffolding tool in the EFL reading classroom. Procedia. Soc. Behav. Sci. 118, 76–84 (2014). https://doi.org/10.1016/j.sbspro. 2014.02.011 29. Azman, H., Bhooth, A.M., Ismail, K.: Readers reading practices Of EFL Yemeni students: recommendations for the 21st century. GEMA Online J. Lang. Stud. 13(3) (2013) 30. Mohammed, M.N.H.: A Critical Study of Crescent English Course for Secondary Stage in Yemen. University of Pune, India (2009) 31. Al-Tamimi, N.O.: The effect of direct reading strategy instruction on students’ reading comprehension, metacognitive strategy awareness, and reading attitudes among eleventh grade students in Yemen. University Sains Malaysia. Malaysia (2006) 32. Richards, J.C.: Curriculum development in language teaching. Cambridge University Press (2001) 33. Charalambous, A.C.: The Role and Use of Course Books in EFL. Online Submission (2011) 34. Lee, S.-M.: The development of evaluation theories for foreign language textbooks. J. Pan-Pac. Appl. Linguist. 17(2), 69–89 (2013) 35. Na’ama, A.: An analysis of errors made by Yemeni university students in the English consonant-clusters system. Damascus Univ. J. 27(3), 145–161 (2011) 36. Al-Ahdal, A.A.M.H.: ELT in Yemen and India-the need for remedial measures. Lang. India 10(11) (2010) 37. Pyle, N., Vasquez, A.C., Lignugaris/Kraft, B., Gillam, S.L., Reutzel, D.R., Olszewski, A., et al.: Effects of expository text structure interventions on comprehension: a meta-analysis. Read. Res. Q. 52(4), 469–501 (2017). https://doi.org/10.1002/rrq.179 38. Hebert, M., Bohaty, J.J., Nelson, J.R., Brown, J.: The effects of text structure instruction on expository reading comprehension: a meta-analysis. J. Educ. Psychol. 108(5), 609 (2016) 39. Weaver, C.A., Kintsch, W.: Expository text. Handbook of reading research, pp. 230–245 (1991)

Analysing English for Science and Technology Reading Texts Using …

561

40. Rohmatillah, R.: Readibility level of reading texts in the english textbook entitled english alive for senior high school grade X published by Yudhistira. English Educ. J. Tadris Bahasa Inggris 7(1), 81–101 (2015) 41. Fetters, C., Ortlieb, E., Cheek, E., Jr.: An exploration of strategy-based reading instruction using expository science texts in the elementary grades. Stud. Literat. Lang. 2(2), 113–126 (2011) 42. Fludernik, M.: Genres, text types, or discourse modes? Narrat. Modalities Generic Categ. Style. 34(2), 274–292 (2000) 43. Richards, J.C., Schmidt, R.W.: Longman dictionary of language teaching and applied linguistics. Routledge (2013) 44. Izgi, U., Seker, B.S.: Comparing different readability formulas on the examples of sciencetechnology and social science textbooks. Procedia. Soc. Behav. Sci. 46, 178–182 (2012). https://doi.org/10.1016/j.sbspro.2012.05.089 45. Mehrpour, S., Riazi, A.: The impact of text length on EFL students’ reading comprehension. Asian EFL J. 6(3), 1–13 (2004) 46. Zantoni, M.: The readability level of reading texts in the english textbook entitled “english on sky 2” used by the eighth grade students of smp budaya bandar lampung in the academic year of 2017/2018. UIN Raden Intan Lampung (2019) 47. Maryansyah, Y.: An analysis on readability of English reading texts for grade IX students at MTsN 2 Kota Bengkulu. J. English Educ. Appl. Linguist. 5(1), 69–88 (2016) 48. Hidayat, R.: The readability of reading texts on the English textbook. In: International Conference: Role of International Languages toward Global Education System. Indonesia (2016) 49. Sidek, H.M.: An analysis of the EFL secondary reading curriculum in Malaysia: approaches to reading and preparation for higher education. University of Pittsburgh (2010) 50. Sheehan, K.M., Kostin, I., Futagi, Y., Flor, M.: Generating automated text complexity classifications that are aligned with targeted text complexity standards. ETS Res. Report Ser. 2010(2), i–44 (2010). https://doi.org/10.1002/j.2333-8504.2010.tb02235.x 51. Klare, G.R.: Measurement of Readability. University of Iowa Press, Ames, Iowa (1963) 52. Greenfield, J.: Readability formulas for EFL. JALT J. 26(1), 5–24 (2004). https://doi.org/10. 37546/JALTJJ26.1-1 53. DuBay, W.H.: Smart language: readers, readability, and the grading of text. Costa Mesa: Impact Information (2006) 54. Heydari, P.: The validity of some popular readability formulas. Mediterr. J. Soc. Sci. 3, 423– 435 (2012). https://doi.org/10.5901/mjss.2012.v3n2.423 55. Leslie, L., Caldwell, J.S.: Qualitative reading inventory. Harper Collins New York (2006) 56. Sani, B.B., Chik, M.N.B.W.: The reading motivation and reading strategies used by undergraduates in university Teknologi MARA Dungun, Terengganu. J. Lang. Teach. Res. 2(1) (2011) 57. Rottensteiner, S.: Structure, function and readability of new textbooks in relation to comprehension. Procedia. Soc. Behav. Sci. 2(2), 3892–3898 (2010). https://doi.org/10.1016/j.sbspro. 2010.03.611 58. Miller, D.: ESL reading textbooks vs. university textbooks: Are we giving our students the input they may need? J. English Acad. Purp. 10(1), 32–46 (2011). https://doi.org/10.1016/j. jeap.2010.12.002 59. Beck, I.L., McKeown, M.G., Sinatra, G.M., Loxterman, J.A.: Revising social studies text from a text-processing perspective: evidence of improved comprehensibility. Read. Res. Q. 251–276 (1991)

Research on Continued Intention to Adopt E-Learning in Beijing University During Covid-19 Epidemic in China Zhao Ming Sheng1 , Poh Hwa Eng2(B)

, and Tat-Huei Cham1

1 UCSI Graduate Business School, UCSI University, Kuala Lumpur, Malaysia

[email protected] 2 Faculty of Business and Management, UCSI, UCSI University, Kuala Lumpur, Malaysia

[email protected]

Abstract. The study aims to apply the TAM model to investigate further about the opinions of students at Peking University on the e-learning platform provided by the university. The research validates the impact of the different variables on students’ intention to continue using e-learning platforms during the Covid-19 epidemic, including perceived usefulness, perceived ease of use, perceived enjoyment, and perceived convenience. The research adopts a quantitative research method and measures the variables using a 5-level Likert scale. The designed questionnaires are distributed online to the targeted respondents through multiple social applications. The research collects 383 valid data and conducts a comprehensive data analysis by using SPSS analysis software. The research indicates that perceived usefulness, perceived ease of use, and perceived convenience all have a significant and positive impact on student’s intention to continue using elearning platforms. However, this research also confirms that perceived enjoyment has no significant effect on student’s intention to continue using e-learning platforms. Therefore, this research validates the hypotheses proposed and achieves the research goals. Furthermore, the research provides useful information and recommendations for the government, education field, IT field and future research. Keywords: E-learning platform · Perceived usefulness · Perceived ease of use · Perceived enjoyment · And perceived convenience

1 Introduction The online-learning approach is adopted at the different academic universities in China to replace the traditional, physical classroom-teaching approach with the intention to allow students to continue their studies and to prevent the spread of the Covid-19 virus during the epidemic. Online learning, defined as “a form of distributed learning enabled by the internet” [1], was previously an optional and secondary paradigm of instruction for institutions of higher education and has presently become the primary and obligatory model for learning. Students and academicians must adjust to this paradigm shift during a tumultuous and uncertain time, with no foreseeable return to traditional, in-person © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 562–572, 2023. https://doi.org/10.1007/978-3-031-20429-6_51

Research on Continued Intention to Adopt E-Learning

563

classroom environment. Therefore, it is important to conduct quality online learning models to ensure satisfactory academic quality. This transition from conventional educational models to online learning models must be as seamless as possible to enable online teaching to act as a substitute for conventional teaching without compromising the quality of education. To ensure a smooth transition with minimal effect on the quality of the academic teaching, it is imperative that both the beneficiaries and implementers of the online learning model are familiarized and trained with the process of implementation. The implementation process will undoubtedly introduce new technologies and protocols for learning, of which both students and faculty members must become sufficiently proficient in the online learning approaches during the transition period. For the case of online learning during the period of the epidemic, the end-user experience and intention to adopt new technology is a primary concern for academic institutions, and the implementation and its perceived success may serve as a precedent for future cases. However, the challenges regarding utility, usability, user experience, and accessibility during the critical period of the epidemic and the transition of online learning as the primary educational paradigm for institutions of higher education, must be addressed in a sufficient manner to ensure widespread adoption and future sustainability of online learning approaches. The aim of this study is to analyze the sustainability of online classes as an effectual substitute for physical teaching and the inherent differences this pedagogical method may incur upon students in Peking University during the COVID19 epidemic. A conceptual framework is developed based on the technology acceptance model (TAM) and information system model proposed by [2] to quantify the measures of user’s intention to adopt new technology. Essential considerations for the intention to adopt online platforms include the student’s perceived usefulness, perceived ease of use, perceived enjoyment, and perceived convenience. The research objectives of the study include the following: RO1: To examine the relationship between perceived usefulness and students’ continued intention to use the online learning platforms of Peking University. RO2: To examine the relationship between perceived ease of use and students’ continued intention to use the online learning platforms of Peking University. RO3: To examine the relationship between perceived enjoyment and students’ continued intention to use the online learning platforms of Peking University. RO4: To examine the relationship between perceived convenience and students’ continued intention to use the online learning platforms of Peking University. The results of this study could potentially serve as a precedent for future implementations of online learning protocols. The conceptual framework developed in this study evaluates the online learning platforms implemented by Peking University as a function of four constructs used within the confines of the TAM: perceived usefulness, perceived ease of use, perceived enjoyment, and perceived convenience. The results of this study will improve the understanding of the implementation of online classes and may be extrapolated to other institutions of higher education that have a similar use case.

564

Z. M. Sheng et al.

2 Literature Review The incidence of epidemic such as COVID-19 and its subsequent impact on academic institutions is a relatively novel occurrence in the digital age. As such, universities are faced with an unprecedented shift in pedagogical paradigm. Due to the mandated change in paradigm, Davis’s 1989 TAM must be adopted to more adequately assess the adoptive success of new technology, in this case, the online learning “5 + N” protocol implemented by Peking University for its enrolled students. Davis’s 1989 TAM is contingent upon user acceptance of new technology as a function of intention to use [2]. Extenuating circumstances resulting from the epidemic have only made the intention to use more important due to mandatory requirements for continued academic learning. Therefore, it is important to investigate on the continuance intention to use online learning platforms among university students. When examining the implementation of the online learning platforms at Peking University from the perspective of sustainability and long-term viability, intention to use must be extrapolated as the continued intention to use. The roles of the endogenous constructs of Davis’s TAM of perceived usefulness and perceived ease of use retains their influences on end-user intention to use, which have been demonstrated extensively by previous literatures as having a positive and significant correlation with user adoption [2, 3]. Additionally, due to the emergence of increasing competition in the digital application workspace, exogenous variables such as perceived enjoyment and perceived convenience of end user must also be accounted for as significant determinants for intention to use, actual use, and long-term adoptive success. Previous studies have already expounded upon the notion that digital media employing a high degree of perceived enjoyment has positive and statistically significant correlations with the adoption of new technology. In Teo and Noyes’s 2011 analysis of perceived enjoyment and user adoption of technology, perceived enjoyment was found to be a primary driver behind both endogenous constructs of TAM [4]. Additional studies have found less conclusive evidence of perceived enjoyment as a primary driver for both the core TAM constructs but have demonstrated statistically significant correlations between perceived enjoyment and intention to use and consequently higher degrees of adoptive success [5, 6]. Modern research into mobile media consumption has also demarcated accessibility or “perceived convenience” as important variable to consider within the TAM model. Expansions of the TAM model to include accessibility and convenience concerns are valid due to the changing paradigm in which information technology is accessed [7]. Chang’s 2014 study regarding perceived convenience as a construct within the TAM found statistically significant correlations between perceived convenience and perceived usefulness and intention to use, justifying accessibility as a valid construct in the expanded TAM model. While greater degrees of accessibility will not necessarily implicate a higher degree of actual use, lower degrees of accessibility will definitively limit actual use scenarios. Therefore, accessibility can be thought of as a floor indicator for intention to use and subsequent user adoption [8]. 2.1 Continued Intention to Adopt In the expanded TAM model utilized in this study, the dependent variable driven by the constructs of the expanded TAM is Peking University students’ continued intention to

Research on Continued Intention to Adopt E-Learning

565

use the online learning platforms implemented by the 5 + N online learning protocol. Continued intention to use encapsulates the end-user’s overall willingness to engage in the continuing use of the new technology. For the context of this study, continued intention to use represents the students’ willingness to continue to use online learning platforms as a substitute for in-person learning. Continued intention to use has previously been evaluated to be statistically correlated to the perceived usefulness and perceived ease of use constructs in many academic studies [2–4, 8]. The relationships between perceived ease of use, perceived usefulness, and intention to use have been demonstrated to be robust, but the influences of perceived enjoyment and perceived convenience have not been extensively validated to the same degree. To ensure a high degree of adoptive success for online learning platforms, the continued intention to use must not be presented with strong restrictive behaviors beyond those present in traditional learning paradigms. Online learning platforms must foster a sustainable environment for continuous usage to ensure user adoption and long-term viability. 2.2 Perceived Usefulness The first of the independent variables to be examined is perceived usefulness, a core construct of Davis’s 1989 TAM. Perceived usefulness, defined as the degree to which the user believes in the utility provided by the introduced technology is purported to be a determinant for the overall user adoption of new technology, which is consistent with the findings from the previous academic literatures [2–4]. For the purposes of this study, perceived usefulness is a measure of the utility provided by the online learning platforms to the end-user (the Peking University students currently undertaking online learning courses). 2.3 Perceived Ease of Use The second independent variable to be examined is perceived ease of use, defined as the degree to which a person believes that using a particular system, in this case, the online learning platforms provided by Peking University, would be free of effort. This core construct of Davis’s 1989 TAM is primarily determined by the technology-to-user interface and deals with aspects regarding the complexity and learning curve of the introduced technology. Previous studies have acknowledged the effect of perceived ease of use on perceived usefulness, attitude toward usage, and intention to use [2]. 2.4 Perceived Enjoyment Perceived enjoyment is a variable introduced after the inception of Davis’s 1989 TAM. Perceived enjoyment is defined as the degree to which the activity of using technology is perceived to be enjoyable apart from any performance consequences that may be anticipated. In the context, perceived enjoyment is the degree of which students at Peking University find using online platforms to be an enjoyable experience. While previous studies have found perceived enjoyment to be a determinant of attitude toward use and intention to use [7], the influence of perceived enjoyment in an academic setting has not been extensively studied.

566

Z. M. Sheng et al.

2.5 Perceived Convenience Perceived convenience is the fourth variable to be examined in this study as a construct of the expanded TAM. It seems intuitive that higher degrees of perceived convenience would yield greater degrees of user adoption and studies appear to corroborate this claim [8, 9]. Chang’s 2014 study found through structural model analysis that perceived convenience had positive impact on attitude toward use. Perceived convenience was also discovered to be positively correlated with both endogenous TAM constructs of perceived usefulness and perceived ease of use [8]. The following depicts the conceptual framework of this study:

Fig. 1. Conceptual framework

Figure 1 The expanded TAM model used in the study; research framework involving the exogenous variables, perceived enjoyment and accessibility, and endogenous variables, perceived usefulness, perceived ease of use, and continued intention to use e-learning. The research framework of the study involves five constructs of the TAM: perceived usefulness, perceived ease of use, perceived enjoyment, perceived convenience, and continued intention to use (Fig. 1). The endogenous variables of Davis’s 1989 TAM have widely been accepted and understood as a robust and effective model for the evaluation of end-user adoption of new technology [2] but are retested here in the context of the online platforms offered by Peking University to evaluate any potential change in the new domain. Perceived enjoyment has also had a demonstrable effect on the intention to use in studies regarding this construct [4]. However, perceived enjoyment has not been evaluated as thoroughly in the context of education and its relationship with perceived ease of use remains a point of contention in the scientific discussion [4, 6]. Accessibility and convenience are critical elements for the widespread adoption of user technology, particularly when considerations are made for the sustainability of

Research on Continued Intention to Adopt E-Learning

567

the implementation. While perceived convenience has been previously demonstrated to have correlations with attitude toward use and intention to use [8], its effect on attitude toward use and sustainability as the primary method for academic acquisition has not been evaluated extensively on a scope as broad as its current implementations. 2.6 Hypotheses Development H1: Perceived usefulness has a positive and statistically significant effect on Peking University students’ continued intention to use e-learning. H2: Perceived ease of use has a positive and statistically significant effect on Peking University students’ continued intention to use e-learning. H3: Perceived enjoyment has a positive and statistically significant effect on Peking University students’ continued intention to use e-learning. H4: Perceived convenience has a positive and statistically significant effect on Peking University students’ continued intention to use e-learning.

3 Research Method Since the purpose of this study is to infer the factors that influence students’ intention to continue using the e-learning platform by understanding the views of the target population on the e-learning platform, this study adopts a quantitative method. This is mainly because it allows researchers to collect data from the target population as a sample, and then perform data analysis to obtain objective results, thereby summarizing the behavior and characteristics of the target population to answer the research questions [10]. At the same time, the results obtained by this method are statistically significant, so it can better help this study to achieve the original goal. Secondly, due to this study ultimately needing to find out and summarize the relationship between independent variables and dependent variables, this study is explanatory research. According to [10], explanatory research is used to promote the progress of the whole research, thereby helping researchers to further explain and summarize the causal relationship between variables and depict the common characteristics of the current population. On the other hand, this research uses deductive approach to maintain the objectivity and scientific of the research. This is mainly because the deductive approach allows researchers to derive research hypotheses based on theory, and to test and demonstrate the hypotheses by analyzing data. Therefore, it can help this study to verify the research model and hypothesis, to achieve the research goals. Thirdly, this study adopts convenience sampling to establish contact with the respondents in the most convenient and fastest way. Because it allows researchers to intercept respondents anywhere in the university and invite them to participate in the questionnaire survey, or invite the target respondents with the nearest geographic location, at the same time, it also allows the respondents to participate in the survey by filling out online [10]. Therefore, convenience sampling can help researchers and respondents to maintain social distance during the epidemic, and to enable respondents to participate in the questionnaire survey at anytime and anywhere.

568

Z. M. Sheng et al.

Due to Peking University currently has a total of 32,777 students and they all are using online learning platforms right now [11]. When the confidence level is 95% and the marginal error is 5%, the sample size should be 381, which is also meet the standard of sample size suggested by [12–15]. Therefore, the sample size plan for this study is 381. Voluntary participants are enrolled in the study by contact through messaging, social media, or email platforms such as WeChat, Weibo, and QQ. Additional survey respondents are recruited through online forums targeted toward Peking University students. The instrument used for data collection in this study is a modified survey questionnaire that combined items from the questionnaires developed by [2–4, 6, 8]. The research performs systematic literature review to ensure the inclusion and exclusion of literatures are consistent with the research objectives when designing the survey questionnaires [16–19]. The survey questionnaire comprising of two domains included four demographic profile items and 25 items regarding the five constructs of the TAM: perceived usefulness (5 items) [2], perceived ease of use (5 items) [2], perceived enjoyment (5 items) [4, 6], perceived convenience (5 items) [4, 8], and continued intention to use (5 items) [4]. SPSS statistical software is used to evaluate the collected data from questionnaire survey and to ensure the quality of the collected data, pilot testing is used to test a small sample of data in advance. Then, reliability and validity tests aim to make sure that the data are reliable and valid. While descriptive statistics aims to describe the basic information of the respondents, and the test of Pearson correlation coefficients aim to verify the direction and closeness of the relationship between variables. Finally, the multiple regression analysis aims to test the validity of the research framework and the relationship between variables, to validate the hypotheses proposed in this study.

4 Discussion and Conclusion 4.1 The Reliability Test According to the Table 1, it showed the final results of 383 data in reliability test. Firstly, the α value of perceived usefulness and ease of use are 0.908 and 0.950 respectively, which are excellent. Then, the α value of perceived enjoyment is 0.795, which is good. Thirdly, the α value of perceived convenience is 0.906, which is excellent. While the α value of continued intention to use is 0.945, which is excellent too. As the α value of all variables in this study are more than 0.7, so the data are all reliable and good based on the standard of reliability test. 4.2 Pearson’s Correlation Analysis According to the Table 2, the r value of perceived usefulness and continued intention to use is 0.816, so there is a high positive correlation between them according to the standard of Pearson’s correlation coefficients mentioned earlier. Then, the r value of perceived ease of use and continued intention to use is 0.581, therefore, there is a moderate positive

Research on Continued Intention to Adopt E-Learning

569

Table 1. Summary of Cronbach’s Alpha results (N = 383) Variable

No. of items

Cronbach’s Alpha

Perceived usefulness

5

0.908

Perceived ease of use

5

0.950

Perceived enjoyment

5

0.795

Perceived convenience

5

0.906

Continued intention to use

5

0.945

correlation between these two variables. While the r value of perceived enjoyment and continued intention to use is 0.453, so there is a low positive correlation between them. On the other hand, the r value of perceived convenience and continued intention to use is 0.598, therefore, there is a moderate positive correlation between these two variables. In general, there are positive correlation between the independent variables and dependent variable in this study. Table 2. Pearson’s correlation analysis (N = 383) Variable

PU

PEOU

PE

PC

CITU

PU

1

0.504**

0.436**

0.487**

0.816**

PEOU

0.504**

1

0.471**

0.751**

0.581**

PE

0.436**

0.471**

1

0.525**

0.453**

PC

0.487**

0.751**

0.525**

1

0.598**

CITU

0.816**

0.518**

0.453**

0.598**

1

Notes ** p < 0.001, * p < 0.05

4.3 Multiple Regression Analysis In the model of this study, the dependent variable is continued intention to use (CITU), while the predictors are perceived usefulness (PU), perceived ease of use (PEOU), perceived enjoyment (PE) and perceived convenience (PC). According to the Table 3, the R2 is 0.722 and F value is 245.96, it means that the continued intention to use can be explained about 72.2% by PU, PEOU, PE and PC, and the effect size of the research model is very large. On the other hand, PU (β = 0.668, t = 20.407, p = 0.000), PEOU (β = 0.090, t = 2.114, p = 0.000), PC (β = 0.196, t = 4.540, p = 0.000). As the p-value of PU, PEOU and PC are all less than 0.05, so there is a positive and significant relationship between them and CITU. But PE (β = 0.017, t = 0.506, p = 0.613). As the p-value of PE is more than 0.05, so there is no positive and significant relationship between it and CITU.

570

Z. M. Sheng et al. Table 3. Multiple regression analysis

Hypothesis

β

t-value

Sig

Perceived usefulness → Continued Intention to Use

0.668

20.407

0.000**

perceived ease of use → Continued Intention to Use

0.090

2.114

0.035*

perceived enjoyment → Continued Intention to Use

0.017

0.506

0.613n.s

Perceived convenience → Continued Intention to Use

0.196

4.540

0.000**

R2 = 0.722, F-value = 245.96 Note ** p < 0.001, * p < 0.05, n.s = not significant, β = standardized beta

This research finally achieved the goal by analyzing the 383 data from respondents. According to the above analysis results, it showed the data of all variables are reliable. Then, the data of this study are roughly normal distributed. At the same time, the research model is a good fit as the R square is more than 0.6, and the dependent variable can be totally explained about 72.2% through the independent variables in this study. Finally, in addition to PE, there is a positive and significant relationship between PU, PEOU, PC and CITU as their p-value are less than 0.05. Therefore, the H1, H2 and H4 are accepted. In the theoretical implication, the TAM model can be expanded by adding two functional factors to predict the intention to adopt of new technology, such as perceived enjoyment and convenience. At the same time, the research related to usage intention or technology could adopt the conceptual model proposed in this study, to better understand the views and feelings of users. This is mainly because ease of use and usefulness have become the basic functions of applications and platforms with the advancement of technology, therefore, modern people pay more and more attention to the entertainment and convenience of this IT software. Hence researchers should adopt as many functional and external factors as possible to improve the research framework. In the practical implication, the education department of government should formulate relevant policies to encourage universities to strengthen the training and introduction of technical talents, to further improve the education platform and optimize its functions to ensure the development of online learning. For example, the government can provide additional economic subsidies and benefits for technical talents who are willing to work at universities and establish a special recruitment system to lower the threshold for examinations. Then, the education department can also set up technology-related training courses and regularly require teachers to participate in learning to improve technology skills. Thirdly, the universities should also further strengthen the echelon construction of technical talents, not limited to full-time teachers, to ensure the digital transformation of teaching. For example, in addition to general IT maintenance personnel, universities can also recruit senior software developers from the society to develop learning applications and platforms that meet their own needs. At the same time, education departments and universities can also develop in-depth cooperation with the Internet or IT companies to accelerate the integration of education and technology, such as through regular exchange meetings, new technology, and new application exhibitions.

Research on Continued Intention to Adopt E-Learning

571

On the other hand, the designers of the e-learning platform should further optimize the ease of use and usefulness of the platform to increase the user’s viscosity as these two factors have a significant and positive influence on continued intention to use. Particularly, when students feel that using the e-learning platform can improve their learning performance, they will be more willing to use it [20]. Then, the designers can optimize account association and one-click login functions, so that students can quickly log in or log out of the platform, thereby improving their perception of ease of use. For example, the login method of the platform can include scanning, entering the account number and verification code to log in. While the login account can include student ID, email address and WeChat account. After that, the designers can add more entertainment features to enrich online learning activities and learning models, thereby attracting more students to use the e-learning platform. For example, set up some simple learning levels to encourage students to learn words or mathematical formulas by clearing the levels. In addition to the flexibility of time and location, the designers can further emphasize the network convenience of the e-learning platform, such as no Internet connection, offline use, so that students can use the platform even when there is no Internet. Finally, the designers can further apply 5G to strengthen the signal when students use the e-learning platform, while further ensuring the stability of video interaction.

References 1. Volery, T., Lord, D.: Critical success factors in online education. Int. J. Educ. Manag. 14(5), 216–223 (2000). https://doi.org/10.1108/09513540010344731 2. Davis, F.D.: Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Q. 13(3), 319–340 (1989). https://doi.org/10.2307/249008 3. Alfadda, H.A., Mahdi, H.S.: Measuring students’ use of zoom application in language course based on the technology acceptance model (TAM). J. Psycholinguist. Res. 50(4), 883–900 (2021). https://doi.org/10.1007/s10936-020-09752-1 4. Teo, T., Noyes, J.: An assessment of the influence of perceived enjoyment and attitude on the intention to use technology among pre-service teachers: a structural equation modeling approach. Comput. Educ. 57(2), 1645–1653 (2011). https://doi.org/10.1016/j.compedu.2011. 03.002 5. Venkatesh, V., Davis, F.D.: Determinants of perceived ease of use: integrating control, intrinsic motivation, and emotion into the technology acceptance model. Manage. Sci. 46(2), 186–204 (2000). https://doi.org/10.1287/mnsc.46.2.186.11926 6. Dickinger, A., Arami, M., Meyer, D.: The role of perceived enjoyment and social norm in the adoption of technology with network externalities. Eur. J. Inf. Syst. 17(1), 4–11 (2008). https://doi.org/10.1057/palgrave.ejis.3000726 7. Cham, T.H., Cheng, B.L., Ng, C.K.Y.: Cruising down millennials’ fashion runway: a crossfunctional study beyond Pacific borders. Young Consumers 22(1), 28–67 (2020) 8. Chang, C.C., Tseng, K.H., Liang, C., Yan, C.F.: The influence of perceived convenience and curiosity on continuance intention in mobile English learning for high school students using PDAs. Technol. Pedagog. Educ. 22(3), 373–386 (2014). https://doi.org/10.1080/1475939X. 2013.802991 9. Yoon, C., Kim, S.: Convenience and TAM in a ubiquitous computing environment: the case of wireless LAN. Electron. Commer. Res. Appl. 6(1), 102–112 (2007). https://doi.org/10.1016/ j.elerap.2006.06.009

572

Z. M. Sheng et al.

10. Sekaran, U., Bougie, R.: Research Methods for Business, 7th edn. Wiley, Chichester, West Sussex, United Kingdom (2016) 11. Peking University: General Information. Peking Universityhttps://xxgk.pku.edu.cn/ (2021) 12. Comrey, A.L., Lee, H.B.: A first course in factor analysis (2nd ed.). New York, NY: Erlbaum, L. (1992) 13. Cham, T.H., Cheah, J.H., Ting, H., Memon, M.A.: Will destination image drive the intention to revisit and recommend? Empirical evidence from golf tourism. Int. J. Sports Mark. Spons. 23(2), 385–409 (2022) 14. Cheng, B.L., Shaheen, M., Cham, T.H., Dent, M.M., Yacob, Y.: Building sustainable relationships: service innovation at the pinnacle of touristic achievement. Asian J. Bus. Res. 11(1), 80–90 (2021) 15. Cham, T.H., Cheah, J.H., Cheng, B.L., Lim, X.J.: I Am too old for this! Barriers contributing to the non-adoption of mobile payment. Int. J. Bank Mark. 40(5), 1017–1050 (2022) 16. Eng, P.H., Tee, W.S.: Impact of consumer privacy concern and privacy related defensive behaviour on the adoption of social media platform. Glob Bus Manage Res. Int. J. 14 (1), 171–184 (2022) 17. Eng, P.H., Chew, B.C., Syaiful, R.H.: Case study for skills management approach to manage and retain the highly-skilled blue collar workers. Int. Bus. Manage. 10(16), 3558–3566 (2016) 18. Eng, P.H., Syaiful, R.H., Md, N.T.: Poverty line income and implementation of minimum wage policy in Malaysia. Int. J. Bus. Technopreneurship. 3(2), 243–260 (June 2013) 19. Eng, P.H., Syaiful, R.H., Md, N.T.: Historical review of minimum wage policy in the developed countries: implementation of national minimum wage policy in Malaysia. Int. J. Bus. Technopreneurship. 3(3), 387–411 (Oct 2013) 20. Hosen, M., Ogbeibu, S., Giridharan, B., Cham, T.H., Lim, W.M., Paul, J.: Individual motivation and social media influence on student knowledge sharing and learning performance: evidence from an emerging economy. Comput. Educ. 172, 104262 (2021)

Intelligent Health Informatics

Real-Time Healthcare Surveillance System Based on Cloud Computing and IoT Radwan Nazar Hamed1 and Azmi Shawkat Abdulbaqi2(B) 1 College of Computing and Information Technology, University of Anbar, Ramadi, Iraq

[email protected] 2 College of Computer Science and Information Technology, University of Anbar, Ramadi, Iraq

[email protected]

Abstract. This article illustrates the use of an IoT-based, real-time remote Biosignals monitoring system with minimal power consumption. This work represents an advancement in the realm of remote health monitoring. Every year, more people need health care, and the traditional bio-signals surveillance systems need patients to be present in person within hospitals. This might make it difficult to care for the patients, particularly those who have serious and fragile medical issues. Internet technology and current electronic devices may thus provide useful solutions in this area. Based on that, this project uses a mobile application as an IoT platform to remotely monitor the heart rate and a live ECG signal. Using a microcontroller-based device, the signals are measured and processed (Arduino). The primary contribution of this study is the transmission of an electrocardiogram (ECG signal) to a particular mobile device (MobiDev) and the transmission of these signals to the Cloud Alarm Server (CAS) for expert observation. This helps to diagnose cardiac disease before the worst may happen. Finally, both mobile devices and personal computers (PCs) are used to show the project’s results. The recommended approach helps specialists keep track of their patient’s problems by providing ECG data to professionals’ MobiDev in an understandable manner. The testing results of the proposed system demonstrated the system’s ease of use and clinicians’ ability to interact with patients directly across vast distances and monitor their health in real-time. Keywords: Electrocardiogram (ECG) · Healthcare Surveillance · Cloud Alarm Server(CAS) · Mobile Device (MobiDev) · Multimedia Messaging Service (MMS)

1 Introduction This article illustrates the use of an IoT-based, real-time remote Biosignals monitoring system with minimal power consumption. This work represents an advancement in the realm of remote health monitoring. Every year, more people need health care, and the traditional bio-signals surveillance systems need patients to be present in person within hospitals. This might make it difficult to care for the patients, particularly those who have serious and fragile medical issues. Internet technology and modern electronics © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 575–584, 2023. https://doi.org/10.1007/978-3-031-20429-6_52

576

R. N. Hamed and A. S. Abdulbaqi

might thus provide practical answers in this field. In order to remotely monitor the heart rate and a live ECG signal, this project makes use of a mobile application as an IoT platform. The signals are measured and processed by a microcontroller-based device (Arduino). This study’s main contribution is sending an electrocardiogram (ECG signal) to a specific mobile device (MobiDev) and sending those signals to the Cloud Alarm Server (CAS) for professional evaluation. This helps to diagnose cardiac disease before the worst may happen. Finally, both mobile devices and personal computers (PCs) are used to show the project’s results. The recommended approach helps specialists keep track of their patients’ problems by providing ECG data to professionals’ MobiDev in an understandable manner. The testing results of the proposed system demonstrated the system’s ease of use and clinicians’ ability to interact with patients directly across vast distances and monitor their health in real-time. IoT applications include smart wearables, smart cities, home automation, remote control, and monitoring systems have been growing over time. IoT technology has also advanced in recent years in healthcare and fitness applications. This is a crucial step in modernizing healthcare systems that now need patients to check their health metrics inside of clinics or hospitals. On the other side, this technology has significant drawbacks, including data management, security, privacy, and the human-cloud interface. To address the aforementioned issues, we devised and built a scheme based on MobiDev that concentrates on surveillance of the ECG signaling, and alarms tasks. In our method, ECG is captured and surveillance in real-time by an ECG instrument and tested in [6]. MobiDev detects alarms and sends alarm messages (AlmMsg) and ECG signaling to the CAS. When CAS receives a message, it sends it to the specialists’ MobiDev via the method for notification on android. Through this notification, clinicians may view the alarm specifics as well as a high-quality ECG signaling. Specialists might get ECG alarm information practically anywhere in time utilizing simply a mobile in this manner. The next shows how the manuscript is organized. Section 2 gives a summary of the proposed system design and operation mode. Section 3 describes the system’s implementation, which includes ECG data processing on MobiDev, CAS, and the alarming procedure. In Sect. 4, we present the operational findings in a real-world setting. Section 5, concludes with a summary of the benefits and drawbacks.

2 Related Works Sudden death might result from acute medical issues. Because one or more bodily organs are malfunctioning, many individuals die abruptly all around the globe. These organs cause aberrant symptoms before they occur. Sensors can identify and gather these signals, also known as bio-signals. Depending on the extent of the search, IoTbased remote healthcare systems rely on microcontrollers and gateways to upload data, and the Arduino microcontroller family is commonly utilized in this sector. Many projects make use of the ESP8266-based NodeMCU board [8]. Others [11] use an ESP8266 module with an Arduino UNO. In certain configurations, an Arduino board and a Raspberry Pi board are coupled. Processing is handled by the Arduino board, while a gateway is provided by the Raspberry Pi [12]. In projects, the Raspberry Pi is also used to analyze data and upload it to an IoT server [14]. On the other hand, some designs

Real-Time Healthcare Surveillance System

577

employ mobile phones as a gateway and depend on the Arduino board for processing [15]. Wired or wireless sensors connected to the microcontroller (as in the majority of projects [8–14]). (as in the case of the HC-06 Bluetooth module [15]). There are several configurations and sizes for IoT systems. Some of them were created from the ground up utilizing web design tools like the Node.js server, the WebSocket library, an HTTP server, and a web-based GUI interface [8, 10, 13]. Additionally, a lot of projects use Android Studio, providing mobile app developers with a fantastic chance to build healthcare apps [9, 12]. For instance, “Abuelómetro” is the name of one smartphone app, while “3rd Nurse” is the name of another. Numerous further smartphone apps have been created for IoT projects in general. This discovery could encourage healthcare institutions to use programs like the Blynk smartphone app [15]. There are two possibilities for storing data. An alternative is to use a web-based (cloud-based) database like MySQL or Google Firebase Database. [10, 12]. The mobile apps server is required by the other projects, on the other hand. This approach is meant to be less complicated than the one preceding it. For instance, the servers of Blynk and Abuelómetro contain healthrelated data [9, 15]. Heart rate and body temperature are only two of the many aspects of a person’s health that are evaluated. The heart rate is measured by analog [8, 10, 11, 15] and digital [9, 14] sensors, the latter of which may track the SPO2. Second, body temperature is utilized using the thermistor NTC [15] and the LM35 [8, 10, 11] sensors. One possible measure for measuring skin conductivity is the galvanic skin response (GSR) [12]. People who are emotionally distressed have higher skin conductivity. Additionally, some apps use the “iBeacon module” rather of GPS, which is less reliable inside, to detect patients in constrained places [14]. The ADXL335 (3-Axis Accelerometer) fall detector sensor is used in many applications to monitor patient safety when a patient falls unexpectedly to the ground [14]. The AD8232 ECG sensor is used by one prototype [13]. This article does not discuss the advantages of real-time ECG monitoring for medical purposes. On a web page, the ECG signal is shown in its simplest form without any additional images. It is challenging to keep track of several ECG signals. Poor quality was also seen in the ECG signal. This problem arose because certain components of the ECG signal, such as P-Q-R-S-T, were absent from the output signal while others were not included. Last but not least, the signal in this inquiry was not precisely and practically tested. As a consequence, it is impossible to rely on or use the signal produced in this research for diagnostic purposes. Finally, there are three categories into which the security and privacy techniques discussed in this article might be divided. Several healthcare apps, for instance, need login information (username and password) in order to access patient health data [9, 12]. On the other hand, other initiatives concentrate on designing their user interface for a certain device in order to ensure that access is limited to people who use that device. Certain mobile apps [15] allow users to accomplish this.

3 The Planned System Model A portable instrument with an ECG sensor for gathering and analyzing ECG signaling, a MobiDev to serve as a center of processing to detect and send alarms, a CAS to respond to alarm alerts and report to specialists, and MobiDevs to be managed by specialists and members of the family to get signals of alarm are the four components of the suggested

578

R. N. Hamed and A. S. Abdulbaqi

technique. The system design is presented in Fig. 1. A wireless communication channel is maintained between the ECG instrument and the MobiDev to send real-time ECG signaling. Due to its high performance on MobiDevs, BT was chosen. A TCP connection is built between the MobiDev and the CAS when an alert occurs, which is utilized to convey AlmMsgs as well as ECG signaling. A long IP connection is maintained between the CAS and the client phones at all times, which could ensure that the AlmMsgs arrive on time. When a client’s phone receives an alarm notification, the image of an ECG alarm is obtained through the connection of an HTTP to CAS. This method allows specialists’ MobiDevs to receive ECG signaling.

Fig. 1. Feature extraction of ECG signal

4 Dataset The MIT-BIH PhysioBank database is utilized to get the signals needed for the investigation. To investigate, signals from two databases were utilized: Congestive Heart Failure RR Interval Database (chf2db) and Normal Sinus Rhythm RR Interval Database (nsr2db). Multiple signals in various forms are stored in the PhysioBank database. It establishes a free-of-charge right to utilize data.

5 System Methodology The proposed technique continuously scans these sample windows, then filters them according to the noise in the ECG signal and the patient’s artifact during the day. The method calls for sending all raw ECG data to a remote server for processing. The sensor node detects medical data transferred from the patient to the specialist and vice versa, and it serves as a link between the transmitter (patient/specialist) and receiver (patient/specialist). Window sampling takes 5 s (1000 samples, 200 Hz sampling rate), while filtering consumes 9Kbytes of RAM.. The next module examines the duration of PQ, QRS, and QT wave patterns and analyzes their changes to monitor cardiac electrical activity for damage or sickness.

Real-Time Healthcare Surveillance System

579

5.1 ECG Signals Filtering Two types of noise may be filtered out of an ECG signal: biological noise and environmental noise. Physiological intervention, for example, low-frequency noise, causes biological noise (baseline drifts less than 0.03 Hz). The human body’s respiration and muscular movement may create a high-frequency EMG noise (1–5000 Hz). An additional source of noise may be the discovery of the ECG electrodes migrating from their original placements, or the discovery of motion artifacts. Instruments and circuit components, for example, interference of the power line (50–60 Hz), noise of the electrode contact, and noise of the electrosurgical, as well as the noise of the radio frequency, all contribute to ambient noise. The range of the frequency of the ECG wave is 0.1 to 250 Hz. The majority of the relevant research utilizes multi-stage filters, for example, low pass, high pass, and notch filters. This approach only recognizes the QRS complex, with the P-wave and T-wave being overlooked due to their low frequency. An additional method is to utilize a wavelet, which is useful at denoising and can identify all of the features of an ECG signal. Both systems, on the other hand, are unsuitable for wearable sensor networks for two causes: first, they demand a large amount of storage capacity, which not all sensor types have; and second, they require a large amount of computational processing, which consumes more power from the sensor. The QRS complex is the high-frequency component of the ECG signal; the derivate action was first established to improve the QRS complex’s high-frequency features and offer information on the complex’s slop. 5.2 Extraction of Features The ECG segment with thousand samples (A five-second window with a sampling frequency of 200 Hz) of data is obtained to get the characteristics; the size of the window is selected such that there are numerous beats present (5 beats at least). Q, R, S, P, and T peaks must be identified to measure ECG features, for example, R-R intervals, QRS complex, and width. We develop a low-cost peak identification technique based on the methodology of lightweight feature extraction. The identification of the R- peaks is the most important feature of this technique; other aspects are dependent on the detection of these peaks. The following is a summary of the search algorithm: (1) Find a maximum value that describes one of the R-peaks inside the sampling frame (MAX). (2) Within the sampling frame, look for a minimum value that represents one of the S-peaks (MIN). (3) Determine thresholding with the following values: thresholding R = MAX/2 and thresholding S = MIN/2. (4) Find Ri peaks that are over the R thresholding throughout the sampled window. (5) Create a sub-window for each Ri [Ri to Ri + 10]. (S-peaks come a few samples after R-peaks) data to look for S-peaks that should be below the thresholding S. (6) Create a sub-window for each Ri [Ri Ri-10]. (Q-peaks usually emerge a few samples before R-peaks) samples to look for Q-peaks. The QRS complex has now been discovered. (7) To locate P-peaks generate a window [Ri-100 to Ri-25] and look for the largest value (P-peaks emerge before Q-peaks). (8) To locate T-peaks Search for the largest value in the timeframe [Ri + 25 to Ri + 100] (T-peaks occur after the S-peaks).

580

R. N. Hamed and A. S. Abdulbaqi

6 System Design and Implementing 6.1 Process of ECG Data Utilizing MobiDevs The authors in [6] define the ECG data frame format. The ECG data frame contains a 7-lead ECG with three types of alerts. MobiDev collects and the Bluetooth (BT) input stream is utilized to extract ECG data. And sets alarms for it. To buffer these data, we built two buffers on MobiDev. The data from the BT channel is read and stored in BUFFER1(Buff-1). On the data in Buff-1, A thread that detects alarms executes a detection action. MobiDev starts the alarming software if an alarm is noticed. In the following part, we’ll go through how it works. The alarm detection thread sends ECG data from Buff-1 to Buff-2 by detecting alarms. The ECG data is obtained from Buff-2 by the user interface thread, which then displays the dynamic ECG on the screen. The process is depicted in Fig. 2.

Fig. 2. The general system architecture

7 Designing and Implementation of CAS In the system, CAS delivers dependable cloud services. It takes care of MobiDev’s AlmMsgs and delivers them to specialists’ phones. As shown in Fig. 3, our CAS is made up of four parts: Message Receive Server (MRS), Alarm Information Management (AIM), Message Push Server (PshSrv), and Web Server (WbSrv). The AlmMsgs are received via MRS. A copy of the alarm will be stored in AIM if one was received. WbSrv will handle ECG signaling, while PshSrv will send the AlmMsg to the specialists’ phones. After receiving this message, the specialist might utilize his or her phone to see ECG signaling from WbSrv. Medical data is personal information, hence the system’s security is critical. We configured authentication keys on both sides of the CAS for the user’s access 1 and 2 in Fig. 3. The CAS could only be accessed by the phone that has been authenticated. For 3), WbSrv keeps an IP list that is authenticated utilizing Key1 and Key2. Only those IP addresses were allowed to access WbSrv’s ECG signaling. Table 1 shows the findings of time delay and alarm reliability testing.

Real-Time Healthcare Surveillance System

581

Fig. 3. The Infrastructure of ECG data processing based on mobile devices

Table 1. Time delay and alarm reliability testing No

Zain network mobile/5G

No

Time of the delay (ms)

Asia network mobile/5G Time of the delay (ms)

A

1903

A

440

B

1707

B

2674

C

1636

C

3226

D

1635

D

2439

E

1620

E

2844

F

7379

F

388

G

2360

G

472

H

1981

H

5906

I

1943

I

762

J

1613

J

3165

8 Processing of Alarms After detecting an alarm, the Android processing center retrieved the necessary characteristics and read 5s’ ECG data in both Buff-1 and Buff-2, yielding 10s’ ECG data. The waveform is then drawn on a bitmap with a resolution of 1400 × 600. Android, on the other hand, utilized the Location API to get geographic data. We utilize BAIDU’s free location API in our system since it integrates nicely with Android Location API and provides a text location message service.

582

R. N. Hamed and A. S. Abdulbaqi

To send this message, MobiDev requires a TCP connection to CAS. CAS creates the connection after authenticating its identification. The AlmMsg is then broadcast across the channel. ECG signaling is then uploaded to a designated directory on WbSrv. We utilize the TCP protocol to transport AlmMsgs and ECG signaling to CAS because it is a dependable protocol that ensures that AlmMsgs and ECG signaling reach safely and on time. A lengthy IP connection is maintained between the CAS and the specialist’s mobile so that PshSrv may send the message and ECG signaling URL to the specialist’s phone on time. An Android WebView might be utilized to display ECG signaling via the URL when the specialist reads the message on his phone. To transport ECG signaling, an HTTP connection is constructed. Figure 3 depicts the whole alerting procedure in the system. Given that the message recipient is a MobiDev, not a PC, and that power and network resources are restricted, we must utilize a sensible method to push messages to the client MobiDev, rather than forcing the phone to fetch messages from CAS on its own. As a finding, the open-source project Android was incorporated into PshSrv [7]. Android is an open-source project that includes an XMPP-based notification server and a client tool kit to offer push notification capabilities for Android. On the Android platform, an effective push mechanism is implemented. It was reformed in our system to meet our objectives, and it performed well in terms of IP long connection maintenance, multi-user administration, and resource usage.

9 Experimentation Based Findings In the ST part, there is an actual test setup as well as an anomalous alert scenario. Alarms are recorded and sent to the CAS. Customers’ alert messages are received on a Samsung Galaxy mobile. The message is sent as a ringing and vibrating notice. When the notice is presented, facts about the alert and ECG signs are displayed on the screen. Two ways encrypt and decode ECG signals. Reserving a domain in the Cloud for the user allows us to conduct all processing within the Cloud. This technology is similar to on-site processing (This is what our paper is based on). Before being delivered to a doctor for diagnosis, ECG signals are uploaded to the cloud for primary treatment, encryption, and feature extraction. After decrypting ECG signals, the expert gives the patient diagnostic feedback. The second method includes conducting all operations locally (on a computer) and transmitting the decrypted ECG signal to an expert for diagnosis.

10 Conclusion In this project, we planned and built an ECG-based surveillance and warning system based on Android, focusing on the system’s architecture and alarm processing, and we tested its reliability and temporization in a real-world setting. Based on the findings, we can conclude that the system has a friendly design and high reliability. Connected to a powerful network environment for mobile networks, of course. There is also room to optimize energy consumption on the client side of MobiDev’s message reception. In all cases, we believe it has some value in terms of applicability for healthcare professionals in the treatment of cardiovascular diseases and diseases of the cardiovascular system.

Real-Time Healthcare Surveillance System

583

References 1. Karaca, Y., Moonis, M., Zhang, Y.D., Gezgez, C.: Mobile cloud computing-based stroke healthcare system. Int. J. Inf. Manage. 45, 250–261 (2019) 2. Thota, C., Sundarasekar, R., Manogaran, G., Varatharajan, R., Priyan, M.K.: Centralized fog computing security platform for IoT and cloud in the healthcare system. In: Fog Computing: Breakthroughs in Research and Practice, pp. 365–378. IGI Global (2018) 3. Liu, Y., Zhang, Y., Ling, J., Liu, Z.: Secure and fine-grained access control on e-healthcare records in mobile cloud computing. Futur. Gener. Comput. Syst. 78, 1020–1026 (2018) 4. Roy, S., Das, A.K., Chatterjee, S., Kumar, N., Chattopadhyay, S., Rodrigues, J.J.: Provably secure fine-grained data access control over multiple cloud servers in mobile cloud computing based healthcare applications. IEEE Trans. Industr. Inf. 15(1), 457–468 (2018) 5. Ali, A.A., Omran, N.F., Alsayied, A.A.: EgyHealth System: A Mobile Healthcare System Application with Android and Cloud Computing. In: Hassanien, A.-E., Chang, K.-C., Mincong, T. (eds.) AMLTA 2021. AISC, vol. 1339, pp. 803–815. Springer, Cham (2021). https:// doi.org/10.1007/978-3-030-69717-4_74 6. Ganatra, N.P., Patel, R.S.: Proposed customized architecture of mobile cloud computing in health care domain. Int. J. Adv. Res. Comput. Sci. 8(5) (2017) 7. Abdulbaqi, A.S., Obaid, A.J., Mohammed, A.H.: ECG signals recruitment to implement a new technique for medical image encryption. J. Discrete Math. Sci. Cryptogr. 24(6), 1663–1673 (2021) 8. Škraba, A. Koložvari, A., Kofjac, D., Stojanovic, R., Stanovov, V., Semenkin, E.: Prototype of group heart rate monitoring with NODEMCU ESP8266. In 2017 6th Mediterranean Conference on Embedded Computing (MECO), pp. 1–4 (2017) 9. Durán-Vega, L.A., et al.: An IoT system for remote health monitoring in elderly adults through a wearable device and mobile application. Geriatrics 4(2), 34 (2019) 10. Wan, J., et al.: Wearable IoT enabled real-time health monitoring system. EURASIP J. Wirel. Commun. Netw. 2018(1), 1 (2018). https://doi.org/10.1186/s13638-018-1308-x 11. Krishnan, D.S.R., Gupta, S.C., Choudhury, T.: An IoT-based patient health monitoring system. In: 2018 International Conference on Advances in Computing and Communication Engineering (ICACCE), pp. 01–07. IEEE (2018) 12. Hamim, M., Paul, S., Hoque, S.I., Rahman, M.N., Baqee, I.-A.: IoT based remote health monitoring system for patients and elderly people. In: 2019 International Conference on robotics, Electrical and Signal Processing Techniques (ICREST), pp. 533–538. IEEE (2019) 13. Yang, Z., Zhou, Q., Lei, L., Zheng, K., Xiang, W.: An IoT-cloud based wearable ECG monitoring system for smart healthcare. J. Med. Syst. 40(12), 286 (2016) 14. Abdelgawad, A., Yelamarthi, K., Khattab, A.: IoT-based health monitoring system for active and assisted living. In: International Conference on Smart Objects and Technologies for Social Good, pp. 11–20. Springer (2016) 15. Tastan, M.: IoT based wearable smart health monitoring system. Celal Bayar Üniversitesi Fen Bilimleri Dergisi 14(3), 343–350 (2018) 16. Aliyu, A., Abdullah, A.H., Kaiwartya, O., Tayyab, M., Joda, U.M.: Mobile cloud computing: layered architecture. In: 2018 Seventh ICT International Student Project Conference (ICTISPC), pp. 1–6. IEEE (2018) 17. Aliyu, A., Abdullah, A.H., Kaiwartya, O., Hussain Madni, S.H., Joda, U.M., Ado, A., Tayyab, M.: Mobile cloud computing: taxonomy and challenges. J. Comput. Netw. Commun. (2020) 18. Abdulbaqi, A.S., Obaid, A.J., Alazawi, S.A.H.: A smart system for health caregiver based on IoMT: toward tele-health caregiving. Int. J. Online Biomed. Eng. 17(7) (2021)

584

R. N. Hamed and A. S. Abdulbaqi

19. Bale, A.S., Gupta, A., Prajwal, R., Manoj, R., Gowda, M.M., Khatokar, J.A., Ramasamy, A.: Mobile cloud computing-enabling technologies and applications. In: 2021 6th International Conference on Signal Processing, Computing and Control (ISPCC), pp. 491–496. IEEE (2021)

A Review of Agent-Based Model Simulation for Covid 19 Spread Samar Ibrahim(B) British University in Dubai, Dubai, UAE [email protected]

Abstract. Containing the implications of the Covid-19 pandemic has and continues to be a priority around the globe. Pursuant to this, scientists have focused on understanding the virus spread’s behavior and patterns to develop mitigation plans. Artificial Intelligence agent-based simulations (ABS) have been used by scientists to simulate virus spread and control. ABS can model different variables and decisions in specific environments and contexts to find ways to reduce the transmission and diminish the severity of this pandemic. This paper presents a review of literature on ABS modeling to contain the virus through pharmaceutical and non-pharmaceutical interventions and the impact of the virus spread on economies. Keywords: Covid-19 · NetLogo · Multi-Agent Simulation · Non-Pharmaceutical Interventions

1 Introduction Corona Virus Disease (COVID-19) is a frightening pandemic discovered in early December 2019 in Wuhan, China, and spreading rapidly across the globe. The World Health Organization (WHO) described this outbreak as a severe global threat [1]. The health crisis caused by this disastrous pandemic has globally and severely affected various sectors. In 2020 only, the global economy diminished by 3%, with a significant loss estimated at around 9 trillion USD [2]. Many measures were taken to stop the pandemic from spreading, such as lockdown, entire whole or part closures, wearing masks, social distancing, and quarantine. Other measures implement methods to track the virus to decrease infection cases and decrease fatality. Many researchers and scientists have embarked on using system simulation to provide different hypothetical scenarios to present and analyze the dynamics of Covid-19 infection. These simulations introduce problems with affecting solutions aiming at finding ways and ideas to reduce the transmission and diminish the severity of this pandemic [3]. The majority of these tools are based on mathematical modeling, including SIR (Susceptible Infected and Recovered) model and its extensions SEIR (Susceptible, Exposed, Infected, and Recovered), which are two compartmental models used in epidemiology [3]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 585–602, 2023. https://doi.org/10.1007/978-3-031-20429-6_53

586

S. Ibrahim

Multi-agent-based systems are a well-known branch of artificial intelligence. This branch of artificial intelligence has existed for a long time [4]. The definition of an intelligent agent may be generalized as a computerized program capable of reasoning, making independent decisions on its own, collaborating with other intelligent agents when necessary, getting information about the environment in which it operates, and taking action to achieve its aim [4]. It has been asserted that Agent-based Simulation is instrumental in assisting all kinds of mitigation strategies during the Covid-19 pandemic. Several studies have investigated the spread, tracing, and transmission of Covid-19 using agent-based simulation models. Since these models consider existing situations that can support policymaking and provide a platform to experiment with different approaches based on an artificial population [5]. They are very diverse in terms of the number of simulated individuals, their goal, the geography they represent, and how they model diseases, transmission dynamics, and human behavior. Their goal is to provide insight into the spread of diseases among autonomous individuals with heterogeneous characteristics and behaviors interacting with each other in a shared environment [1]. Hence, these models allowed for cost and time-efficient evaluation of various possible interventions and developed a better understanding of the pandemic that could identify different interventions for halting its spread [3]. Generally, most studies related to agent-based model simulation and covid 19 virus have the same predominant objective; they seek to estimate the spread of Covid-19 over time [6]. In addition, many of these studies have a specific purpose to simulate models and explore scenarios for Covid 19 and the impact of non-pharmaceutical interventions (NPI), such as quarantine, isolation, social distancing, using masks or contact tracing on the virus spread [7]. Some of which investigate and recommend a scenario of a particular combination of these interventions, such as simulating the universal use of masks with social distance. Other agent-based model simulation studies investigated the lockdown strategy to mitigate the pandemic and its impacts on various health and economic issues and other studies that simulate testing policies or reopening universities or towns. Very limited number of studies develop agent-based model and simulate the effect of the virus spread by implementing pharmaceutical interventions (PI) such as vaccination strategies. Some reviews examine studies implementing artificial intelligence (AI) techniques to detect and classify Coronavirus disease 2019 (COVID-19) [8, 9]. Other reviews analyze studies related to using sentimental analysis to mitigate the covid-19 [9]. This critical survey aims to provide a critical review of studies that analyze and simulate Covid 19 spread over time and investigate different methods of interventions. Section 2 presents a literature review on agent-based simulation for modeling pedestrian dynamics during the Covid 19 pandemic Papers are categorized according to the objective of the simulation in these studies: Contain through Non-Pharmaceutical Interventions, the virus spreads impact on the economy, contain through Pharmaceutical Interventions. Section 3 presents conclusions based on the reviewed papers. Finally, Sect. 4 suggests a proposition for future studies.

A Review of Agent-Based Model Simulation

587

2 Literature Review This section provides an overview of the reviewed studies arranged by their general objectives. These studies are surveyed and coded as listed in Appendix A. This appendix also captures the domain, aim, case studies, simulator used, the key features of agent type, and any additional simulation model used in these papers. 2.1 Contain the Virus Through Non-pharmaceutical Interventions (NPI) The different studies reviewed share the same general objective to find the best NonPharmaceutical Intervention methods to control and contain the spread of the virus. Some of these studies discuss the effect of lockdown on the spread of the virus. Other studies simulate models using other interventions such as social distancing or contact tracing. In addition, other studies use simulation to combine some of these interventions and investigate their impact on the virus spread. Shamil et al. [10] Propose a microsimulation model to evaluate the effect of NPIs for Covid 19. This model aims to identify the impact of a city-wide lockdown and the protective measures against spreading the virus infection. The model simulates the dynamic spread of the disease in any given city. Agents represent individuals in a city associated with a family with a particular profession with tasks or behaviors they can perform daily. The simulation provides the total number of cases of Covid-19. The model allows the infected agent to transfer the disease by numerous acts every hour. The tasks can be summarized as such: Tasks for students, tasks for doctors and healthcare services, tasks for service holders, and tasks for the unemployed. Agents are divided into five groups, stay home (F), Commute (T), Work or attend school (W), Attend an event (E) and Stay at the hospital (H). Furthermore, three environments are represented: Awareness and lockdown, contact tracking and quarantine, and Digital herd immunity. In addition, the model uses five possible states for each agent: not Infected, infected, contagious and asymptomatic, infected not contagious and asymptomatic, dead or recovered. The parameters used in the model are the three environments in addition to the five states applied to the agents. Finally, the interaction between agents and disease transmission cases is implemented to make the model more realistic, although this would add more intensive computation for the study. Four scenarios were simulated with the four interventions: None Intervention, contact tracing lockdown, contact tracing no lockdown, and lockdown only. The scenario of Lockdown resulted in fewer people being infected. The best scenario happened when contact tracing and lockdown were implemented, allowing herd immunity. This scenario would result especially if contact tracing is used with smartphones and the smartphone owner population is more than 75%. In addition, the model concludes that the best approach to be implemented is to track the emergency service provider during the lockdown. The model is suitable for any given city or realistic scenario with the relevant parameters for this city or area. In addition to being a threshold model with many computations involved, this study has a limitation in the sense that not all populations own or can afford smartphones. Figure 1 depicts one of the scenarios.

588

S. Ibrahim

Fig. 1. Total infections vs. intervention scenarios

Mukherjee et al. [11] Developed a model to simulate strategies for reopening after a significant lockdown in educational institutions and found the safest way to resume normal operations. The model explores ways for NPIs to implement such as wearing masks, social distancing, Contact tracing, and sanitizing. The model also studied rapid tests that provide fast results as a strategy to enable resuming normal operations. The model’s objective is to combine an agent-based simulator model with the analytical epidemiology model to derive the best policies to contain the spread of the virus. The SHEILD model incorporates key parameters such as bulk testing, contact tracing, reduced infectivity, and contact rates in various educational institutions. A nonlinear regression technique that estimates the parameter from the data was used. The infectivity rate was analyzed in relation to the bulk of testing. However, the external infectivity that depends on internal and external mobility can hold and slow the process and affect wearing masks and sanitizing. The size of the institutions or organizations would matter and impact the process. The data were collected from several US universities. The model event entails several steps: Bulk testing, then positive test results are isolated, contact tracing is implemented to trace interactions that took place before test results are published, and finally, self-isolation would also occur as a consequence of a positive test. The key feature of the model is considering the asymptomatic cases. Preventive measures such as using masks and social distancing and the rapidity of the test result are key observations for safer reopening. The model designates that weekly testing is a suitable indicator for the threshold used. Figure 2 presents an example of infection cases with bulk testing. However, the cost of bulk testing is an issue, and institutions must invest in this area to have a safer opening. As a limitation, the model uses a threshold to simulate the results. Catching et al. [12] Developed an agent-based model to simulate and assess the effectiveness of wearing masks and social distancing on the infection rate and reducing the virus spread in a closed area. The model takes into consideration the existence of the asymptomatic infected individuals. The model is a stochastic model that presents heterogeneous individuals as an agent. An extended features SEAIR model was implemented, Susceptible (S), exposed (E), Asymptomatically infected (A), symptomatically infected (I), and Recovered (R). The model includes the extended feature as a parameter in addition to the mask-wearing and infection probability. The simulator model was developed using python with the NumPy library. Different scenarios of agents wearing masks at a different period of time versus implementing social distancing were simulated. Each agent behaved dynamically according to environmental changes in population density

A Review of Agent-Based Model Simulation

589

Fig. 2. The scenario used with bulk testing in the threshold

and a certain velocity. In addition, agents moved with a randomly oriented trajectory with a fixed diameter with other agents. A simulation model was conducted to analyze the asymptomatic cases. The result showed that the number of infected asymptomatic patients is linear. The model suggests that high dependability on social distance would curve the infection spread. More importantly, masks protect from the virus even if only a small percentage follow social distancing. Finally, the model provides recommendations for policies for reopening social gatherings; a critical limitation in this model is the lack of consideration of reinfection (Figs. 3 and 4).

Fig. 3. Presents the trajectory of an individual agent

Fig. 4. Represents the infection using masks

590

S. Ibrahim

Li et al. [13] exploited an agent-based model to simulate and study the effects of NPI interventions to manage and control the virus. This model is based on SIRD (susceptible, Infected, Recovered, Dead) and uses the NetLogo as a simulator. One of the model’s objectives is to investigate these interventions as a combination of each other. Therefore, different aspects of the agent-based simulation were presented in this model. One aspect is the state in which each agent can be healthy, asymptomatic, light asymptomatic, severe asymptomatic, recovered, and dead. Another aspect is the different virus transmission methods and their effect, according to an agent’s duration and age difference and the fatality rate concerning the agent’s age. A third aspect is the healthcare system. Modeling the lockdown and mask usage are other aspects used in the simulations. The model provides distinct parameters such as social distancing, mask usage rate, lockdown delay, and symptomatic isolation rate. These parameters were analyzed against the range of age groups. Model Sensitivity is also applied in the study. The model simulation concluded that enforcing a lockdown has relatively lower effects than social distancing and mask use. The data were derived from Hong Kong, Italy, and the UK. This model was a general study, and its limitation was that it didn’t look at areas of schools or universities, transport, and venues to make the model more realistic. Nevertheless, the model was applied in general (Fig. 5).

Fig. 5. Present agent-based graph based on three parameters.

Philip et al. [7] developed a full-scale stochastic agent-based model system (ABM) to simulate a university model environment to help the administration evaluate the possibility of safe in-person instruction while assessing various necessary interventions to contain the spread of the virus. The simulation model aims to study the virus’s spread and observe the efficiency of interventions such as quarantining, contact tracing, and maskwearing policies. At the same time, larger classes are moved to online instructions. The model has several agents where behaviors are very structured and fully individualized. This approach incorporates contact heterogeneity within a population of students and instructors, resulting in a more dynamic situation for the disease. The modeled university presents a sequence of events on any typical day: Illness testing, where three predetermined percent of the tested population. And a required Quarantining for individuals who test positive. And status updates for each individual according to susceptible (S), Infected (I), and Removed (R)) followed by traceable vs. non-traceable contact tracing

A Review of Agent-Based Model Simulation

591

cases. This model also incorporates a global parameter of non-pharmaceutical interventions (NPI), such as wearing masks and social distancing. Using these parameters and other aspects of the model, such as contact tracing, online transition, and high accuracy testing, individuals are divided into clusters and models according to total infection and quarantine population. Figure 1 presents a log scale of the model with different cases: (1) intervention versus (2) no intervention or (3) intervention with high social contacts (Fig. 6).

Fig. 6. Presents another log where the different clusters given with the total infection and quarantining time

The model’s goal is to have a realistic view and an understanding of the university situation during the pandemic. Consequently, the model allows visualization of the disease dynamics and suggests several interventions. For example, Universities have to focus on their testing regimes and accuracy, quarantining, and contact tracing while encouraging large classes to switch to online courses to decrease the disruption caused by the quarantine period [7]. Among its limitations, the study did not consider seasonal differences in transmission patterns or off-campus infections. Moreover, large group sessions and seminars were not considered. Almagor et al. [14] used the ABM agent-based model to explore the effectiveness of contact tracing smartphone application (CTA) on a population on an urban scale to contain the spread of the virus. The model incorporates part of the population that uses the application, the availability of testing if required, and the agent behavior, which is the willingness to self-isolation if needed. The model explicitly simulates the interaction between heterogeneous agents in a social network. Agent represents individuals with their specific characteristics such as age and sex. Furthermore, model applications can detect other smartphone applications directly contacting possible infection transmission. The model uses the social structure to simulate daily connections between individuals within the population (household, workplace/school, friends, relatives, and random contacts). The model implemented two scenarios; The first used the initial baseline scenario to implement a social distancing and face mask policy with no CTA and no test. The second scenario is an experimental model that used a baseline scenario with tracing application and detection tests implemented as mitigating strategies to track and test infected agents. Self-isolation is required after an agent test positive. However, there is no guarantee that all individuals will comply with self-isolation. Therefore, CTA proved its efficiency in reducing the spread of the virus, which was dependent on many factors:

592

S. Ibrahim

first, the testing capacity, second, the availability of testing resources, in addition, to prioritizing the symptomatic cases testing policies, and finally, the compliance with the self-isolation. This model has certain limitations where all aspects of covid-19 cannot have all the parameters and evidence. It is very complex to simplify reality concerning self-isolation and compliance percent [14] (Fig. 7).

Fig. 7. Presents the percentage of CTA compliances of symptomatic priority, & the % of test availability

D’Orazio et al. [15] incorporate and modify the ABM model and use this model to estimate the spread of the virus in touristic areas depending on different critical scenarios, density conditions, tourists’ condition, and pandemic conditions. D’Orazio et al. [15] use a probabilistic approach using simulation software, the NetLogo R script with NLRX package. The model applies the various epidemic rules such as the proximity-based contagion spreading of the virus, the exposure time, and the mask filter protection, representing direct and indirect contact between simulator agents. The model has been adjusted to include two main typologies: Tourists and the residents created in an area called the world. The daily movement of the tourist and the residents were defined. In addition, the model incorporates the following set of parameters, such as initial infector, wearing masks, an asymptomatic ratio of infected cases, fever, and disease delay. A case study of 10 hotels was simulated and used in scenarios. The result shows that nonpharmaceutical measures such as social distancing are more effective when there is a high infection rate. However, this would affect the economy. However, results show that when the infector rate decreases, protection measures, such as wearing masks could become preferable [15] (Fig. 8). Asgary et al. [16] developed an agent-based model extended from SEIR to simulate different testing strategies that can be implemented for various classes with varying sizes in schools. The model includes symptomatic and asymptomatic infectious populations. Student and class are two agents represented in the model: The teacher is not represented in the model. In this model, a pre-symptomatic phase exists that proceeds with symptomatic infectious cases. Agents with a positive test are self-isolated and move from Not Tested to Quarantine. Class Agent determines the location of the student agent. Different simulation testing strategies were implemented, with a different number of tests per day for each class. Some challenges can be derived from the simulation: the need to perform

A Review of Agent-Based Model Simulation

593

Fig. 8. Presents the infection rate concerning % using mask parameter

a substantial number of tests, the resources needed for these tests, and the possibility of false-positive versus false-negative test results. In conclusion, the simulation model concluded that testing is conducted regularly and results are expedited. As a result, cases are self-isolated for infected students at home, and then the strategy will be effective. However, the rapid testing strategy implemented may create testing precision concerns in which, as a result, a considerable percentage of students will be in self-isolation (Figs. 9 and 10).

Fig. 9. Covid-19 testing simulation

Fig. 10. School simulation tool

2.2 Virus Spread’s Impact on the Economy A limited number of academic studies simulated the economic impact of covid 19 spread. One of these studies is reviewed below.

594

S. Ibrahim

An agent-based model (ABM) was developed. The simulations are intended to analyze the dynamics of COVID-19 epidemics and the epidemiological and economic impacts of social distance interventions [17]. The model simulates a closed society in a given environment where agents may denote people, businesses, houses, government, and the healthcare system. Agents possess defined attributes and behaviors. This society is presumed to live in an environment of a dynamic system. The interaction between the agents in the system has many variables. Their behavior is nonlinear and stochastic; The model is implemented in the Python programming language and is a free, open-source software library. Furthermore, their characteristics evolve over time with the system. The main objective of ABS is to model the temporal evolution of the system, capturing the internal states and global behaviors resulting from the interactions between the agents over many repetitions. In this approach, even nonlinear systems, complex conditions, and restrictions are simulated, some of which may be difficult to describe mathematically. This system models the economy in this society of agents, it also assesses the economic effects of seven various scenarios with different social-distancing interventions, using the COVID-ABS simulator: (1) do nothing, (2) lockdown, (3) conditional lockdown, (4) vertical isolation, (5) partial isolation, (6) use of face masks, and (7) use of face masks together with 50% of linkage to social isolation. The study supported the idea that lockdown is the best for diminishing the infected and death rate. The model contains 29 epidemiological, social, demographic, and economic input parameters and output response variables. Lockdown and conditional lockdown were the most effective scenarios for preserving lives. However, both of these scenarios present a slower evolution of the epidemic. As lockdown policies cannot be implemented, the system with 50% social isolation with masks and physical distance was the most effective. Results showed that the COVID-ABS approach effectively simulates social intervention scenarios and can be extended and customized with a specific parameter for each city (Fig. 11).

Fig. 11. The economic relationship between agent

A Review of Agent-Based Model Simulation

595

2.3 Contain the Virus Through Pharmaceutical Interventions A couple of studies that simulate optimal strategies that contain the virus using pharmaceutical interventions such as vaccines are reviewed below. Castro et al. [1] developed an agent-based model to analyze the spread of the pandemic. This model is based on different social scenarios of viral transmission. The model has multiple region spaces and open regions and have agents present dynamic individuals in a society with a very diversified environment. The agents have a wide range of features representing health conditions, purchasing power, awareness, mobility, professional activity, age, and gender. Two social scenarios were simulated with or without using a social distancing policy in viral transmissibility behavior. The simulation models the physiological and socioeconomic differences between individuals or agents within the same population. Each agent has four detrimental factors: the probability of contracting the disease, rules of movement, recovery time, and the likelihood of death. In addition, the approach permits the agent to move between different regions and with further infection exposure making the scenario closer to real-life scenarios. The simulation investigates the progress of susceptible (S), infected (I), deceased (D), and recovered (R) populations in each scenario. The model defines two types of parameters: The agent parameter similar to the agent features and the environmental parameter that include risk and crowd level. In order to endorse the model, a simulator was developed in MATLAB called M2COVIDSIM. This study demonstrates that the pandemic spread is stabilized for a more extended period of time when the social distance policy is present and the curve is more flattened. Social distancing is presented in figure four, which defines the distance between agents. The study’s conclusion indicates that the model can be implemented in different situations, and decision-makers can better understand the virus spread with the individual, social and regional parameters to better interventions against the disease. This model can be reused to implement PI, such as medical interventions and vaccination scheduling where the age parameter is available. So that is why this model was included in this category (Figs. 12 and 13).

Fig. 12. Represents evolution scenario with social distance policy of the spreading virus; the infected agent is in red

596

S. Ibrahim

Fig. 13. Presents infection of agents, the distance between agents, infected agent is presented in red

Truszkowska et al. [18] developed ABM agent-based model with human mobility, incorporating data from different databases to simulate the best optimal complex vaccine strategies with the possibility of reopening. Through this model, they examine the interplay between the transmission risk resulting from ongoing reopening efforts and the enhanced immunity brought by the vaccine rollout. In this model, the presymptomatic and symptomatic stages, as well as the outcomes of recovery and death, were included. When an agent exhibits covid 19 symptoms, it can be tested, quarantined, hospitalized, and treated in an intensive care unit. Vaccination campaigns can also be simulated, and a wide variety of non-progressive interventions, including school closures, lockdowns, and social distancing, can be implemented. Many Simulations are used to evaluate the vaccine rollout and the reopening of the economy. The finding suggests that opening can be accelerated without the risk of another breakdown. The results pinpoint the value of accurate testing. In comparison, limitations revolve around the absence of the initial health state of the town—the information of the new variant is all the time unforeseen (Fig. 14).

Fig. 14. Reopening % with different stages of testing, Low, Moderate, perfect testing efficacy

Sulis and Terna [19] conduct an agent-based model simulation to find the optimal vaccine strategy. This model adopts an artificial intelligence technique to adopt the best effective policies. In addition, this model uses the multiagent modeling environment NetLogo with some genetic algorithms to study the evolution of vaccination strategies. The diffusion of news and rumors via social media is affecting vaccination strategies. ABM is investigating the complex behavior of individuals that happens due to their social

A Review of Agent-Based Model Simulation

597

and physical environment. The model is designed with the idea of what-if scenarios. S.I.s.a.R that represent susceptive (S), Infected (I), symptomatic (s), asymptomatic (a) and Recovered (R). Agents are computational entities that have various interactions: in houses, in hospitals, in nurseries, in schools, and in the same places and outer spaces. The model’s primary objective is to find the best strategy for selecting the vaccine group is to find the optimal parameter for this simulation and find the best vaccine campaign. The model simulates a baseline without any implementation of any vaccine. And then run the model of different scenarios with different vaccine strategies. The advantage of this model is more relevant to the realistic case as the virus is still spreading. However, the model can be improved by adding different probabilities of infection for immunized individuals and by adding and finding the best replicate tests by changing the people and keeping the same parameters. Therefore, this would provide an excellent option for further research in this area.

3 Conclusion The newest and fast pace events and aspects of the Covid-19 pandemic presented the world with a complex challenge to understand and contain the spread of the virus. Modeling the infection and spread of the virus is needed to handle many aspects of the pandemic and its inter-relationships and dependencies. The review of the different studies confirms that these complexities translated into various limitations across the different models. For example, models using thresholds included lots of computation. Other simulations depend on situational practicalities, such as smartphone access and costly bulk testing; hence, their outcome is not replicable. Similarly, Rapid testing models are affected by test accuracy issues. Emerging changes like reinfection, changing weather, or exceptional cases couldn’t be simulated, given the inconsistency of their occurrence. Despite all these limitations, models did provide guidance to help policymakers develop vaccine strategies, reopening strategies, and progress in achieving immunity.

Appendix A

Covid 19 Spread Analyze the with Preventive effectiveness testing of different testing strategies in school

Covid 19 Multi-region space

Covid-19 Spread Examine the efficacy of social distancing and face Mask usage in containing the virus spread

[16]

[1]

[12]

To understand the virus spread Actions were taken against the pandemic

Covid 19-Spread To explore the urban scale effect of contact tracing application on containing the virus

[14]

Aim

Domain

Paper

Close space

Open space, multi-region space, Civil society

School

Social structure: Household’s schools organization

Case study

SEAIR

M2COVIDSIM

Simulator

Available at GitHub

Available/open source

Student = (and class,

Individual Heterogeneous

Agents

Population of asymptomatic

Individual heterogeneous

Health condition, Individual Purchasing power, Heterogeneous Awareness, Mobility patterns, Social interaction

Symptomatic and asymptomatic Social distance Hygiene practice

Behavior factor

Feature

(continued)

Python NumPy library

MATLAB

Other modeling approaches

598 S. Ibrahim

Domain

Covid 19 spread

Covid 19 Spread

Covid -19 Spread of reopened school

Covid 19 in small society

Covid-19 spread

Paper

[15]

[13]

[11]

[7]

[17]

(continued) Case study

To control the virus spread using NPI scenarios to look economy

Containing campus outbreak, i.e., size outbreak And quarantining

Explore strategies to reopen and restart the operation

The effect of NPI on reducing the spread

NetLogo

Simulator

Society

University & Colleges

Institutions

Python

SHIELD program

Open Society NetLogo

Effect of risk Tourist city mitigation measure: social distancing and masks policy

Aim

Available open-source

Available

Available

Available/open source

closed society

Large University.Highly structured Contact Heterogeneity between individual

Age, gender, infection status

tourist and resident

Feature

People, government business

Full-scale stochastic agent-based model Individual

Autonomous agent heterogeneous

Two mains Heterogeneous tourist and resident

Agents

(continued)

Python programming language

Removing large classes to online: issues

Rapid bulk testing

Probabilistic approach R version

Other modeling approaches

A Review of Agent-Based Model Simulation 599

Domain

Optimal for a vaccination campaign

Covid-19 spread

Optimal for a vaccine campaign

Paper

[19]

[10]

[18]

(continued)

To control the virus spread using NPI

To find the best vaccine strategy

Aim

City

Case study NetLogo

Simulator

Available

Available

Available/open source

Feature

Agents

Python programming language

Other modeling approaches

600 S. Ibrahim

A Review of Agent-Based Model Simulation

601

References 1. Castro, B.M., de Abreu de Melo, Y., Fernanda dos Santos, N., Luiz da Costa Barcellos, A., Choren, R., Salles, R.M.: Multi-agent simulation model for the evaluation of COVID-19 transmission. Comput. Biol. Med. 136 (2021) 2. Villavicencio, C., Macrohon, J.J., Inbaraj, X.A., Jeng, J., Hsieh, J.: Twitter sentiment analysis towards COVID-19 vaccines in the Philippines using Naïve Bayes. Informatiom 12 (2021) 3. Lorig, F., Johansson, E., Davidsson, P.: Agent-based social simulation of the covid-19 pandemic: a systematic review. Jasss 24(3) (2021) 4. Cardoso, R.C., Ferrando, A.: A review of agent-based programming for multi-agent systems. Computers 10(2), 1–15 (2021) 5. Sebastien, R., Olivier, M., Andrei, D.: Use of fuzzy sets, aggregation operators and multi agent systems to simulate COVID-19 transmission in a context of absence of barrier gestures and social distancing: Application to an island region. Proceedings - 2020 IEEE International Conference Bioinformation Biomedical BIBM 2020, pp. 2298–2305 (2020) 6. Wei, Y., Wang, J., Song, W., Xiu, C., Ma, L., Pei, T.: Spread of COVID-19 in China: analysis from a city-based epidemic and mobility model. Cities 110 (2021) 7. Philip, J.R.P., Gressman, T.: Simulating Covid -19 in a university environment. Math. Biosci. 388, 539–547 (2020) 8. Albahri, O.S., et al.: Systematic review of artificial intelligence techniques in the detection and classification of COVID-19 medical images in terms of evaluation and benchmarking: Taxonomy analysis, challenges, future solutions, and methodological aspects. J. Infect. Public Health 13(10), 1381–1396 (2020) 9. Alamoodi, A.H. et al.: Sentiment analysis and its applications in fighting COVID-19 and infectious diseases: a systematic review. Expert Syst. Appl. 167 (2021) 10. Shamil Salman, I., Farhanaz, F.: An agent-based modeling of COVID-19 validation, analysis, and recommendations. Cognit. Comput. (2021) 11. Mukherjee, U.K. et al.: Evaluation of reopening strategies for educational institutions during COVID-19 through agent based simulation. Scientific Reports 11(1) (2021) 12. Catching, A., Capponi, S., Te Yeh, M., Bianco, S., Andino, R.: Examining the interplay between face mask usage, asymptomatic transmission, and social distancing on the spread of COVID-19. Scientific Reports 11(1) (2021) 13. Li, K.K.F., Jarvis, S.A., Minhas, F.: Elementary effects analysis of factors controlling COVID19 infections in computational simulation reveals the importance of social distancing and mask usage. Comput. Biol. Med. 134 (2021) 14. Almagor, J., Picascia, S.: Exploring the effectiveness of a COVID-19 contact tracing app using an agent-based model. Scientific Reports 10(1) (2020) 15. D’Orazio, M., Bernardini, G., Quagliarini, E.: Sustainable and resilient strategies for touristic cities against COVID-19: an agent-based approach. Safety Sci. 142 (2021) 16. Asgary, A., Cojocaru, M.G., Najafabadi, M.M., Wu, J.: Simulating preventative testing of SARS-CoV-2 in schools: policy implications. BMC Public Health 21(1) (2021) 17. Silva, P.C.L., Batista, P.V.C., Lima, H.S., Alves, M.A., Guimarães, F.G., Silva, R.C.P.: COVID-ABS: an agent-based model of COVID-19 epidemic to simulate health and economic effects of social distancing interventions. Chaos Solitons Fractals 139 (2020)

602

S. Ibrahim

18. Truszkowska, A. et al.: Designing the safe reopening of US towns through high-resolution agent-based modeling. Adv. Theory Simul. 4(9) (2021) 19. Sulis, E., Terna, P.: An agent-based decision support for a vaccination campaign. J. Med. Syst. 45(11) (2021)

Dengue in Bangladesh: Strategic Assessment Considering the Future Outbreak and Hospital Scenario Md. Zahidur Rahman(B) and Nur Mohammed Department of CSE, Britannia University, Cumilla 3500, Bangladesh [email protected]

Abstract. Dengue is one of the most important global health problems caused by mosquito bites. In recent years, the incidence of dengue has increased dramatically all over the world, especially in Bangladesh. There have been several outbreaks of dengue in Bangladesh. Recently, the incidence of dengue is much higher in Bangladesh. The government is trying its best to eradicate this epidemic, but dengue is on the rise despite efforts taken by the government. One of the main reasons for this failure is the ignorance about the upcoming dengue outbreak and appropriate actions. To detect and deal with future dengue outbreaks on time, in this paper, we will forecast the upcoming dengue outbreak using SARIMA (1,0,0)(1,1,1)12 model. Furthermore, we will discuss the current hospital scenario for dengue patients and predict the future risks. Finally, we will discuss some strategies to help the government to control the situation. Keywords: Dengue · Forecast · Control

1 Introduction Dengue is one of the most fatal public health problems in Bangladesh and many other countries around the world. It is mainly transmitted by the mosquito Aedes aegypti, a well-known principal vector of dengue fever in South-East Asia, including Bangladesh. Dengue is caused by a virus of the Flaviviridae family and there are four serotypes of the virus which are (DENV-1, DENV-2, DENV-3, and DENV-4) [1, 2]. All four serotypes cause a full spectrum of diseases from subclinical infection to mild self-limiting diseases dengue fever (DF), dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS), etc. Diseases cause a spectrum of illnesses ranging from mild symptoms such as fever, body ache, joint pain, headache, and body rashes to severe hemorrhagic complications. There is no specific treatment for dengue or severe dengue, but early detection and access to proper medical care lowers fatality rates to below 1% [3]. Dengue virus (DENV) is transmitted when a person is bitten by an infected female Aedes mosquito, predominantly Aedes aegypti and Aedes albopictus. Aedes aegypti is the principal urban vector and is highly competent as an epidemic vector of DENV. Aedes aegypti females lay eggs in artificial receptacles and usually rest inside houses. Aedes albopictus females rest outdoors where they also lay eggs in artificial and natural containers [4]. Clinical manifestations of © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 603–612, 2023. https://doi.org/10.1007/978-3-031-20429-6_54

604

Md. Zahidur Rahman and Nur Mohammed

dengue range from asymptomatic infection to life-threatening shock and hemorrhage. Dengue fever manifests typically following a 4–7 day (maximum 10 days) intrinsic incubation period in a human host. During the viral period of the following 4–5 days (maximum 12 days) the virus can be transmitted to mosquitoes through an infectious blood meal. After 8–12 days of extrinsic incubation, mosquitoes become infectious and can transmit the virus to humans; the mosquito remains infectious for life [5]. The objective of this study is to forecast monthly dengue incidences for the year 2022. Forecasting is needed to determine when disease outbreaks or events will occur or arise. Forecasting will also help to provide effective and efficient management and administration planning for the country. Another objective is to describe the hospital scenario, which will help the government to increase hospital facilities for dengue patients. The remaining paper is organized as follows. Section 2 provides summary of dengue epidemiology worldwide and in Bangladesh and also provides factors influencing dengue transmission. In Sect. 3, we forecast the dengue cases for the year 2022 using the SARIMA model. Current and future hospital scenarios discussed in Sect. 4. Section 5 provides the discussion regarding the strategies and guidelines to reduce the impact of dengue. A conclusion is presented in Sect. 6.

2 Dengue Epidemiology 2.1 Global Situtation of Dengue Dengue, a mosquito-borne viral illness, is an important cause of morbidity and some mortality in many countries, mostly in Asia and Latin America, and is continuing to expand globally. From only nine countries before 1970, the disease is now endemic in over 100 countries. Around 390 million infections (95%) occur each year with approximately 500,000 hospital admissions with potentially life-threatening forms of the disease. Approximately 12,000 deaths, mostly among children, occur worldwide every year. An estimated 50% of the global population is at risk of acquiring dengue and over half reside in the World Health Organization’s South-East Asia Region. Outbreaks of dengue fever were recorded in Southeast Asia in the late 1940s. Nevertheless, dengue has reemerged worldwide since the 1980s with severe epidemics and geographical spread and has become a rapid epidemic threat in the Asia Pacific and South America. According to the World Health Organization (WHO), the DF is currently endangering local and an estimated 3 billion people in more than 128 countries. The annual number of infections worldwide is estimated at 50 million, with 500,000 severe cases. About two-thirds of the global understanding of dengue is on the shoulders of Asia Pacific and about 72% of the world’s 3 billion population is at risk [6]. A recent study indicated 390 million dengue infections per year, of which 96 million were clinically revealed. Another study found that 3.9 billion people in 128 countries were at risk of contracting the dengue virus [7]. The WHO reported the number of dengue cases was 505,430 in 2000, 2.4 million in 2010, and 5.2 million in 2019. In 2021, 1.4 million cases have been reported all over the world.

Dengue in Bangladesh: Strategic Assessment Considering ...

605

2.2 Dengue in Bangladesh In Bangladesh, Dengue was first reported as “Dacca fever” in 1964 by Aziz and his colleagues [8]. The subsequent reports recommended that dengue could occur periodically in Bangladesh from 1964 to 1999. The primary epidemic of dengue fever was rumored within the capital town, Dhaka within the year 2000. Between January 2000 to December 2007, Bangladesh recorded a total of 21847 cases and 233 deaths (1.066%). Between January 2008 to December 2016, Bangladesh recorded a total of 15412 cases and 29 deaths (0.188%). Recently, Bangladesh has experienced several dengue outbreaks: in 2017, the country reported 2,769 cases and 8 deaths and in 2018, the annual incidence increased to 10,148 cases and 26 deaths. In 2019, the Directorate General of Health Services (DGHS) reported an incidence rate 10 times higher than the previous year, with 101,354 cases and at least 179 deaths (0.162%). This outbreak had the best-case burden among all infectious disease outbreaks in Bangladesh. DENV-3 was delineated because of the dominant serotype within the 2019 outbreak [9]. Meanwhile, COVID-19 has been a global concern since January 2020, and the number of COVID-19 cases has been on the rise in Bangladesh since May 2020. In 2020, during the covid situation, 1193 cases and 3 deaths were reported. Between January 2021 to November 2021, Bangladesh has reported over 27,368 dengue cases and 96 deaths. According to the DGHS, Dhaka accounts for 88% of the total cases since January [10]. At this moment, Bangladesh is on the verge of facing another dengue outbreak. 2.3 Factors Influencing Transmission Climate Determinants. Climate influences the ecology of dengue. Low temperature slows adult mosquito development and prevents Aedes mosquitoes from transmitting DENV [11]. Consistent with these observations significantly reduced dengue transmission in cold climates in seasonal areas [12]. Conversely, dengue infection increases with increasing temperature, which accelerates mosquito development; increases bite frequency and shorten external incubation time [13]. Survival of adult and aquatic mosquitoes is reduced to temperatures above 30 °C and 35 °C, respectively [14]. The ambient temperature also affects the ovulation rate of female Ae. Aegypti but the intensity is affected by relative humidity. The highest rate is reported at 25 °C with 80.0% relative humidity [15]. Variations in daily temperature also affect dengue survival and vectorial ability of aegypti and hence the infection of DENV. In warmer environments, small fluctuations in temperature increase the vectorial capacity and in cooler environments, fluctuations in temperature reduce the vectorial capacity. DENV infections are usually highest during the rainy season [16–19]. Non-climate Determinants. Socio-economic and behavioral factors affect DENV infection. For example, with rising temperatures, people tend to live in air-conditioned rooms where they are less exposed to mosquito bites, resulting in a reduced incidence [20]. Conversely, in many low-income tropical areas housing is not air-conditioned, and the risk may be further increased by clothing appropriate to high ambient temperature. High population density increases the incidence of dengue. Small distances between large families and homes increase the likelihood of multiple infections in a single-family and the development of urban pockets of dengue infection [21]. Unplanned urbanization

606

Md. Zahidur Rahman and Nur Mohammed

is associated with inefficient urban management resulting in a lack of water supply, drainage, and waste disposal. Inadequate access to piped water encourages water storage that supplies breeding sites. In the absence of proper waste management, artificial pots are provided where Aedes mosquitoes lay their eggs [22].

3 Forecasting Dengue Statistical models like Seasonal Autoregressive Integrated Moving Average (SARIMA) can be used to analyze the data and make the prediction on data. The seasonal SARIMA (1,0,0)(1,1,1)12 model has been used here to observe and forecast data. This model has been found as the most suitable model with least Normalized Bayesian Information Criteria (BIC) of 18.549 and Root Mean Square Error (RMSE) of 9073.657. The model was further validated by Ljung-Box test (Q18 = 15 and p > 0 .05) with no significant autocorrelation between residuals at different lag times. We discussed the details of the SARIMA model and time series analysis in our other paper [23]. Using data on dengue infection from 2017 to 2021, we fit a SARIMA model to dengue incidence, and then we use the best model SARIMA (1,0,0)(1,1,1)12 to predict the dengue cases for the year 2022. First, we observed the series of dengue incidents from January 2017 to December 2021. The observed Fig. 1 shows that the series is nonstationary and there are seasonal fluctuations in the dataset.

Fig. 1. Dengue incidence: observed SARIMA(1,0,0)(1,1,1)12 [23]

values,

fit

value

and

predicted

values

of

Finally, we used the identified SARIMA (1,0,0)(1,1,1)12 model to calculate predicted values. Using the model, we find the predicted values from January 2022 to December 2022, as shown in Fig. 2. The graph shows that, the monthly predicted incidents where the number of dengue cases from January to May is low and starts to increase from June. The peak comes in July and continues to rise until November, and starts declining afterward. This study found that the trend of dengue fever outbreaks occurred between

Dengue in Bangladesh: Strategic Assessment Considering ...

607

August and November. The forecasted number of infected patients indicates a seasonal pick in August 2022, with the estimated number of patients as 6410 [23].

Fig. 2. Forecast of dengue incidence from January 2022 to December 2022 by SARIMA(1,0,0)(1,1,1)12 [23]

4 Hospital Scenario 4.1 Current Situation Dhaka is currently the hotspot of dengue outbreaks, reporting more than 90% of the total cases every day. In this situation, the reduced medical facilities resulted in another treatment crisis with Covid-19. Treatment of dengue patients in hospitals where covid patients have also been treated increases the risk of coronavirus infection. If dengue patients are kept in the same hospital, there is a risk of Covid-19 infection. Some hospitals such as Bangabandhu Sheikh Mujib Medical University (BSMMU), Shaheed Suhrawardy Medical College Hospital, and Mugda General Hospital have also turned their dengue units into covid wards. These hospitals are currently only providing first aid to dengue patients and referring them to other government facilities. On 23 August 2021, the Directorate General of the Health Services (DGHS) announced six hospitals in Dhaka were dedicated to treating dengue patients. The six dedicated hospitals are: • • • • •

Sir Salimullah Medical College Hospital Shaheed Ahsan Ullah Master General Hospital in Tong Railway General Hospital at Kamalapur 20-bed Aminbazar Govt. Hospital Lalkuthi Hospital in Mirpur

608

Md. Zahidur Rahman and Nur Mohammed

• 31-bed Kamrangirchar Hospital Seat capacities of these six hospitals are shown in Fig. 3.

Fig. 3. Dengue dedicated hospitals & their seat number

4.2 Future Scenario (Prediction) Figure 4 shows a comparison of predicted dengue patients, the availability of seats, and the scarcity of seats in dengue dedicated hospitals from January 2022 to December 2022. The graph shows that from January to May we have no shortage of seats. But from June to December, we have a huge deficit of 243, 1942, 5452, 4181, 4883, 3754, and 683 seats respectively. This huge deficit will be a major factor for the next dengue outbreak in Bangladesh in 2022.

Dengue in Bangladesh: Strategic Assessment Considering ...

609

Fig. 4. Comparison Chart of affected number, total seats and number of shortage seats based on forecasted data

5 Instructions to the Government Realizing the depth of the Dengue epidemic, we think it is high time for the government to take certain initiatives to decrease the effect of Dengue. The following aspects can be considered. 5.1 Enrichment of Data Data plays a very important role in research. The dataset we worked on collected from the Directorate General of the Health Services (DGHS) is not well enough. We need two types of data for our research: (1) Dengue Data. All the data in our country is not properly formatted. Moreover, there is no specification of data in certain cases like which area, which age, etc. Several categories are also missing including the age and sex of dengue patients. (2) Hospital Data. The DGHS announced six hospitals in Dhaka dedicated to treating dengue patients. The exact number of patients admitted to hospitals other than these dengue-dedicated hospitals is not accurately recorded. As a result, the future prediction could not be performed properly. Very little data is available on the website of the DGHS; we have to collect data from several online news portals by using the web scraping technique. More data needs to be added to the DGHS website to solve this problem of lack of data. 5.2 Regularity of Data It is not possible to come to the right conclusion from any survey with irregular data. There is a lot of missing data in the central database. For example, accurate data from the Covid-19 period was not available on DGHS’s website. During certain periods, such

610

Md. Zahidur Rahman and Nur Mohammed

as March 2020 to December 2020, there is no data available at DGHS; we collected data manually to perform our dengue case prediction for the year 2022. The government needs to be more vigilant in maintaining the regularity of data. This is a major problem that should be resolved because mass people should know about the current scenarios about dengue, medication, and hospital arrangement. 5.3 Taking Initiatives to Face the Upcoming Outbreak This research shows that the number of dengue patients in Bangladesh will gradually increase from June 2022, and it will continue to increase till December 2022. The maximum number of victims will be 6410 in August 2022. The government needs to look into this and take the necessary initiatives to eradicate this outbreak properly. To overcome the impending outbreak, the government should increase hospital facilities and maintain an accurate database of daily dengue cases. Hopefully, this research will help the authority to create an early warning system. 5.4 Establishment of New Unit Under DGHS Since no antidote to dengue has been discovered yet, we need to take appropriate steps to prevent dengue. The government should open a new unit under DGHS at the district level which especially works for Dengue, monitors the dengue situation every year, collects preliminary data and contributes to the central database. This will help in data enrichment and data normalization and using this data we can predict impending outbreaks and control outbreaks on time. 5.5 Development of a User-Friendly Application If a central database can be managed by DGHS, the government can develop a userfriendly application, maybe an android app, where general people can track the current situation of dengue, gets general instructions, gets information about dengue-dedicated hospitals, connect with telemedicine service, etc.

6 Conclusion Dengue is a mosquito-borne viral disease that is considered an epidemic in our country. Many people have died due to this dengue in Bangladesh. Since no vaccine has been discovered, it is not possible to eradicate it. However, we can deal with dengue outbreaks if we know the outbreak in advance and are aware of the steps that need to be taken. In our study we found that in August 2022, there will be a major dengue outbreak in Bangladesh. On this basis, we tried to give some instructions to help the government to deal with dengue. Hopefully, if the government follows this guidance, Insha’Allah we will be able to properly deal with the next dengue outbreak.

Dengue in Bangladesh: Strategic Assessment Considering ...

611

References 1. WHO, Dengue and severe dengue, https://www.who.int/news-room/fact-sheets/detail/den gue-and-severe-dengue. Accessed 10 May 2021 2. Gupta, N. et al.: Dengue in India. Indian J. Med. Res. 136(3), 373–90 (2012) 3. Health Organisation, Fact sheet. Updated April 2017, https://reliefweb.int/report/world/den gue-and-severe-dengue-fact-sheet-updated-april-2017. Accessed 10 May 2021 4. Bonizzoni, M. et al.: The invasive mosquito species Aedes albopictus: current knowledge and future perspectives. Trends Parasitology 29(9) (2013) 5. Centers for Disease Control and Prevention, Dengue 2010, http://www.cdc.gov/dengue/epi demiology/index.html. Accessed 10 May 2021 6. Hii, Y.L., Ng, N., Ng, L.C., Zhu, H., Rocklöv, J.: Dengue risk index as an early warning. In: DiVA (2013) 7. Rojas, A., Moreira Soares, A., Mendoza, L.P. et al.: Revisiting the dengue epidemic of 2011 in Paraguay: molecular epidemiology of dengue virus in the Asuncion metropolitan area. In: BMC Infect Dis 21, 769 (2021) 8. Aziz, M.A., Graham, R.R., Gregg, M.B.: Dacca fever -an outbreak of dengue. Pak. J. Med. Res. 6, 83–92 (1967) 9. Dengue outbreak in 2019 crosses all previous records, https://bdnews24.com/health/2019/10/ 10/dengue-outbreak-in-2019-crosses-all-previous-records. Accessed 10 May 2021 10. Dengue outbreak shatters all records, Bangladesh, https://www.dhakatribune.com/bangla desh/2019/10/13/dengue-outbreak-shatters-all-records. Accessed 10 May 2021 11. Blanc, G., Caminopetros, J.: Experimental researches on dengue. Ann Inst Pasteur 44, 367– 436 (1930) 12. Kuno, G.: Review of the factors modulating dengue transmission. Epidemiol. Rev. 17(2), 321–35 (1995) 13. Focks, D.A., Brenner, R.J., Hayes, J., Daniels, E.: Transmission thresholds for dengue in terms of Aedes aegypti pupae per person with discussion of their utility in source reduction efforts. Am. J. Trop. Med. Hyg. 62(1), 11–8 (2000) 14. Yang, H., Macoris, M., Galvani, K., Andrighetti, M., Wanderley, D.: Assessing the effects of temperature on the population of Aedes aegypti, the vector of dengue. Epidemiol. Infect. 137(8), 1188–202 (2009) 15. Costa, E.A., Santos, E.M., Correia, J.C., Albuquerque, C.M.: Impact of small variations in temperature and humidity on the reproductive activity and survival of Aedes aegypti (Diptera,Culicidae). Revista Brasileira de Entomologia 54(3), 488–493 (2010) 16. Carrington, L.B., Armijos, M.V., Lambrechts, L., Scott, T.W.: Fluctuations at a low mean temperature accelerate dengue virus transmission by Aedes aegypti. PLOS Neglected Tropical Diseases (2013) 17. Carrington, L.B., Seifert, S.N., Willits, N.H., Lambrechts, L., Scott, T.W.: Large diurnal temperature fluctuations negatively influence Aedes aegypti (Diptera: Culicidae) life- istory traits. J. Med. Entomol. 50(1), 43–51 (2013) 18. Lambrechts, L., Paaijmans, K.P., Fansiri, T., Carrington, L.B., Kramer, L.D., Thomas, M.B., Scott, T.W.: Impact of daily temperature fluctuations on dengue virus transmission by Aedes aegypti. P Natl. Acad. Sci. USA 108(18), 7460–5 (2011) 19. Liu-Helmersson, J., Stenlund, H., Wilder-Smith, A., Rocklöv, J.: Vectorial capacity of Aedes aegypti: effects of temperature and implications for global dengue epidemic potential. Plos One (2014) 20. Gage, K.L., Burkot, T.R., Eisen, R.J., Hayes, E.B.: Climate and vectorborne diseases. Am. J. Prev. Med. 35(5), 436–50 (2008)

612

Md. Zahidur Rahman and Nur Mohammed

21. Scott, B.: Halstead: Dengue virus-mosquito interactions. Annu. Rev. Entomol. 53, 273–91 (2008) 22. WHO and TDR: dengue guidelines, for diagnosis, treatment, prevention and control. Technical document (2009) 23. Mohammed, N., Zahidur Rahman, Md.: Forecasting dengue incidence in Bangladesh using seasonal ARIMA model, A time series analysis. In: International Conference on Machine Intelligence and Emerging Technologies 2022 (MIET 2022). Springer, in Press

Covid-19 Vaccine Public Opinion Analysis on Twitter Using Naive Bayes Samar Ibrahim(B) and Sheriff Abdallah The British University in Dubai, Dubai, UAE [email protected], [email protected]

Abstract. Twitter is a viable data source for studying public opinion. The study aims to identify public opinion and sentiments toward Covid-19 vaccine and examine conversations posted on Twitter. The study examined two Datasets; one of 7500 tweets collected using RapidMiner from June 7–17, 2021, and 9865 tweets collected from Kaggle on the 3rd of January 2021. It used Naive Bayes model to classify, analyze and visualize tweets according to polarity, K-means clustering, and key tweet topics. The study showed that positive sentiments were dominant in both times; it also realized that positive polarity increased over time from January to June 2021. In addition, vaccine acceptance became more prevalent in the tweets’ discussions and topics. Understanding sentiments and opinions toward Covid-19 vaccine using Twitter is critical to supporting public health organizations to execute promotions plans and encourage positive messages towards Covid-19 to improve vaccination mitigation and vaccine intake. Keywords: Covid-19 Vaccine · Twitter · Public Opinion · Datasets · Naïve Bayes

1 Introduction Corona Virus Disease (COVID-19) is a frightening pandemic discovered in early December 2019 in Wuhan, China, and spread rapidly across the globe. The World Health Organization (WHO) described this outbreak as a severe global threat [1]. The health crisis caused by this disastrous pandemic globally severely affected various sectors. In 2020, the global economy diminished by 3%, with a significant loss estimated at around 9 trillion USD [2]. Amid the unprecedented crisis, significant efforts have been embarked on in order to mitigate disease spread and find a cure or protection. Since December 2020, many COVID-19 vaccine candidates have been verified to be safe and effective in protecting against COVID-19 and were approved by regulators to be used [3]. Hence, many governments across the globe devised vaccination rollout plans according to priorities population and conducted a vaccination campaign where their residents were encouraged to be vaccinated and acquire immunity to tackle the pandemic [4]. In addition, Governments and health organizations have also used a variety of communications channels, traditional and nontraditional, such as social media, to share and provide learning knowledge related to the COVID-19 virus and its various vaccines [3]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 613–626, 2023. https://doi.org/10.1007/978-3-031-20429-6_55

614

S. Ibrahim and S. Abdallah

Since the beginning of the COVID-19 pandemic, information circulating via social media networks such as Twitter, Facebook, YouTube, etc., has significantly affected public opinions. Furthermore, people’s use and dependency on social media as a source of news and information have intensified because of the many implemented lockdown [1]. These social media platforms also provided access to unprecedented information and knowledge through which users could express sentiments, views, and opinions. As a result, these platforms have become a viable source of data for researchers to analyze posts and predict trends for human sentiments and use these for better mitigating crises, especially pandemics [1]. For example, Twitter, one of the world’s leading microblogging social media platforms, has over 330 million active users, over 500 million tweets per day, and more than a billion visits per month, and people can genuinely express themselves in a timely mode [5]. The Emergence of mobile and ubiquitous computing and the advance in big data analytics and storage platform makes it more feasible for more social media data to be generated, collected, and analyzed [6]. There is a great advantage in using these platforms, especially Twitter, to analyze public sentiments and opinions instead of traditional face-to-face interviews. As a result, users can express themselves genuinely and freely share their locations, opinions, and feelings. Whereas in traditional ways, people’s response is typically affected by the nervousness brought on by live communication situations [7]. Many researchers use many methods and techniques to analyze these tweets. They prove that an automated data mining system with real-time analysis of these posts is highly efficient for identifying people’s opinions during times of disaster [8]. They mainly use sentiment analysis and topic clustering methods. By defining the topic used in these tweets using LDA (Latent Dirichlet allocation), they comprehend more the conversation subjects and sentiments about COVID-19 [9]. Moreover, many researchers have examined the effect of social media data on vaccine hesitancy. It is the role in spreading misinformation about Covid-19, such as the virus was the result of a 5G social network or the idea that this virus was created as a government bioweapon [1], to name a few examples. This discouraged immunization perception or decreased the vaccine’s benefits perception, negatively affecting the vaccine uptake [10]. Therefore, detecting trends using social media data would provide public health authorities with essential opportunities to create policy decisions, provide new plans, set up arrangements or interventions, and make people aware of all public changes [5]. The objective of the research study is to compare public emotion toward Covid-19 vaccine using the microblog Twitter data and test its sentimental analysis across two different timelines. 1. The first research question compares the different polarities that change over time, using Twitter as a social media towards Covid-19 vaccine. 2. The second question compares different polarities using Twitter as social media data towards Covid-19 vaccine. 3. The third question is to examine the main topical issues in the tweet’s conversations about the Covid-19 vaccine?

Covid-19 Vaccine Public Opinion Analysis

615

The rest of the paper is organized as follows: Sect. 2 looks at all previous researchrelated work analyzing public opinion and sentimental analysis towards Covid-19 and its vaccines using Twitter as a data source. Section 3 provides an overview of the methodology, methods, and algorithms used in the study. Section 4 presents the results and a detailed discussion of the results. Finally, Sect. 5 provides concluding remarks and suggested future areas for research.

2 Related Work Many researchers investigate public opinion towards Covid-19 using social media data. They use various methods and techniques to analyze social media data and study public attitudes toward Covid-19. One of the studies looked at public perception towards some non-pharmaceutical interventions (NPIs) used during the pandemic to mitigate it in different countries using the topic clustering technique applied to extracted tweets. They identified that some countries have more concerns and attention than other countries. People in New Zealand, for example, care more about using hand sanitizer and wearing masks than in the US [11]. Another study investigated different topics in the tweets related to Covid-19 and showed that analyzing the topic would allow a sense of trends of people’s perceptions on Twitter. Tweets were investigated in Portuguese and English and compared to discuss the effectiveness of topic identification and sentiment analysis in these languages. Topics were ranked, and tweets’ content was analyzed while providing an analytical assessment of the discourse evolution over time [12]. (2021) proved a high correlation between public attention level, analyzed on Twitter, and Covid-19 related to case number; this was done using topic analysis and sentiment analysis methods. At the early stage of the pandemic, researchers explored COVID-19-related social media posts describing symptoms in order to identify Covid-19 different symptoms. As a result, a total of 36 symptoms were extracted and identified [13]. In addition, other researchers looked for frequent users. They analyzed their usage patterns and overall emotion during this period to understand the changing mood of the people in relation to the disease [14]. Other researchers designed and developed a novel learning machine based on classification, clustering, and topic extraction to analyze tweets and obtain significant sentiments. This model extracts main topics from the clusters using the K-means algorithm divides these topics into positive, neutral, and negative sentiments, and rapidly identifies commonly dominant characteristics of public opinions and attitudes related to COVID19 [15]. Other studies examined tweets to explore sentiment analysis using a support vector machine (SVM), k-nearest neighbor (KNN), and Naïve Bayes [15]. Likewise, Samuel et al. (2020) in [15] analyze the tweets, provide non-textual variables using N-Gram, and analyze the sentiments using NB, Linear regression, LR, and KNN. Fake news during the pandemic was another focus for researchers to explore ways to detect and reasons for misleading the targeted population using a classification approach that implements natural language processing, machine learning, and deep learning [16]. Very few and limited research investigates the public opinion and perception of the vaccine on Twitter. One study assesses the opinion of the Indonesian people using social network analysis of the COVID-19 vaccine by implementing sentiment analysis

616

S. Ibrahim and S. Abdallah

using the Naïve Bayes Algorithm. Likewise, Yousefinaghani et al. (2021) study public sentiments and opinions toward COVID-19 vaccines based on the content of Twitter compared to their progression over time between January 2020 and January 2021, their geographical distribution, and provide discussion on vaccine rejection and hesitancy and show different patterns in various countries. This study also investigated public opinion and sentimental analysis but compared recent Twitter data collected in June 2021 to data collected in January 2021. In addition, it examined the topics of discussion included in these tweets.

3 Methodology and Methods The study was conducted following a 5-stage successive process, as depicted in Fig. 1. First, data was collected, annotated, and pre-processed using Natural Language Processing techniques (NLP), then classified using sentimental analysis utilizing the Naïve Bayes classifiers algorithm in order to train a sentiment classification model. A Naive Bayes classifier is commonly used for solving classification problems and has verified to be accurate at determining the true polarity of sentences as compared to other techniques [2]. Next, data used performance classification to test the performance of the model. Furthermore, K-means clustering was implemented in the study to analyze the topics and apply them in visualization. Finally, visualization was used to present the result of the study.

Data Collection Data Annotation Data Preprocessing Data Processing

Performance Evaluation Visualization

Fig. 1. Represents the different five stages of the study

3.1 Data Collection Two datasets were used in the study, one set of 9,862 data tweets collected from the Kaggle website dated from the third week of January of 2021. The second was collected directly from Twitter using RapidMiner searching twitter using different keywords, such as covid19-vaccine or #covid vaccine; English attribute was also used. Specific attributes were collected, including but not limited to creation date, username, user ID, user ID, language, source, text, geolocation, retweet count, and tweet ID. Only text and user ID

Covid-19 Vaccine Public Opinion Analysis

617

attributes are used in the study. The tweets were collected between 10 and 17 June 2021, duplicate tweets were removed, and a total of 7500 tweets were collected. Tweets were collected at different times, the files were appended and tabulated in one data file, and Retweet and duplicate data were removed and prepared for annotation and preprocessing. Two extraction subsets of 900 tweets from each dataset were prepared for the second phase of the study. 3.2 Data Annotation After duplicate data was removed, the data was manually annotated, where tweets are classified into three polarities: Positive, neutral, and negative. Positive polarity, where tweets indicate a willingness to take the vaccine; Neutral polarity, \where tweets show neutral opinion towards the Covid19 vaccine; Finally, negative polarity, where tweets have a very clear opinion against the Covid-19 vaccine. Of the entire set, 60% of the data was annotated, and the other 40% was left for testing. Examples of the annotated tweets are depicted in Table 1. Table 1. Examples of three annotated Tweets that correspond to three different kinds of polarity: Positive, Neutral, Negative Polarity

Text

1-Positive

RT @CSi: No appointment required! Get vaccinated today! All individuals age 12 and older are currently eligible for a free vaccine

2-Positive

Chi @ PA: Need another reason to get vaccinated? We have more data showing that getting a vaccine may keep you out of the hospital

3-Positive

RT robinmonotti2: VACCINE ENHANCED DISEASE: “Almost one third of people in the UK who have so far died from the Indian variant Covid19”

1-Neutral

RT CT: Are you a youth aged 12–17 or do you work with youth who have questions about the #COVID19 vaccine. Toronto PFR &

2-Neutral

RT PU: The #COVID19 Dashboard has been updated On Thurs 17 June, 11,007 new cases & 19 deaths within 28 days

3-Neutral

RT thetruthin: #COVID19 Situation in Kamrup (M) seems to be under control as it reported only 158 new COVID-19 cases. According to official

1-Negative

RT drjohnm: Another paper chronicling myocarditis after mRNA #covid19 vaccine in 8 young people. Some were admitted to ICU

2-Negative

DrEricDing: Two patients positive for the #DeltaVariant in Calgary have died. One death was in patient with 2 doses of #COVID19 vaccine

3-Negative

RT @abirballan: @afneil Many individuals don’t benefit from the vaccine It is unethical to ask an individual to take a risk for somebody else

618

S. Ibrahim and S. Abdallah

3.3 Data Preprocessing Post annotation, research progressed to the data processing stage. Before performing any semantic data analysis, datasets were preprocessed to reduce all possible data noise. RapidMiner process Document operator was used to preprocess the data using NLP, depicted in Fig. 2. This operator produced word vectors from string attributes and used the term frequency-inverse document frequency (TF–IDF) to preprocess the data. The first stage of preprocessing was tokenizing in order to create a list of single tokens that can include stand-alone words. The second stage was to filter many unnecessary items, URLs, special characters, emojis, non-ASCII codes, non-English letters, tabs, symbols, punctuations, and abbreviations. Then all tweets were converted to lower case, and all StopWords that occur frequently but don’t carry distinct semantical meaning were removed. In addition, replace operator was used to replace all hashtags and special characters such as “@,” “?”, etc. These characters don’t have significant meaning on NLP. Finally, the two extracted data sets were preprocessed. In addition, the complete dataset collected from Twitter was also preprocessed for this data to be used in the K-means clustering model and implemented in visualization techniques such as WordCloud.

Fig. 2. Representation of the subprocess of the process document that performs preprocessing.

3.4 Data Processing Sentiment analysis. Sentimental analysis, known as opinion mining, deals with text classification that divides text between positive, negative, and neutral. It is a computational study of opinions, where emotions are expressed in given words or sentences and are analyzed by NLP techniques [2]. Pang, Lee and Vaithyanathan (2002) were the first to use a sentiment analysis approach to movie classification using machine learning approaches [8]. Vaccine sentiment analysis was implemented in order to explore the evolution of public opinion and classify sentiments of the extracted tweets related to Covid-19. This analysis utilized the Naïve Bayes classification algorithm, which was applied to classify the tweets according to their polarity. This algorithm is based on Bayes’ Theorem: P(A|B ) = (P(B|A )P(A))/P(B) Naïve Bayes is generally used in the classification technique, specifically on Twitter. Its main feature is to have a strong hypothesis of any condition. Using the annotated developed data, the Naïve Bayes operator in RapidMiner was implemented in the study to train and develop a model in order to classify tweets according to their polarity (positive, neutral. Negative). The naïve Bayes has two-level, the training level, and the

Covid-19 Vaccine Public Opinion Analysis

619

classification level; the training level will predict the input according to its polarity. We based on training 60% of the data and the remaining for testing the data for our model. Extraction data from the two datasets, the Dataset dated to January and the one collected from Twitter, were applied to the Naïve Bayes and tested for performance. For the purpose of the study, these extractions have the exact count of tweets for both subsets. Figure 3. Depicts part of the process used with Naïve Bayes.

Fig. 3. Representation of the Naïve Bayes classifier and the performance operator

Evaluation. Performance classification was implemented to evaluate the statistical performance of the model. K-fold Cross-validation Performance operator with a single K parameter from RapidMiner was used, where data was split between training and testing. The output of the performance operator is a confusion matrix. Clustering is also a machine learning method that groups data according to common characteristics. K-means is a clustering technique implemented in the study to group data into K mutually exclusive clusters. With the least mean distance between the data points and the centroid. The objective is to apply this method in this study and find out the most popular topic and conversation in the tweets. It is a method of vector quantization. It has an objective to partition N observations into K clusters, with each term in a document assigned a score based on how prevalent the paper is. Any word with a high score implies more significance or representative than the other words. Therefore, on average, all top words in the different clusters are the most significant in the document. Cluster operator was also implemented using Rapid miner on the complete datasets from Twitter that counts for 7500 tweets. The number of clusters was achieved after consecutive iterations of trying for K number of clusters to reach the number 5.

4 Results and Discussions In this study, the two extracted datasets that count for 900 tweets were implemented and applied to the Naïve Bayes classifier model. As a result, 400 out of 900 or 44.44% of the extracted Datasets from January showed a positive attitude towards the Covid-19 vaccine. Whereas 350 out of 900, 38.89%, showed a neutral attitude or opinion toward the datasets in June. The remaining 150 making 16.67%, were negative. Whereas for the datasets collected in June 2021, 550 out 900 or 61.12% adopted a positive attitude,

620

S. Ibrahim and S. Abdallah Table 2. Percentage of polarity results for the two Datasets.

Percentage Polarity in January

Polarity percentage in June

Positive Polarity: 44.44%

Positive Polarity: 61.12%

Neutral Polarity: 38.89%

Neutral: Polarity 27.89%

Negative Polarity:16.67%

Negative Polarity: 11%

251 out 900 or 27.89% related to a neutral count of polarity, and 99 out 900 adopted a negative count of polarity or 11%. A summary is provided in Table 2. These results showed an increase in positive polarity count, indicating an increase in the positive attitude towards the Covid-19 vaccine over time (from Jan to June 2021). Analyzing the data from both sets will show words such as “take the vaccine,” “appointment,” “doses,” and “received.” Also, there was a decrease in neutral polarity, which could mean that the neutral attitude towards the Covid19 vaccine was changing and decreasing. Looking at tweets with neutral polarity showed no implication for or against the vaccine. Finally, negative polarity counts decreased between data collected in January and June. In addition, from analyzing the tweets, these tweets included words such as “rejection” or mentioning “death cases.” Therefore, the decrease in negative polarity could be translated as fewer people rejecting the vaccine and more willing to take the covid19 vaccine. Two graphs that depict the sentiment analysis result of the tweets for January and June according to polarity are represented in Figs. 4 and 5.

Fig. 4. Visualization of the data of January according to polarity

Fig. 5. Visualization of the data of June according to polarity

Covid-19 Vaccine Public Opinion Analysis

621

As mentioned earlier, the model is evaluated by K fold cross performance. The result was the Confusion matrix to estimate the classifier’s performance by indicating the number of correct versus incorrect predictions. It identified true values such as True Positive, True Neutral, and True Negative. The confusion matrix’s objective is to test the accuracy of the training and the testing data. We have two datasets for January and June that use two different Naïve Bayes classifiers, so we had two annotated Datasets used to train the classifiers; after performing training and testing, the accuracy of the confusion matrices is depicted in Tables 3 and 4, respectively. The Accuracy of the model for data in January is equal to 78.889%, and its Kappa = 0.665, 330, 280, and 100 from the confusion matrix for data selected in January are correct True polarity for tweets, i.e., True positive, True neutral, and True negative. The rows presented the tweets for positive, neutral, and negative polarity. Similarly, for the data in June, the Accuracy = 80.556%, and the Kappa = 0.645489, 480, 170, 75 are the correct polarity for tweets, i.e., True positive, True neutral, and True negative, respectively. Table 3. Confusion matrix resulted after implementing the performance operator: accuracy = 78.889%, & Kappa = 0.665. True positive

True neutral

True negative

Class precision (%)

Prediction Positive/Actual

330

35

35

82.5

Predictive Neutral/Actual

30

280

40

80

Predictive Negative /Actual

20

30

100

66.667

Class Recall

86.842%

81.15%

7.143%

Table 4. Confusion matrix resulted after implementing performance operator accuracy = 80.556% & Kappa = 0.645489 True positive

True neutral

True negative

Class precision (%)

Prediction Positive/Actual

480

40

30

82.5

Predictive Neutral/Actual

51

170

30

80

Predictive Negative /Actual

9

15

75

66.667

Class Recall

88.889%

75.556%

55.556%

This study addressed people’s opinions and attitudes towards Covid-19 by analyzing Twitter posts. Results showed an increase in positive polarity as time progressed and a decrease in neutral polarity and negative polarity on Twitter. It was noticeable that

622

S. Ibrahim and S. Abdallah

tweet conversations tend to be more toward positive or pro-vaccine attitudes in different times where positive polarity exceeded neutral and negative polarity in Datasets. This also shows that people’s attitudes were more pro-vaccine than anti-vaccine or neutral vaccine attitudes. And from the analysis of the result of the positive polarity, there was a more common general acceptance in the tweets for vaccination becoming a must. However, some of the analyses of negative tweets showed extreme side effects that were very rigid and refused the Covid-19 vaccine. It was also important to look at tweet conversations and topics to investigate people’s opinions and attitudes. Using K-means clustering to analyze the tweet, the most frequent 25 topics used in the clusters were selected from the tweets and ranked. Many graphical visualizations were used in this study using k means clustering methods. Table 5 provides the top 25 popular frequent words with their scores in the five clusters related to the Covid-19 vaccine. Topics such as” third,” “dose,” “received,” “appointment,” dates,” “vaccination,” and “available” infer that people’s tweets related to the pro-vaccine conversation. Also, words such as “total, “Pfizer,” “sputniK,” “Moderna,” “AstraZeneca” showed that the tweets’ conversations are more inline of pro-vaccine. In addition, a graph that depicts the ranking of the various topics is presented in Fig. 6. The graph shows “dates” is the first in ranking and “availability “ranks last. This implies that people are looking to book appointments to register for the vaccine. The other topic in ranking stands for “coronavirus,” “Africa,” and “important,” indicating a sentiment that vaccine is important. “third” where people are considering a third vaccine, other topics such as “delta” variant where it seems that people are becoming more aware of the Indian variant and are more worried about it.

Fig. 6. Represents the top 25 topics according to io its rankings with their frequency

The visual graph in Fig. 7 depicts how each topic is represented according to its usage percentage, and different colors indicate each cluster. For example, it is clear from the graph that “dates” are very common and frequently used in clusters 1,2,3 and 4. Similarly, “coronavirus” and “third” are used in clusters 1,2,3 and 4. While some other topics, such as covishield, are the least used in all clusters, “capacity” has minimal use in clusters 4,3,2, etc. Furthermore, if you examine cluster 0 is barely presented. Similarly, Fig. 8 visualizes K-means clustering. Still, this graph represents the frequent use of the topic using the intensity of the shade of color and according to each cluster where each cluster is presented in a single row. It is so clear that cluster 0 is unshaded or has white color and cluster one has the most frequent topic such as “dates’,

Covid-19 Vaccine Public Opinion Analysis

623

Table 5. The most common word used in the 5 clusters.

Fig. 7. Prevalent of the most 25 frequent topics and their prevalence in each cluster

“vaccine,” “variant,” and “vaccination.” This clearly shows that more conversation leads to more positive polarity and a positive attitude towards the vaccine. WordCloud was also implemented in the study. The tool is a standard text analysis tool that visualizes the word of the most frequent term used in the tweet. It was created using the K-means clustering results using the most popular topic in these 5 clusters. It is used according to the frequency of occurrence of these topics. If you examine the word cloud, words such as “appointment,” “vaccine,” “take,” “third,” “vaccinated,” “Sinopharm,” Pfizer, and Moderna stand out. This word cloud can create a story to tell.

624

S. Ibrahim and S. Abdallah

Fig. 8. Heat map Visualizing the different topics represented in different clusters described as a shade of color.

Examining this word cloud tells a clear story to understand and analyze the situation. Here again, the story will continue towards more acceptance of the vaccine words such as “vaccinated,” “dose,” etc. The words at the center and most in each cluster are depicted in Fig. 9.

Fig. 9. WordCloud for the most 25 frequent words

These five graphs show each cluster and represent the two center words prevalent in each cluster. For example, cluster 0 has “sputnik” and “pandemic” forming its center. Whereas cluster 1 has “doses” and “capacity,” similarly, cluster 2 has “dates” and “third,” cluster 3 has “coronavirus” and “Africa,” and cluster 4 has “Moderna” and “Pfizer” (Fig. 10).

5 Conclusion Social media data such as tweets on Twitter is becoming a platform for disseminating ideas and individual opinions, enabling this platform to forecast and shape behavior.

Covid-19 Vaccine Public Opinion Analysis

625

Fig. 10. Scatter plots that visualize the center of the cluster using two words

With the current pandemic, understanding and reacting to people’s emotions, attitudes and opinions are a key component of public health plans, and it is very critical for all organizations and governments to encourage mitigation and strategic public health promotion plan in order to encourage vaccine acceptance and vaccine intake and decrease vaccine hesitancy and opposition. Twitter is an excellent viable source for testing and examining issues related to people’s attitudes and emotions. Insights gained from analyzing real-time conversations create more opportunities and measures for public health organizations to promote and affect possible change in a more dynamic and timely manner. This study used Naïve Bayes to investigate public opinion and attitudes related to the Covid-19 vaccine. In addition, our study examined the clustering technique using the K-means algorithm to look at the diverse topical issues on people’s minds and expressed in tweets. Our research examines the progression of people’s attitudes and opinions by comparing two different datasets from different times. Our findings demonstrated that people’s opinions and attitudes show that tweets’ positive polarity towards the Covid19 vaccine increases over time. On the other hand, negative and neutral opinions are decreasing, and people on Twitter express more willingness to take the vaccine. This is aligned with findings in other recent studies on the vaccine [17]. The visualization technique using k-means clustering methods provides a way to extract popular topics and better understand the tweets, opinions, conversations, and topics relating to the Covid-19 vaccine. Limitations: Few limitations are affecting this study. The data size that can be applied on a Naive Bayes classifier on a desktop is minimal, increasing the chance that the system will not accomplish the objective with a moderate data size. Second, English is not the only language used in the tweets.

626

S. Ibrahim and S. Abdallah

These limitations will provide opportunities for further research to analyze tweets from different languages, or in other geographical locations, and with different cultures that might affect public opinion and attitude.

References 1. Cinelli, M. et al.: The COVID-19 social media infodemic. Sci. Rep., 1–11 (2020) 2. Villavicencio, C., Macrohon, J.J., Inbaraj, X.A., Jeng, J., Hsieh, J.: Twitter sentiment analysis towards COVID-19 vaccines in the Philippines using Naïve Bayes. Informatiom 12 (2021) 3. Eibensteiner, F., Ritschl, V., Nawaz, F.A., Fazel, S.S.: People’s willingness to vaccinate against COVID-19 despite their safety concerns: twitter poll analysis. J. Med. Int. Res. 23(February), 1–10 (2021) 4. Pristiyono, Ritonga, M., Al Ihsan, M.A., Anjar, A., Rambe, F.H.: Sentiment analysis of COVID-19 vaccine in Indonesia using Naïve Bayes Algorithm. IOP Conf. Ser. Mater. Sci. Eng. 1088(1) (2021) 5. Sooknanan, J., Mays, N.: Harnessing social media in the modelling of pandemics—challenges and opportunities. Bull. Math. Biol. 83(5), 1–11 (2021) 6. Mutanga, M.B., Abayomi, A.: Tweeting on COVID-19 pandemic in South Africa: LDA-based topic modelling approach. African J. Sci. Technol. Innov. Dev., 1–10 (2020) 7. Hou, K., Hou, T., Cai, L.: Public attention about COVID-19 on social media: an investigation based on data mining and text analysis. Pers. Individ. Dif. 175, 110701 (2021) 8. Ragini, J.R., Rubesh Anand, P.M., Bhaskar, V.: Big data analytics for disaster response and recovery through sentiment analysis (2018) 9. Alshalan, R., Al-khalifa, H., Alsaeed, D., Al-baity, H., Alshalan, S.: Detection of Hate Speech in COVID-19 – related tweets in the arab region : deep learning and topic modeling approach. J. Med. Internet Res. 22(12) (2020) 10. Id, R.P. et al.: Examining the effect of information channel on COVID-19 vaccine acceptance. PLoS One, 1–14 (2021) 11. Lyu, J.C., Luli, G.K., Lyu, J.C.: Understanding the public discussion about the centers for disease control and prevention during the COVID-19 Pandemic using twitter data: text mining analysis study. J. Med. Internet Res. 23 (2021) 12. Garcia, K., Berton, L.: Topic detection and sentiment analysis in twitter content related to COVID-19 from Brazil and the USA. Appl. Soft Comput. J. 101, 107057 (2021) 13. Guo, J., Christina, A., Ms, L.R., Bsn, S.E.W., Cloyes, K.G.: Mining twitter to explore the emergence of COVID-19 symptoms. Willey, pp. 934–940 (2020) 14. Mathur, A., Kubde, P., Vaidya, S.: Emotional analysis using Twitter data during pandemic situation: Covid-19. Proceedings 5th International Conference Communication Electronics System ICCES 2020, ICCES, pp. 845–848 (2020) 15. Satu, M.S., et al.: TClustVID: a novel machine learning classification model to investigate topics and sentiment in COVID-19 tweets. Knowl. Based Syst. 226, 107126 (2021) 16. Madani, Y., Erritali, M., Bouikhalene, B.: Using artificial intelligence techniques for detecting Covid-19 epidemic fake news in Moroccan tweets. Results Phys. 25, 104266 (2021) 17. Yousefinaghani, S., Dara, R., Mubareka, S., Papadopoulos, A., Sharif, S.: An analysis of COVID-19 vaccine sentiments and opinions on Twitter. Int. J. Infect. Dis. 108, 256–262 (2021)

Medical Application of Deep Learning-Based Detection on Malignant Melanoma Abdulkader Helwan1

and Mohamad Khaleel Sallam Ma’aitah2(B)

1 School of Engineering, Lebanese American University, Beirut, Lebanon 2 Near East University, TRNC, Near East Boulevard Mersin 10, P.O. BOX: 99138, Nicosia,

Turkey [email protected]

Abstract. The application of machine learning and deep learning has over the years revealed a high level of image accuracy in medical diagnosis and classifications. This study was focused on the application of deep learning in the staging of Malignant Melanoma coupled with the use of VGG-16 and AlexNet architecture in the training over different datasets, for comparison purposes. The first objective of the study was to analyze the impact of deep learning in Melanoma identification. This was accompanied with the acquisition of datasets for analysis. The dataset used in this study consists of 5342 images that are classified as melanoma and non-melanoma. Moreover, the two models (AlexNet and VGG-16) were used for the classification 7 different classes using 7 datasets consisting of actinic keratosis; Basal cell carcinoma (bcc), benign keratosis-like lesions, dermatofibroma, Melanoma, Melanocytic nevi and Vascular lesions images. The result from this study showed a highly precise performance with an accuracy of 96.3%, recall (96.5%), specificity (96%), precision (96%), and a F1 score (96.2%). For the 7 datasets model, AlexNet had an accuracy of 90.2%. Experimentally, the result obtained from this study has revealed that the possible use of deep learning in Malignant Melanoma detection is a highly effective and accurate technique. Keywords: Deep Learning · Malignant Melanoma · Visual Geometry Group (VGG-16) · AlexNet Architecture

1 Introduction Melanoma has been regarded to be one of the deadliest forms of skin cancer in both male and female. However, recent incidence of Melanoma has increased drastically, thus making high demands for remediating solutions. Melanoma, is known for affecting the melanocytes cells, thus inhibiting melanin synthesis. It has been recorded that about 75% of deaths relating to cancer has been linked to Melanoma diagnosis [1]. The categories of skin lesions can be divided into Melanocyte (melanocytic nevi and melanoma) as well as non-melanocyte (“actinic keratosis”, “basal cell carcinoma”, “vascular” and “dermatofibroma” lesions). The two lesion types are known to have malignant, which is composed of “melanoma”, “actinic keratosis” and “basal cell carcinoma”) coupled with benign Melanoma [2]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 627–637, 2023. https://doi.org/10.1007/978-3-031-20429-6_56

628

A. Helwan and M. K. S. Ma’aitah

Fig. 1. Melanoma skin lesion (Kassani and Hosseinzadeh, 2019).

Melanocytes are cell regions in the skin that are responsible for the production of Melanin (see Fig. 1) to give colour to the skin, eyes and hair [3, 4]. Melanoma is known to commonly affect different regions of the skin such as the arms, face, neck as well as the legs, which is usually exposed to sunlight. It has been reported in the United States of America that a life is lost in every one hour as a result of Melanoma, hence annually 87,110 cases are recorded with 9730 mortality rates. A similar case of 6800 Melanoma related deaths occurs in Canada [5, 6]. The initial diagnosis measure for Melanoma was carried out by dermatologists through oral skin examinations, thus taking measurement of the colour variations, diameter, irregularity of border and asymmetric measurements. This method was relatively slow and early diagnosis was impossible as well as the pain associated with biopsies. Moreover, several reasons such as the high cost for skin cancer treatment especially with the malignant type of melanoma has propelled scholars and researchers to initiate a more flexible and accurate algorithm that can be very effective in the early diagnosis of melanoma [7]. Hence, the early detection of Melanoma can drastically reduce the mortality rate due to the fact that early detected Melanoma can be treated successfully. Another advantage of the use of innovative (dermoscopic imaging) techniques can reduce the effect of invasive measures usually employed by many dermatologists. The removal of hair and ruler marking is known as the pre-processing phase, the separation of the lesion skin from other skin is termed segmentation [8]. Deep learning has been known to be a very effective and quick method in the staging of Malignant Melanoma and other forms of carcinogenic lesions. In deep learning process, raw images are imputed into the system for analysis, thus bypassing the complex process of pre-processing and segmentation. The use of convolutional neural network in diagnosis has been in play for a quite good time [9]. Since the use of AlexNet architecture in imaging net classification, the utilization of neural network systems has gained increased attention. Furthermore, deep learning has eliminated the handcrafting computer techniques and has modulated an engineering learning process. Hence, laboratory extraction is no longer required in diagnosis but rather the use of labelled sets of data [10–13]. This research study investigates the application of deep learning in identifying the malignant Melanoma through the training of deep networks on large datasets of

Medical Application of Deep Learning-Based Detection

629

Melanoma versus non-Melanoma images. Diagnosis. Moreover, in achieving the desired aim of this study, several objectives were outlined: • • • •

Analyzing the impact of deep learning in malignant Melanoma diagnosis. Acquisition of Melanoma dataset for deep learning analysis. AlexNet and VGG-16 deep learning simulation analysis of acquired datasets. Interpretation of analyzed results and comparison of the proposed methods to other related works.

The limitations of the study were also discussed as well as future recommendations in getting a more accurate imaging in the diagnosis of malignant Melanoma.

2 Related Works Over the years, the use of “deep convolutional neural network” has gained a lot of attention especially in the area of image detection for medical diagnosis [14]. Kawahara et al. carried out a pre-trained convolutional neural network skin classification images on a dataset, which was faster than training a new parameter [15]. An accuracy of 85.8% for 5 classes was obtained with a reduction in training time due to the use of a pre-trained parameter. The use of colour features in the detection of skin cancer was developed by He et al. [16]. An approximate amount of 82% accuracy was achieved with this technique. The highest and lowest accuracy was for both ‘intradermal nevi and ‘melanoma’ with an image of 100% and 77% respectively. Lopez et al. [17] also conducted a study on melanoma detection through the use of deep learning technique [17]. The use of transfer learning approach and VGGNet architecture was also effective in classifying melanoma lesion. A sensitivity record of 78.66% was achieved in the study. Moreover, the utilization of diagnostic tree for skin tumour classification based on fuzzy logic and neural networking. The study included four different lesion class: “basal cell carcinoma, intradermal and malignant melanoma. The use of transfer technique was also used by Menegola et al. [18] to enhance deep learning for the staging of melanoma skin lesion. However, the experimental results obtained in their study showed that sensitivity can highly be screened using this technique [18]. The utilization of segmentation technique was also conducted by [19] in the staging of melanoma via the combination process of genetic algorithm and neural network. The proposing of border detection in this study did not yield regularity in image identification, hence causing difficulty in melanoma detection [20]. Li and Shen [21] analyzed a deep convolutional neural network using large dataset that contained a high imaging resolution. A (Fully Connected) FC layer was initiated into the study to resolve the issues of over fitting [21]. In the classification of skin lesion, Ayan [22] used augmented and nonaugmented dataset in analyzing convolutional neural network performance [22]. However, the studies showed that inadequate datasets can be analyzed with the augmentation technique, in the development of classifiers. The detection of melanoma through the use of multi-scaled lesion biased as well as a classification measure using joint reversed technique, was proposed by in [23]. In this study, the lesion was divided into two groups (melanoma and non-melanoma)

630

A. Helwan and M. K. S. Ma’aitah

with a 92% accuracy in imaging. The use of “Kullback leibler colour histogram” was utilized in the classification of skin lesion. This was used to test for colour irregularity. However, an 80% rate of classification was obtained at the end of the study [24]. The use of computer aided systems in cancer diagnosis such as the use of deep learning algorithm has been beneficial and effective in the resonating of poor image qualities and resolutions, other visual obstacles that including noise, hairs, shadow, reflections, and artefacts, thus eliminating any obstacle that will interfere with the diagnostic imaging of the skin lesion. The review discussed above has revealed the crucial act of gaining accuracy in the detection of melanoma as well as other skin cancer types. However, the emergence of high accuracy in the use of deep learning technique for cancer diagnosis can also be identified from studies conducted by different researchers [25]. Furthermore, studies by [26] have revealed the high level of accuracy that can be obtained through the use of AlexNet architecture.

3 Methodology AlexNet transfer learning knowledge is employed in this research. The major aim is to use relatively complex and successful pre-trained model which has been trained from large data source such as ImageNet. These pre-trained models are trained on a subset of ImageNet which has 1000 categories. The learnt knowledge is then transferred to a relatively smaller task which is the classification of skin images into two categories: Melanoma and Non-Melanoma. Our result was compared with support vector machine and other deep neural approaches. 3.1 Dataset The skin cancer dataset from Kaggle has 5342 images that are classified as melanoma and 5342 images that are classified as non-melanoma. We used 1000 each from the two classes. We further divided it into training and testing in the ratio 0.8 to 0.2 (Figs. 2 and 3).

Fig. 2. Melanoma images.

3.2 Networks Fine-Tuning Since none of the pre-trained classes of the AlexNet, we could not directly apply AlexNet as the feature extractor. In our transfer learning scheme as shown in Table 1, we used a

Medical Application of Deep Learning-Based Detection

631

Fig. 3. Non-melanoma images.

new randomly initialized fully connected layer with two neurons and a new classification layer with just two classes (Melanoma and non-melanoma). AlexNet Softmax layer and classification layer was fine turned to fit this research. As shown in table below, in our transfer learning scheme, we used fully connected layers with two neurons and a new classification layer with two classes (melanoma and non-melanoma). Table 1. Modified layer of AlexNet. Layer

Original

Used

23

FCL (1000)

FCL (2)

25

Classification Layer (1000)

Classification Layer (2)

We set the training options as shown in Table 2. For a transfer learning, the epoch should be small, hence we set training epochs to 20. In addition, a very small learning rate of 0.001 was used. Table 2. AlexNet training options. Parameter

Value

Epoch

20

Learning Rate

0.001

Minimum batch size

64

The 600 by 450 original images were scaled down to 227 × 227 which was used as input to the AlexNet model. 3.3 Evaluation Metrics The metrics used for evaluating the employed models in this study are the following: Precision (PRC), accuracy (ACC), Recall (RCL), specificity (SPC) and F1 Score. ACC =

TP + TN TP + TN + FP + FN

(1)

632

A. Helwan and M. K. S. Ma’aitah

PRC =

TP TP + FP

(2)

SPC =

TN TN + FP

(3)

RCL =

TP TP + FN

(4)

F1 Score = 2 ∗

PRC ∗ RCL . PRC + RCL

(5)

where TP = True Positive TN = True Negative FP = False Positive FN = False Negative

4 Result and Discussion The models employed in this study were trained and tested on two datasets. The first dataset contains images of Melanoma and Non-Melanoma (Dataset 1). Dataset two contains images of seven different skin diseases: Actinic keratosis and intraepithelial or Bowen’s disease (akiec); Basal cell carcinoma (bcc); Benign keratosis-like lesions (bkl); Dermatofibroma (df); Melanoma (Mel); Melanocytic nevi (nv) and Vascular lesions (angiomas, angiokeratomas, pyogenic granulomas and hemorrhage; vasc) both models were trained and tested on these two datasets as shown in Table 3. Table 3 shows the comparison of the two approaches used in this study. It is seen that transfer learning of VGG-16 has outperformed the AlexNet on dataset 1 which includes the classification of images into Melanoma and Non-Melanoma. Similarly, on dataset 2 which includes the classification of skin images into seven different classes, VGG-16 could achieve better generalization capability than that reached by AlexNet. The confusion matrices of AlexNet and VGG-16 for dataset 1 (Melanoma and Not Melanoma) are shown in Figs. 4 and 5 respectively. Figure 6 shows the confusion matrix of the models on dataset 2. AlexNet achieved an accuracy of 96.3% and VGG-16 had an accuracy of 98.5%. It is obvious that VGG-16 showed the best performance compared to AlexNet. In this study, different trainings were carried out from Melanoma and Non-melanoma skin images. Also, a training was also conducted for 7 data’s: Actinic keratosis and intraepithelial or Bowen’s disease (akiec); Basal cell carcinoma (bcc); Benign keratosislike lesions (bkl); Dermatofibroma (df), Melanoma (Mel); Melanocytic nevi (nv) and Vascular lesions (angiomas, angiokeratomas, pyogenic granulomas and hemorrhage; vasc). This comparison was done for AlexNet and VGG-16. For the analysis between melanoma and non-melanoma, AlexNet had the highest accuracy of 96.3% (Fig. 3) compared to an accuracy of 90.2% for the other 7 dataset. However, this result revealed that a lesser class of classification during the training of datasets gives a better reading and higher accuracy.

Medical Application of Deep Learning-Based Detection

633

Table 3. Training model of dataset. Dataset 1

AlexNet

VGG-16

Training accuracy

96.3%

98.5%

Testing accuracy

92.5%

94.8%

Dataset 2

AlexNet

VGG-16

Training accuracy

90.2%

96.0%

Testing accuracy

88.0%

95.1%

Fig. 4. AlexNet confusion matrix.

Comparisons were also made with different studies on the numbers of images and accuracy (Table 5). A study conducted by Lopex et al. [27], it was observed that the accuracy for VGG-16 in the test dataset was 81.33% which is lesser compared to the result observed in this study [27]. A reason for this was as a result of a training between the two datasets (Malignant and Benign) which was also observed in the test of Malignant and Non-malignant in this study for VGG-16 (67.3%).

5 Conclusion In this study, we proposed transfer learning method using AlexNet and VGG-16. This solution can be used by dermatologist during the diagnosis of skin lesions. The results

634

A. Helwan and M. K. S. Ma’aitah

Fig. 5. VGG-16 confusion matrix.

in Table 4 show that this proposed VGG-16’s approaches achieved promising result over dataset 1 and 2, with accuracy of 98.5%, a recall or sensitivity of 95.1%, a specificity of 95.5%, a precision of 97.8% and F1 score of 96.3%. Moreover, conclusions can be drawn from the findings in this study that due to the fact that melanocyte and nonmelanocyte consist of multiple skin lesion, it is more difficult to classify and detect them, hence the level of accuracy observed in this study has revealed the high level of efficiency and accuracy in the use of deep learning for melanoma staging. Future study can compare other transfer learning models such as VGG-19, ResNet, DenseNet etc. Also, good data augmentation techniques can also be implemented to improve the network’s performance.

Medical Application of Deep Learning-Based Detection

Fig. 6. Dataset 2.

Table 4. Evaluation Metrics. Dataset 1

ACC

SPC

RCL

PRC

F1-SCORE

ALEXNET

96.3%

0.960

0.965

0.96

0.962

VGG-16

98.5%

0.955

0.951

0.978

0.963

Dataset 2

ACC

SPC

RCL

PRC

F1-SCORE

ALEXNET

90.2%

0.929

0.944

0.904

0.933

VGG-16

96.0%

0.955

0.951

0.978

0.962

ACC: Accuracy, SPC: Specificity, RCL: Recall

635

636

A. Helwan and M. K. S. Ma’aitah Table 5. Comparison of Model Performance with Previous Studies.

Parameters AlexNet VGG-16

AlexNet-SVM VGG16-SVM CNN VGG16 AlexNet

Number of 15,420 images

15,420

15,420

14,110

CNN [27]

Accuracy

96.0%

92.6%

97.1%

CNN 81.33% 84.7%

90.2%

[1]

Conflict of Interest. The author of this paper declares that there is no conflict of interest regarding the publication of this paper.

References 1. Jojoa Acosta, M.F., Caballero Tovar, L.Y., Garcia-Zapirain, M.B., Percybrooks, W.S.: Melanoma diagnosis using deep learning techniques on dermatoscopic images. BMC Med. Imaging 21(1), 1–11 (2021) 2. Sertan, K., Parvaneh, E., Ali, S.: Deep learning for two-step classification of malignant pigmented skin lesions, member, IEEE. Symposium on neural networks and applications (Neurel) Belgrade. Serbia 1, 20–21 (2018) 3. Al-masni, M., Al-antari, M., Choi, M., Han, S., Kim, T.: Skin lesion segmentation in dermoscopy images via deep full resolution convolutional networks. Comput. Methods Programs Biomed. 3, 76–98 (2018) 4. Flores, E., Scharcanski, J.: Segmentation of melanocytic skin lesions using feature learning and dictionaries. Expert Syst. Appl. 56, 300–309 (2016) 5. Siegel, R., et al.: Colorectal cancer statistics, CA: a Cancer. J. Clin. 67, 177–193 (2017) 6. Siegel, R., Miller, K., Jemal, A.: Cancer statistics, CA: A Cancer. J. Clin. 66, 7–30 (2016) 7. Pathan, S., Prabhu, K.G., Siddalingaswamy, P.C.: Techniques and algorithms for computer aided diagnosis of pigmented skin lesions—a review. Biomed. Signal Process. Control. 39, 237–262 (2018) 8. Bush, I.J., Abiyev, R.H., Mohammad, K.M.: Intelligent machine learning algorithms for colour segmentation. WSEAS Trans. Signal Process. 13, 232–240 (2017) 9. LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. nature 521(7553), 436–444 (2015) 10. Ma’aitah, M.K.S., Abiyev, R., Bush, I.J.: Intelligent classification of liver disorder using fuzzy neural system. Int. J. Adv. Comput. Sci. Appl. 8(12) (2017) 11. Helwan, A., El-Fakhri, G., Sasani, H., Uzun Ozsahin, D.: Deep networks in identifying CT brain hemorrhage. J. Intell. Fuzzy Syst. 35(2), 2215–2228 (2018) 12. Helwan, A., Ma’aitah, M.K.S., Hamdan, H., Ozsahin, D.U., Tuncyurek, O.: Radiologists versus deep convolutional neural networks: a comparative study for diagnosing COVID-19. Comput. Math. Methods Med. (2021) 13. Abiyev, R.H., Ma’aitaH, M.K.S.: Deep convolutional neural networks for chest diseases detection. J. Healthcare Eng. (2018) 14. Helwan, A., Ma’aitah, M.K.S., Uzelaltinbulat, S., Sonyel, B., Altobel, M.Z.Z., Darwish, M.: Stacked autoencoders deep learning approach for left ventricular localization in magnetic resonance slices. In: International Conference on Emerging Technologies and Intelligent Systems, pp. 225–234. Springer, Cham (2021) 15. Kawahara, J., BenTaieb, A., Hamarneh, G.: Deep features to classify skin lesions. 2016 IEEE 13th International Symposium on Biomedical Imaging (ISBI), pp. 1397–1400 (2016)

Medical Application of Deep Learning-Based Detection

637

16. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2016) 17. Lopez, A.R., Giro-i-Nieto, X., Burdick, J., Marques, O.: Skin lesion classification from dermoscopic images using deep learning techniques. 2017 13th IASTED International Conference on Biomedical Engineering (BioMed), pp. 49–54 (2017) 18. Menegola, A., Fornaciali, M., Pires, R., Bittencourt, F.V., Avila, S., Valle, E.: Knowledge transfer for melanoma screening with deep learning. Proceedings - International Symposium on Biomedical Imaging (2017) 19. Celebi, M., Wen, Q., Iyatomi, H., Shimizu, K., Zhou, H., Schaefer, G.: A state-of-the-art survey on lesion border detection in dermoscopy images. Dermoscopy Image Analysis 1, 97–129 (2015) 20. Codella, N., et al.: Deep learning ensembles for melanoma recognition in dermoscopy images. IBM J. Res. Dev. 61(4), 1–5 (2017) 21. Li, Y., Shen, L.: Skin lesion analysis towards melanoma detection using deep learning network. Sensors 18(2), 556–569 (2018) 22. Ayan, E., Ünver, H.M.: Data augmentation importance for classification of skin lesions via deep learning. In: 2018 Electric Electronics, Computer Science, Biomedical Engineerings’ Meeting (EBBT), pp. 1–4. IEEE (2018) 23. Chakravorty, R., Liang, S., Abedini M., Garnavi, R.: Dermatologist-like feature extraction from skin lesion for improved asymmetry classification in PH2 database. 38th International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 3855– 3858 (2016) 24. Jafari, M.: Skin lesion segmentation in clinical images using deep learning. 2016 23rd International Conference on Pattern Recognition (ICPR), pp. 337–342 (2016) 25. Hu, Z., Tang, J., Wang, Z., Zhang, K., Zhang, L., Sun, Q.: Deep learning for image based cancer detection and diagnosis − a survey. Pattern Recognit 83, 134–149 (2018). “ISIC 2018 [Online]. Available: https://challenge2018.isicarchive.com/ 26. Dorj, U.-O., Lee, K.-K., Choi, J.-Y., Lee, M.: The skin cancer classification using deep convolutional neural network. Multimedia Tools Appl. 77(8), 9909–9924 (2018). https://doi.org/ 10.1007/s11042-018-5714-1 27. Lopez, A.R., Giro-i-Nieto, X., Burdick, J., Marques, O.: Skin lesion classification from dermoscopic images using deep learning techniques. In: 2017 13th IASTED International Conference on Biomedical Engineering (BioMed), pp. 49–54. IEEE (2017)

Privacy-Preserving Data Aggregation Scheme for E-Health Matthew Watkins1 , Colby Dorsey1 , Daniel Rennier1 , Timothy Polley1 , Ahmed Sherif1(B) , and Mohamed Elsersy2 1 School of Computing Sciences and Computer Engineering, University of Southern

Mississippi, Hattiesburg, MS, USA {Matthew.Watkins,Colby.Dorsey,Daniel.Rennier,Timothy.Polley, ahmed.sherif}@usm.edu 2 Computer and Information Systems Deparment, Higher Colleges of Technology, Al Ain, Abu Dhabi, UAE [email protected]

Abstract. E-Health is using digital services and communication technology to support healthcare. E-Health services are becoming increasingly popular. With EHealth, large amounts of data need to be collected, stored, and sent to other places while remaining private. This raises the need for privacy-preserving data aggregation schemes to be implemented. Many other privacy-preserving data aggregation schemes already exist for E-Health services utilizing tools such as homomorphic encryption, which can be slow with large amounts of data. This paper proposes a privacy-preserving scheme to aggregate data in an E-Health setting. Our scheme allows all patients’ data to remain private. Doctors can utilize partial decryption in our scheme to collect specific patient information, such as how many patients have high blood pressure, without seeing all patients’ data. Keywords: Aggregation over encrypted data · E-Health · k-Nearest Neighbor

1 Introduction E-Health is a rapidly expanding healthcare service. E-Health involves using electronic systems and communication in healthcare that carries many benefits, one of them being easy access for doctors to view and modify patient information. To preserve the patient’s privacy, all the data should be encrypted before sending this information to the E-health system. Within E-Health, there is a need to collect lots of information to get statistics that call for data aggregation over encrypted data. Data aggregation is a technology that needs to be used in E-Health architectures to collect mass amounts of medical data for storage and processing [1–7]. Several data aggregation schemes over encrypted data have been proposed to be used in today’s burgeoning E-Health industry. One major category of these schemes uses Homomorphic encryption, which permits users to perform computations on encrypted data without decrypting it [8]. These resulting computations are left in an encrypted © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 638–646, 2023. https://doi.org/10.1007/978-3-031-20429-6_57

Privacy-Preserving Data Aggregation Scheme for E-Health

639

form which, when decrypted, results in an identical output if these operations have been performed on plaintext data. These schemes use Asymmetric encryption, also known as Public-key encryption, which uses pairs of keys. Each pair consists of a public key (which may be known to others) and a private key (which may not be known by anyone except the owner). One privacy problem with data aggregation is collecting and aggregating the information without knowing the individual data. In E-Health architectures, privacy preservation and security are of extreme importance. This means that it is necessary to be able to complete data aggregation over encrypted data without knowing any individual data. A solid and secure privacy-preserving data aggregation scheme in E-Health needs to be secure to maintain the confidentiality of the user’s medical data to preserve privacy. The scheme should also maintain the integrity of the user’s medical data to ensure that the data has not been tampered with or altered in any way. This means that a strong encryption technique must be used to encrypt and protect the data. In this paper, we proposed a privacy-preserving data aggregation scheme over encrypted data using the k-Nearest Neighbor encryption technique (kNN) [9]. The kNN encryption algorithm can perform aggregation on individual bits, unlike other aggregation schemes that perform the addition by using the regular binary addition. This allows for more scalability within the scheme, which means that the aggregation can be performed on data of different types and sizes. The benefit of using the kNN algorithm is that the aggregation can be done over the patients’ encrypted data, and the contents of the individual patient’s data are not revealed. The users of the E-Health system (doctors, nurses, etc.) can request the aggregated data, and no individual data is revealed, allowing for more privacy and security of patients’ data. In addition, by keeping the security of the encryption key, no one can decrypt the patient’s data or the E-Health user’s aggregation request. The remainder of this paper is organized as follows: The related work is discussed in Sect. 2. The network model, attack model, and design goals are discussed in Sect. 3. The proposed scheme is discussed in Sect. 4. The security and privacy analysis is discussed in Sect. 5. The performance evaluation is presented in Sect. 6. The conclusions are made in Sect. 7.

2 Related Work Several aggregation schemes over encrypted data have been proposed [1–5]. In [1, 2], privacy-preserving data aggregation schemes are proposed utilizing homomorphic encryption. Utilizing homomorphic encryption comes with heavy computational overhead, especially when dealing with large amounts of data; however, it allows for strong privacy preservation. In [3], the authors provide a survey on frameworks for secure data aggregation in smart cities, all using Fog architecture. Multiple schemes are proposed for many applications, not only for E-Health. Smart devices communicate autonomously to transmit sensitive data securely across network nodes. In [4], the wireless body sensor network (WBSN) technology is an application of IoT in healthcare, whereas data security and privacy impediments have raised some concerns. The main contribution of this research paper is a cryptographic accumulator based on the

640

M. Watkins et al.

authenticated additive homomorphic encryption, which can collect and accumulate data from IoT wireless wearable devices. These encrypted data can be used for analysis in an encrypted form so that the information is not revealed [5]. Proposes a privacy-preserving health data aggregation scheme that securely collects health data from multiple sources and guarantees fair incentives for contributing patients. The proposed scheme combines a Boneh-Goh-Nissim cryptosystem and Shamir’s secret sharing to keep data obliviousness security and fault tolerance. However, all previous schemes require heavy computational overheads as it depends on asymmetric encryption techniques. Our proposed scheme depends on the kNN encryption technique to aggregate the patients’ data, where the required operations are very light and requires minimal computational overhead.

3 System Models and Design Goals This section will discuss the proposed scheme’s network model, attack model, and design goals. 3.1 Network Model The network model is shown in Fig. 1 consisting of three entities the hospitals, the aggregation server, and the health center.

1 The health center distributes encryption keys 2 The hospital uploads patients’ encrypted data 3 The health center sends aggregation query 4 The server performs aggregation and sends result back Fig. 1. Network model

Privacy-Preserving Data Aggregation Scheme for E-Health

641

– Hospitals: The hospitals collect the data from the patients, encrypt it, and then send it to the aggregation server. – Server: The aggregation server stores the patients’ data and is able to perform an aggregation over encrypted data. The aggregation server sends the aggregation results to the health center. – Health Center: The health center is a federal research center. It is responsible for distributing encryption keys. The health center sends queries to the aggregation server to receive aggregated data to calculate required statistics. 3.2 Attack Model Our proposed scheme could involve many possible attackers, including internal and external attackers. Attacks on the scheme could either be passive or active. In a passive attack, the attacker only wishes to collect the hospitals’ data that are being aggregated. In an active attack, the attacker could collect and modify the data to produce false information or insert new data. In our scheme, the aggregation server is a third party that is not related to the health center but is known to all users. With this, a possible attack method would be honest but curious. Honest but curious is a legitimate participant, such as the aggregation server, in a communication protocol who will not deviate from the defined protocol but will attempt to learn all possible information from the network. 3.3 Design Goals The following three design goals are to be achieved by this privacy-preserving data aggregation scheme. – Efficiency: The first design goal is efficiency, with the idea of reducing the computational and communication costs of accessing and manipulating data while at the same time providing quick and timely access to the authorized personnel. The encryption and aggregation should be performed efficiently and in a timely manner. – Scalability: To allow for encryption and aggregation of data being different types and sizes. – Privacy Preservation: The patients’ individual information should remain private to all users with only the specific aggregated information being accessible to the health center upon its request.

4 Proposed Scheme The proposed scheme consists of four phases. In this section, each phase is discussed with more details.

642

M. Watkins et al.

4.1 Overview A privacy-preserving data aggregation scheme is essential in E-Health for many reasons. A large amount of encrypted data for each patient is collected, but health centers/doctors may not want or need to receive all of the data. This is where aggregation becomes helpful. The health centers/doctors can receive only the desired patient information by receiving the aggregated data from the aggregation server. In our scheme, the key distribution is handled by the health center. The health center distributes keys to the hospitals that will be used to encrypt patients’ data. The encrypted data about each patient is sent to the server from the corresponding hospital. The aggregation of encrypted data is performed at the request of the health center, and the aggregated information is sent to the health center. The health center can then decrypt the information and get the required aggregated data. In the example given below, there are two binary vectors containing some data about each patient P1 and P2 . Each vector cell contains specific information about the patient, such as their blood pressure (BP ) or blood sugar (BS). A value of 1 would indicate that the patient has high blood pressure or blood sugar, and a value of 0 would indicate that the patient has low/normal blood pressure or blood sugar. The binary vector could also contain information about health problems such as if the patient has diabetes (D), cancer (C), lung disease (LD), etc. BP

BS

D

C

LD

P1

1

0

0

1

0

…

P2

0

1

0

1

0

…

4.2 Key Distribution The primary user, which is the health center, will have a key as follows: [SI, X 1 Y 1 , X 1 Y 2 , X 2 Y 3 , X 2 Y 4 ] with SI being a binary vector of size n, where n is the number of health problems/conditions in each patient’s vector. [X 1 , X 2 , Y 1 , Y 2 , Y 3 and Y 4 ] are random invertible matrices with a size of  the hospital, will  n × n. Each secondary user, being have different key as follows: SI , Xi Y1−1 , Xi Y2−1 , Xi Y3−1 , Xi Y4−1 . For the secondary user (i), (X i + X i ) is equal toand (X i + X i ) is equal to X2−1 ) are random invertible i i 2, where (X i , X i , X i , X i ) matrices with a size of n x n. 4.3 Data Encryption and Submission The hospital is responsible for submitting encrypted patient data to the aggregation server. For the encryption in our scheme, a binary vector SI will be used as a splitting indicator. SI is used to split the data vectors vi into two random vectors vi and vi , where vi is the patient (i) data vector. If the jth bit of SI is one, vi (j) and vi (j) are set similar to vi (j). If the jth bit of splitting indicator SI is zero, vi (j) and vi (j) are set to two random numbers that add up to equal vi (j). The  data vector pair, (vi , vi ) is encrypted  to Ci = Xi Y1−1 vi , Xi Y2−1 vi , Xi Y3−1 vi , Xi Y4−1 vi . The ciphertext C i is a column vector with size 4n. Then, C i will be submitted by the hospital to the aggregation server.

Privacy-Preserving Data Aggregation Scheme for E-Health

643

4.4 Aggregation on Patient Data by the Server The aggregation server performs the data aggregation. For the aggregation, C i is a column vector containing data about the patient. The number of elements inside C i is 4n. There are m users, where U i creates an index C i as shown below. k i,j is the jth element in the ith index. T  C1 = k1,1 , k1,2 , . . . , k1,4n T  C2 = k2,1 , k2,2 , . . . , k2,4n C agg is the aggregated index that can be computed by summating all m users’ indices, as shown below. This could give the health center information such as the number of patients with high blood pressure. C agg is then submitted to the health center/hospital. m Ci i=1 m m m m ki,1 , ki,2 , ki,3 , . . . ki,4n = i=1 i=1 i=1 i=1 = [k1 , k2 , . . . k4n ]

Cagg =

4.5 Sending Aggregation Request and Data Decryption

P1

BP

BS

D

C

LD

1

0

0

0

0

…

The decryption can be performed by the health center/doctor (decryptor) who wishes to receive the aggregated patient information. The decryptor uses its key to decrypt and retrieve back the desired aggregated data. The decryption can be full or partial. In full decryption, the decryptor can know the number of patients in all conditions/problems. In partial decryption, the decryptor can get only the number of patients with a specific condition or problem, such as the number of patients with high blood pressure or high blood sugar. To enable partial decryption, the decryptor needs to decrypt the k th element in the aggregated index, for example, to receive the number of patients with high blood pressure. The decryptor should create a binary vector, d with the same size n, where the k th bit in d is set to 1 (representing the high blood pressure) while all other bits are set to 0 as shown above. E(d) will serve as the encryption of vector d. The decryptor will create E(d) by using its secret key. To decrypt the data and view the patients’ blood pressure, the dot product between C agg and E(d) would need to be conducted. Doing so would allow for partial decryption to take place.

644

M. Watkins et al.

5 Privacy and Security Analysis Efficiency. The use of a scheme based on the kNN algorithm ensures efficiency. In our scheme, the time required to encrypt and aggregate data remains low, even with an increased number of data, which allows for these operations to be performed in a timely manner. Scalability. This scheme is highly scalable and allows for data of different types and sizes to be efficiently encrypted and aggregated. Patients’ data privacy. Our scheme preserves the privacy of patients’ data that is being transmitted. The aggregation server is not able to find out any information about the patients’ data because of the encryption used. The patients’ data is encrypted using their keys, and it is infeasible to decrypt the data without knowing the keys. This ensures that the patients’ data remains private to those involved in the scheme. The keys remain private between the users. The secret keys used for encryption in this scheme involve random matrices that are unknown between different users. Since these matrices are unknown, the necessary computations to compute the keys cannot be performed by any attackers. Due to this, the patients’ data cannot be decrypted by any other patient/hospital and remains secure and private.

6 Performance Evaluation Our scheme was implemented using MATLAB. Random binary vectors of 50 elements were used in different amounts for the patients’ data. The execution time of the encryption, aggregation, and decryption was tested five times with the four different amounts of binary vectors, and then the results were averaged and plotted. The code was tested using 50, 100, 150, and 200 vectors to simulate different patients’ data. The performance evaluation of the MATLAB code was conducted on a PC with an Intel Core i7-10700F processor @ 2.90 GHz and 16.00 GB of RAM. In MATLAB, a double is stored using 8 bytes. After performing encryption, the data size is 200 elements which takes up to 1600 bytes of memory. Figure 2 shows the encryption time with different sizes of patient data vectors: 50, 100, 150, and 200. With an increased amount of data vectors, the number of vectors containing patient information, the encryption time also increases proportionally. Although the time increases, it remains low to perform encryption with less than 35 milliseconds. Figure 3 shows the aggregation time with different amounts of patient data vectors: 50, 100, 150, and 200. The aggregation time is the time it takes for the aggregator to sum the encrypted vectors from different hospitals. The time it takes for aggregation increases proportional to the amount of data vectors being aggregated. Figure 4 shows the time required for the partial decryption operation to be performed with different amounts of patient data vectors: 50, 100, 150, and 200. This figure is related to the dot product operation performed for partial decryption where a value of 1 is placed in a specific location with 0’s placed in all other locations to get the number of patients with a particular disease.

Privacy-Preserving Data Aggregation Scheme for E-Health

Fig. 2. Encryption time

Fig. 3. Aggregation time

Fig. 4. Partial decryption time

645

646

M. Watkins et al.

7 Conclusion In this paper, we proposed a privacy-preserving data aggregation scheme for E-Health. Our scheme utilizes techniques based on the kNN encryption technique. Using the kNN technique ensures that patients’ data remains private and can be aggregated efficiently. The techniques utilized in our scheme allow for data of different sizes to be aggregated, while also maintaining our original design goals of efficiency, security, and reliability. With the E-Health industry expected to grow exponentially in the coming years, schemes such as the one we are proposing are of the utmost importance to the continued growth of the field. Patients and doctors expect nothing less than the complete safety and reliability of their sensitive health information, and our kNN-based scheme is the one we feel will be the best fit for them and the entire industry going forward.

References 1. Almalki, F.A., Soufiene, B.O.: EPPDA: an efficient and privacy-preserving data aggregation scheme with authentication and authorization for IoT-based healthcare applications. Wirel. Commun. Mobile Comput. 2021, Article ID 5594159, 18 p., March 2021 2. Ara, A., Al-Rodhaan, M., Tian, Y., Al-Dhelaan, A.: A secure privacy-preserving data aggregation scheme based on bilinear elgamal cryptosystem for remote health monitoring systems. IEEE Access 5, 12601–12617 (2017) 3. Ullah, A., Azeem, M., Ashraf, H., Alaboudi, A.A., Humayun, M., Jhanjhi, N.Z.: Secure Hhealthcare data aggregation and transmission in IOT—a survey. IEEE Access 9, 16849–16865 (2021) 4. Rezaeibagha, F., Mu, Y., Huang, K., Chen, L.: Secure and efficient data aggregation for IoT monitoring systems. IEEE Internet Things J. 8(10), 8056–8063 (2021) 5. Tang, W., Ren, J., Deng, K., Zhang, Y.: Secure Data Aggregation of lightweight E-healthcare IoT devices with fair incentives. IEEE Int. Things J. 6(5), 8714–8726 (2019). https://doi.org/ 10.1109/JIOT.2019.2923261 6. Ren, H., Li, H., Liang, X., He, S., Dai, Y., Zhao, L.: Privacy-enhanced and multifunctional health data aggregation under differential privacy guarantees. Sensors (Basel) (2016) 7. Othman, S.B., Almalki, F.A., Chakraborty, C., Sakli, H.: Privacy-preserving aware data aggregation for IoT-based healthcare with green computing technologies. Comput. Electr. Eng. 101 (2022) 8. Yi, X., Paulet, R., Bertino, E.: Homomorphic encryption. In: Homomorphic Encryption and Applications. SpringerBriefs in Computer Science. Springer, Cham (2014) 9. Wong, W.K., Wai-lok Cheung, D., Kao, B., Mamoulis, N.: Secure kNN computation on encrypted databases. In: Proceedings of the 2009 ACM SIGMOD International Conference on Management of data (SIGMOD ’09). Association for Computing Machinery, New York, NY, USA, pp. 139–152 (2009)

Classification of Skeletal Muscle Fiber Types Using Image Segmentation Mehdy Mwaffeq Mehdy1(B) , Sarah Raad Mohammed2 , Nasser N. Khamiss3 , and Anam R. Al-Salihi4 1 Department of Medical Instrumentation Techniques Engineering, Dijlah University College,

Baghdad, Iraq [email protected] 2 Department Medical Laboratory Techniques, Dijlah University College, Baghdad, Iraq [email protected] 3 Department of Information and Communication Engineering, Al-Nahrain University, Baghdad, Iraq [email protected] 4 Deparment of Anatomy, College of Medicine, Al-Nahrain University, Baghdad, Iraq [email protected]

Abstract. Muscle fibers can be classified in a variety of ways based on their different anatomical and histochemical features and these many classifications usually depend on subjective observations and may contradict each other. Thus, an objective method of grouping is always preferred to have a standardized reference for this complicated human tissue. Microscopic images with appropriate staining techniques are proven to be highly reliable in studying the histology of human body, with different image segmentation techniques are applied on these types of images and provide an encouraging outcome to further analyze the composition of the human tissues. In this study, Fuzzy C-Means algorithm was applied on muscle specimen images stained by alpha naphthyl acetate esterase (ANAE) with different number of muscle fibers within the specimen and different reaction times used for the staining step. The results shown were recommended by anatomical experts to rely on and to further develop with other types of tissues. Unsupervised classification techniques based on fuzzy c-means clustering algorithm have proved successful in segmenting of biopsy image of human skeletal muscle tissue into different fiber types. Keywords: Skeletal Muscle Fiber Types · Image Segmentation · Fuzzy C-means Algorithm

1 Introduction Skeletal muscles fibers are classified into two main types each with its own subdivisions. These types differ in tone and speed of contractions. As the types of muscle fibers in most of the animals are clear and well-defined, human muscle fibers are complicated and heterogenous due to the multitasked movement nature of these muscles [1, 2]. The © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 647–656, 2023. https://doi.org/10.1007/978-3-031-20429-6_58

648

M. M. Mehdy et al.

classification of muscle fibers is a controversial subject because of the disagreements between the different methods used in the classifications as each method rely on a certain property within the muscle fiber such as: histochemistry, morphology or physiology [3, 4]. The importance of the classification was shown by several studies which revealed that the fiber types proportions are affected by the routine of the human body such as in exercises [5], as well as the abnormalities in the musculoskeletal system [6, 7]. The use of biopsy in to image the microscopic structure of the muscles is an effective tool to analyze abnormalities of the muscular system [8]. Many enzymes were used in muscle biopsy and one the most commonly used ones are the esterase enzymes. Esterase enzymes, belonging to the hydrolase enzymes, break down the ester bonds and, thereby, catalyze the hydrolysis of esters. Esterases are classified under 2 groups, nonspecific esterases and specific esterases. Esterases, which hydrolyze simple esters such as naphthylacetate, are classified as nonspecific esterases. Nonspecific esterases are widely distributed in the body and are found in several different types of cells. Practical staining methods, based on cytochemical esterase activity, are commonly used [9]. It’s an Enzyme histochemical stain that relies on endogenous esterase activity to hydrolyze exogenous alpha naphthyl acetate substrate, which yields naphthol, a reddish-brown product visible under light microscopy [10]. Exposure to heat during paraffin processing can denature the enzyme so it does not fit its substrate, which is why these specimens are usually frozen sectioned. Enzymes are also typically sensitive to pH changes [9]. Fuzzy C-Means is an algorithm used for clustering process based on the fuzzy set theory, where the group of the input values are classified into given number of clusters by calculating the centroid of each cluster and the membership function of the input values [11]. The application of fuzzy-based image segmentation techniques on medical images has been widely used for decades with proven high accuracy [12]. Such medical applications include Magnetic Resonance Imaging (MRI) [13, 14], Computed Tomography (CT) [11]. Nowadays, the algorithms of deep learning are considered a basic component in the medical images system [15]. Several studies have been made to apply the image processing methods on muscle microscopic images. Some of which focused on the histological analysis of the muscles like determining the orientation of the myosin filaments [16], edge detection of the muscle fiber borders [17], reducing the time and effort of the tissue staining procedure [18], registration of muscle tissue sections for automated characterization of the muscle fibers [19], different myofibers identification [20] as well as comparing different factors that affect the laboratorial and segmentation procedures [21, 22]. While other studies develop the segmentation algorithms to either approach fully automated process [23], multi-channel microscopic image segmentation [24], processing time improvement [25], accuracy and efficacy enhancement [26]. The summary of the related work is illustrated in (Table 1). This study aims to create a prototype of an objective classification of skeletal muscle fiber types based on the given staining method.

Classification of Skeletal Muscle Fiber Types

649

Table 1. Related work Authors

Segmentation algorithm

Staining method

Papastergiou et al. [17]

Sobel and Laplace

Hematoxylin and eosin

Sertel et al. [19]

Fourier shape descriptors

ATPases

Janssens et al. [23]

Top-down

Hematoxylin and eosin

Liu et al. [18]

Seed detection, contour evolution

Hematoxylin and eosin

Nguyen et al. [24]

Superpixel-based

Fluorescent

Vu et al. [21]

Multiscale deep residual aggregation network

Hematoxylin and eosin

Cui et al. [25]

Parallel framework

Hematoxylin and eosin

Cui et al. [26]

Deep hierarchically connected networks

Hematoxylin and eosin

Achouri [22]

Fiji software

Red azorubin

Rahmati [20]

Neutrosophic set

Immunofluorescent

Our proposed method

Fuzzy C-Means

ANAE

2 Materials and Methods The proposed system in this study is composed mainly of 3 steps: Slice preparation, Image Segmentation and Image Analysis. 2.1 Slice Preparation The data was provided by previous retrospective studies which were done in the specific laboratory in the college of medicine/Al-Nahrain University. The muscle samples were taken from the rectus femoris muscle of 10 healthy adult patients, trimmed and cut into appropriate blocks then frozen to be prepared for histochemical staining using ANAE (Table 2). After the staining process, the sections were examined and photographed at low and high magnification by “freeze” then “capture” function. In order to investigate the effect of the time of the histochemical reaction, several rounds were made with different durations (15, 30 and 45 min). The slices were examined using Reichert Jung Polyvar Microscope. 2.2 Image Segmentation The resultant photos of the slice preparation were the input data of the segmentation process, giving initially the number of clusters to be extracted from the photo, applying the FCM segmentation technique and then analyzing the shape of each object to decide its eligibility to be classified as a muscle fiber type. So, if the given number of clusters exceeds the actual number of muscle fiber types, then the extracted cells would be distorted because their pixels would not probably be in

650

M. M. Mehdy et al. Table 2. Alpha Naphthyl Acetate Esterases Histochemistry

1.

2.

3.

0.2M Phosphate Buffer (pH 7.4) Sodium phosphate dibasic (anhydrous) Na2 HPO4

11.4 g

Potassium phosphate monobasic

2.7 g

Distilled Water

500 ml

Sodium Nitrate 4% Sodium Nitrate

80 mg

Distilled Water

2 ml

Alpha naphthyl acetate 1% Alpha naphthyl acetate

50 mg

Acetone

2.5 m

Pararosaniline Stock Solution: Pararosaniline hydrochloride (Basic fuchsin)

1g

Distilled Water

20 ml

Concentrated HCL

5 ml

The solution was warmed slowly, cooled to room temperature, filtered and stored in refrigerator. It will remain stable for weeks or until depleted Hexazotized pararosaniline Pararosaniline Stock Solution

1 ml

4% Sodium Nitrite (Freshly prepared)

1 ml

Pararosaniline was added to sodium nitrate, it will precipitate if mixed in other direction, then the solution was shaken and kept in ice cold bath till no further NO2 gas can be smelled evolving from the solution Incubation Solution Mixed in order 0.2 mol/L phosphate buffer pH 7.4

20 ml

1% alpha naphthyl acetate

0.5 ml

Hexazotized pararosalinine

1.6 ml

pH adjusted to 6.8 – 7.2 Staining Procedure 1.

The sections were incubate in a 37C – water bath in it for 30 min;:

2.

The sections were washed in tap water for 10 min

3.

The sections were counterstained with 2% methyl green for 5 min

4.

The sections were mounted in PVP

the same clusters. On the contrary, the distortion would be also found if the clusters were fewer than fiber types within the slice as the cells would not be separated accurately and then their borders (i.e., membranes) unnoticeable.

Classification of Skeletal Muscle Fiber Types

651

2.3 Image Analysis Finally, for proper objective classification of the fiber types, a histogram-based analysis was made on the clusters to find the mean and the band of the gray levels for each type of fibers as well as the other statistical factors.

3 Results and Discussion 3.1 Slice Preparation The histochemical reaction for ANAE in skeletal muscle fibers produces segregation of muscle fiber types depending on the intensity of the color characteristics of the final reaction product. Different shades of color can be obtained (depending on the muscle fibers and histochemical parameters). The muscle fibers can be easily classified into two or more types as shown in (Fig. 1).

a

b

c

Fig. 1. Color shades of muscle fibers

The results of variation in incubation time are illustrated in serial sections. Same thickness (7µm thickness) of serial sections of muscle fibers underwent histochemical reaction for 15, 30, 45, 60 min and the results are shown in (Fig. 2). 3.2 Image Segmentation This stage includes implementation of fuzzy c-means algorithm on an input biopsy image. This algorithm separates a set of data points into similar groups such that the points that belong to the same group are more similar than the points belonging to a different group. The result of the FCM algorithm is shown in (Figs. 3, 4 and 5), through which, the extracted object is shown black (object) and the other clusters are shown white (background). The algorithm was implemented by MATLAB programming language. The image results of the segmentation have shown clear shape of the cells of each muscle fiber. Excessive number of clusters (i.e., classified objects) would lead to distorted contour of the cell borders as the pixels in the individual cells would likely be classified into different clusters.

652

M. M. Mehdy et al.

a

b

c

d

Fig. 2. Skeletal Muscle Tissue (ANAE) reaction showing four serial sections of muscle fibers with different reaction times a. 15 min b. 30 min c. 45 min d. 60 min

Fig. 3. FCM result on 2-type fibers (a. Type I, b. Type II, c. Endomysium)

3.3 Image Analysis From the results shown, FCM can identify between 2–4 types of skeletal muscle fibers, as well as the endomysium (intercellular tissue). The same algorithm was applied on the slices with different reaction time (shown in Fig. 2), and the histogram feature analysis was as shown in (Tables 3 and 4).

Classification of Skeletal Muscle Fiber Types

653

Fig. 4. FCM result on 3-type fibers (a. Type I, b. Type IIA, c. Type IIB)

Fig. 5. FCM result on 4-type fibers (a. Type I, b. Type IIA, c. Type IIB, d. Type IIC)

The histogram features showed the complex nature of the input images. While the unsupervised segmentation manages to classify the fiber types with mostly clear cell borders, there are few uncertain portions of the pixels that require more effective tools as well as more data input to increase the efficiency of the algorithm.

4 Conclusions Unsupervised classification techniques based on fuzzy c-means clustering algorithm using ANAE in staining have proved successful in segmenting of biopsy image of human

654

M. M. Mehdy et al.

Table 3. Gray levels and percentage of four muscle tissue images with four different reaction times Reaction Ttime

Type I

Type II

Gray levels

Endomysieum

Percentage (%)

Gray levels

Percentage (%)

Gray levels

Percentage (%)

15 min

99–148

26.1

149–185

55.2

186–220

18.7

30 min

117–146

25.6

147–172

28.4

173–218

46

45 min

89–150

20.6

151–180

41

181–226

38.4

60 min

106–151

22.4

152–179

55.8

180–229

21.8

Table 4. Histogram feature analysis of four muscle tissue images with four different reaction times Reaction time 15 min

30 min

45 min

60 min

Type I

Type II

Endomysieum

Mean

128

168

202

Standard Deviation

12.674

9.5134

12.419

Skewness (*103 )

2.6266

1.1928

2.8647

Flat (*104 )

4.6654

1.7176

5.7644

Mean

133

160

184

Standard Deviation

9.0355

10.077

8.2558

Skewness (*103 )

1.1032

1.4359

0.9509

Flatness (*104 )

1.6220

2.2136

1.6389

Mean

130

170

190

Standard Deviation

12.689

8.3404

8.0086

Skewness (*103 )

3.0001

0.8309

0.8551

Flatness (*104 )

6.4124

1.1287

1.4606

Mean

135

166

192

Standard Deviation

9.7876

8.7811

18.008

Skewness (*103 )

1.2915

0.9351

7.1181

Flatness (*104 )

1.9420

1.2346

16.277

skeletal muscle tissue into different fiber types. Change in the reaction time does not affect the results of the images, so the minimum reaction time (which is 15 min) is the suitable one for this work.

Classification of Skeletal Muscle Fiber Types

655

5 Recommendations In case of big data availability, artificial neural networks (ANN) are recommended in the learning process of the algorithm. Cooperating of microscopic and ultrasound images in neuromuscular diseases is recommended to visualize the progress in the treatment procedure of the muscle fibers from both anatomical and physiological aspects. Acknowledgment. The study was done in both Al-Nahrain University and Dijlah University College.

References 1. Kugelberg, E., Edström, L.: Differential histochemical effects of muscle contractions on phosphorylase and glycogen in various types of fibres: relation to fatigue. J. Neurol. Neurosurg. Psychiatry 31(5), 415–423 (1968) 2. Bottinelli, R., Reggiani, C.: Human skeletal muscle fibres: molecular and functional diversity. Prog. Biophys. Mol. Biol. 73(2–4), 195–262 (2000) 3. Scott, W., Stevens, J., Binder-Macleod, S.A.: Human skeletal muscle fiber type classifications. Phys. Ther. 81(11), 1810–1816 (2001) 4. Sjostrom, B.E.M., Friden, J.: Fine structural details of human muscle fibres after fibre type specific glycogen depletion. Histochemistry, 425–438 (1982) 5. Wilson, J.M., Loenneke, J.P., Jo, E., Wilson, G.J., Zourdos, M.C., Kim, J.S.: The effects of endurance, strength, and power training on muscle fiber type shifting. J. Strength Cond. Res. 26(6), 1724–1729 (2012) 6. Wang, Y., Winters, J., Subramaniam, S.: Functional classification of skeletal muscle networks. I. Normal physiology. J. Appl. Physiol. 113(12), 1884–1901 (2012) 7. Kinugawa, S., Tsutsui, H.: Skeletal muscle abnormalities in heart failure. Int. Heart J. 56(5), 475–484 (2015) 8. Joyce, N.C., Oskarsson, B., Jin, L.W.: Muscle biopsy evaluation in neuromuscular disorders. Phys. Med. Rehabil. Clin. N. Am. 23(3), 609–631 (2012) 9. Aydin, M.F., Celik, I., Sur, E., Oznurlu, Y., Telatar, T.: Enzyme histochemistry of the peripheral blood lymphocytes in arabian horses. J. Anim. Vet. Adv. 9(5), 920–924 (2010) 10. Cai, C.: Nonspecific esterase. PathologyOutlines.com website. https://www.pathologyout lines.com/topic/stainsnonspecificesterase.html. Accessed 16 June 2022 11. Mohammed, R., Ajwad, A.: CT image segmentation based on clustering methods. J. Fac. Med. Baghdad 52(2), 232–236 (2010) 12. Pednekar, A., Kakadiaris, I.A., Kurkure, U.: Adaptive fuzzy connectedness-based medical image segmentation. Indian Conf. Comput. Vision, Graph. Image Process. (2002) 13. Bensaid, A.M., Hall, L.O., Clarke, L.P., Velthuizen, R.P.: MRI segmentation using supervised and unsupervised methods. Proc. Annu. Conf. Eng. Med. Biol. 13(pt 1), 60–61 (1991) 14. Khamiss, N.N.: Unsupervised segmentation method for brain MRI based on fuzzy techniques. Nahrain Univ. Coll. Eng. J. 13(1), 108–115 (2010) 15. Hesamian, M.H., Jia, W., He, X., Kennedy, P.: Deep learning techniques for medical image segmentation: achievements and challenges. J. Digit. Imaging 32(4), 582–596 (2019). https:// doi.org/10.1007/s10278-019-00227-x 16. Yoon, C.H., Bones, P.J., Millane, R.P.: Image analysis for electron microscopy of muscle fibres. Proc. Digit. Imaging Comput. Tech. Appl. DICTA 2005, vol. 2005, no. Dicta, pp. 551– 557 (2005)

656

M. M. Mehdy et al.

17. Papastergiou, P.T.A., Hatzigaidas, A., Cheva, A.: A sophisticated edge detection method for muscle biopsy image analysis. Proc. 7th WSEAS International Conference Signal, Speech Image Process, pp. 118–123 (2007) 18. Liu, F., Mackey, A.L., Srikuea, R., Esser, K.A., Yang, L.: Automated image segmentation of haematoxylin and eosin stained skeletal muscle cross-sections. J. Microsc. 252(3), 275–285 (2013) 19. Sertel, O., Dogdas, B., Chiu, C.S., Gurcan, M.N.: Muscle histology image analysis for sarcopenia : registration of successive sections with distinct atpase activity * Dept. of electrical and computer engineering, The Ohio State University, Columbus, OH ( USA ) Dept . of Biomedical Informatics, The O. Science (80):1423–1426 (2010) 20. Rahmati, M., Rashno, A.: Automated image segmentation method to analyse skeletal muscle cross section in exercise-induced regenerating myofibers. Sci. Rep. 11(1) (2021) 21. Vu, Q.D. et al.: Methods for segmentation and classification of digital microscopy tissue images. Front. Bioeng. Biotechnol. 7 (2019) 22. Achouri, A., Melizi, M., Belbedj, H., Azizi, A.: Comparative study of histological and histochemical image processing in muscle fiber sections of broiler chicken. J. Appl. Poult. Res. 30(3), 100173 (2021) 23. Janssens, T., Antanas, L., Derde, S., Vanhorebeek, I., Van den Berghe, G., Güiza Grandas, F.: Charisma: an integrated approach to automatic H&E-stained skeletal muscle cell segmentation using supervised learning and novel robust clump splitting. Med. Image Anal. 17(8), 1206– 1219 (2013) 24. Nguyen, B.P., Heemskerk, H., So, P.T.C., Tucker-Kellogg, L.: Superpixel-based segmentation of muscle fibers in multi-channel microscopy. BMC Syst. Biol. 10(Suppl 5) (2016) 25. Cui, L., Feng, J., Zhang, Z., Yang, L.: High throughput automatic muscle image segmentation using parallel framework. BMC Bioinformatics 20(1), 1–9 (2019) 26. Cui, L., Feng, J., Yang, L.: Towards fine whole-slide skeletal muscle image segmentation through deep hierarchically connected networks. J. Healthc. Eng. 2019 (2019)

Modeling the Intention to Use AI Healthcare Chabot’s in the Indian Context Aishwarya Nagarathinam(B) , Aarthy Chellasamy, N. Elangovan, and Sangeetha Rengasamy School of Business and Management, Christ University, Bengaluru, India [email protected]

Abstract. Covid-19 has accelerated the need and use of artificial Intelligencebased healthcare Chabots. Penetration of the internet, smartphone, computational capability and machine learning technology brings healthcare services close to the patients. The penetration of AI healthcare Chatbot technology worldwide is on the rise. However, the healthcare ecosystem in India is unique and poses challenges in the adoption of healthcare chatbots. The demographic characteristics, economic conditions, diversity, belief systems on health-seeking, and alternative medical practices play a role in accepting and using chatbots. In this study, we attempt to model the factors influencing the intention and the purpose of using the chatbot. Through a literature review, we identify the variables related to the adoption of healthcare chatbots. We then focus on the more relevant concepts to the Indian context and develop a conceptual model. Through cases and literature, we frame the propositions of the study. We look at the awareness of chatbot features, perception towards the chatbot, trust and mistrust of the healthcare system, the doctors and the chatbots, health-seeking behavior, and the belief in traditional, complementary, and alternative medicine prevalent in India. This study contributes by developing an initial conceptual model for healthcare chatbots adoption in the Indian context. In the future, we plan to operationalize the study and test the propositions through an elaborate survey to validate the model empirically. Keywords: AI Healthcare chatbots · Intention to use · Indian Context · Conceptual model

1 Introduction Covid-19 has virtually disrupted all sectors, but the digital health sector has seen a boost amid the situation. When the entire health care system was dedicated to the covid treatment, the regular healthcare requirements were shifted to the digital mode. Digital technology in healthcare encompasses telemedicine and telehealth. The growth of big data and Artificial Intelligence (AI) usage in healthcare has enhanced the decisionmaking and analytical processes, a strong pillar of support in digital healthcare. As a part of AI, Chatbots act as virtual conversational agents and mimic human behavior by providing health advice, giving remainders, booking appointments, and monitoring [1]. The adoption of healthcare chatbots is not the same across the globe. In countries © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 657–666, 2023. https://doi.org/10.1007/978-3-031-20429-6_59

658

A. Nagarathinam et al.

like India, the digital and economic divide between rural and urban makes healthcare challenging in terms of affordability, quality, and inequity. On the one hand, India is home to some of the best medical hospitals in the world and on the other hand, there is an acute shortage of doctors (1: 1,596) [2]. AI for healthcare will be a boon for the underserved rural population, which lags in infrastructure, physicians, and financial means to access medical facilities [3]. However, the adoption of digital healthcare depends on cultural and social attitudes. India is known for its food, ayurvedic, yogic practices, and traditional home remedies being deep-rooted among the population may also be why people still don’t rely on medical practitioners and use chatbots in such cases. A recent study stated that rural women seek out informal healers over formal health care providers because of ease of communication, cultural familiarity, social stigma, and geographical distance [4]. More than 80% of people cannot afford the most basic medical treatments, medications, or vaccines in impoverished countries. Hence, Complementary and alternative medicines (CAM) are popular among affluent populations in both developed and developing countries, despite limited evidence of their safety and effectiveness. Approximately 60% of the world’s population uses alternative medications. The rural masses of underdeveloped nations utilize these medicines for primary health care and even in developed countries where modern medicines predominate [5]. Ayurveda, Siddha, and Unani are three historic Indian medicinal systems described in some of the ancient Vedas and other scriptures. Around 70% of India’s rural population relies on the ancient Ayurvedic medical system [6]. It’s assumed that those who adopt these methods consider them less expensive than western medical approaches. Also, confidence in the treatment, ease of access, fewer side effects, and choice of a traditional healer are other reasons CAM is popular. Although chatbots have been around for a while in the field of information technology, their Healthcare applications are relatively new. A lot of interest is shown in AI assisted Healthcare and their implications to consumers post COVID-19. However, chatbot based studies on Indian context and the factors determining them are very minimal and in light of this, the current study proposes a conceptual framework in Indian context for determining whether (1) consumers’ attitudes toward chatbots influence their intention to use this technology; (2) consumer trust and mistrust of healthcare systems; and (3) consumer health seeking behaviour explicitly deter the intention to use chatbots. The variables connected to the adoption of healthcare chatbots are identified using a literature review.

2 Theoretical Background As the study focuses on adoption and intention to use the AI Healthcare Chatbots, popular theories of adoption and the human-machine interaction are referred. Following are the theories related to technology adoption. 2.1 TAM Model One of the most important technology acceptance models is the Technology Acceptance Model [7], which states that two key elements influence an individual’s intention to utilize

Modeling the Intention to Use AI Healthcare Chabot’s

659

new technology: perceived ease of use and usefulness. Recent studies have identified various factors which determine the intention to use the technology or not, such as social influence, culture, and the like. Another well-known theory from social psychology inspired the TAM, the theory of reasoned action [8], which explains a person’s activities through their intentions [9]. The intention is explained by individual attitudes and social norms [7]. Innovation adoption is a “the generation, development and adaptation of novel ideas on the part of firm” [10] which acts as a response to the environments or controls the environment. Diffusion of Innovation theory (DoI) proposed in 1962, explains how a product or idea gains momentum and diffuses over a specific social system [11]. Another well-known theory is Theory of Reasoned Action (TRA) [8]. This theory states that an important construct of individual behaviour is the intention to accomplish that behaviour which is a function of subjective norm and attitude towards behaviour [12]. Interaction with a computer was mentioned in the science fiction category in 1945. However, it has become a reality with the introduction of conversational agents [13]. Conversational agents include service robots that can provide service interactions through voice or text-based [14]. The following Universal theories investigate the aspects of human-machine interaction. 2.2 Anthropomorphism Theory Anthropomorphism theory can be used to understand how different levels of robot humanness and social interaction opportunities affect consumers’ liking for service robots/conversational agents/Chatbots [15]. Anthropomorphism is more specifically attributing human mental states or effects to non-humans. E.g. sometimes, while observing the moving clouds, most individuals relate them to an object, mountain, animal, or person [16]. As per this theory, humans perceive there are possibilities as if they are consulting with another human. 2.3 Social Presence Theory Social Presence Theory (SPT), is defined as “the degree of salience of the other person in the interaction and the consequent salience of the interpersonal relationships”. However, it was improvised believing that interpersonal emotional connection is a prerequisite for interaction on the one hand and perceiving that a person is present on the other end [17]. Health seeking behavior theories are unique for Indian population as the country has unique, recognized medicinal systems.

3 Methodology This study uses a qualitative method for building the conceptual framework for adopting healthcare chatbots in the Indian context. This paper’s methodology was guided by using past studies. [18, 19] “Conceptual frameworks seek to identify ‘presumed relationships’ among key factors or constructs to be studied, and that the justification for these

660

A. Nagarathinam et al.

presumptions may come from multiple sources such as one’s own prior research or ‘tentative theories’ as well as established theoretical or empirical work found in the research literature [20]”. We start by reviewing the theoretical background related to Diffusion of innovation, Technology adoption, and Intention to use. Various concepts are extracted from the literature under different theories and are reflected in the Indian context. The most relevant and important concepts unique to the Indian context are selected from the list of identified concepts. After the logical arrangement of the concepts on a temporal basis, the relationship between the concepts is proposed in propositions. Further, justification for the propositions is presented with the support of literature.

4 Developing Conceptual Framework 4.1 Awareness and Perception of Chatbots Chatbots are artificial intelligence agents with whom users can communicate via natural language dialogue, text, or voice. Chatbots can be categorized based on purpose (assistant or conversation) and communication (text-based or voice-based). Originally, chatbots were created for fun and entertainment purposes and used simple keyword matching techniques [21]. Later, chatbots started using natural language processing systems to analyze users’ inputs. Chatbots who comprehend and employ human humor are seen as more personable, capable, and cooperative and are thought to perform better than those that don’t [22]. Chatbots are getting popular among tech-savvy consumers in all the fields of business. Thanks to personal virtual assistants like Siri, Alexa, and Cortana, there is much awareness and attention about these conversational agents. 4.2 Trust & Mistrust in Chatbots The user’s trust in chatbots is determined by factors specific to the chatbot, such as how much it responds like a human, how it presents itself, and how competent it appears. However, it also depends on factors specific to its service contexts, such as the chatbot host’s brand, privacy, security, and other risk issues related to the request’s topic [23]. However, not many users trust chatbot usage since human–chatbot communication varies based on disclosing who their conversational partner is. Also, Chatbots still lack empathy when it comes to understanding meaning, and they aren’t as good at interpreting conversational undertones as humans are, which might bring mistrust to AI chatbots [24]. 4.3 Trust & Mistrust in Doctors and the Healthcare System Trust plays a major role in all medical relationships and is an important contributor to the best health outcomes for patients. Patients frequently disclose sensitive personal information to their doctors to expect it would be kept private. Clinicians strive to earn patients’ trust to accept their diagnosis and suggested treatments, including invasive procedures and long-term pharmaceutical use for chronic diseases [25]. To gain this level of confidence, practitioners and healthcare institutions must first persuade people

Modeling the Intention to Use AI Healthcare Chabot’s

661

that they prioritize patients’ needs over their own financial or nonfinancial self-interest. However, people’s trust in health care has eroded over the last half-century. Patient distrust is linked to fewer doctor-patient interactions, poor clinical relationships with less continuity, worse adherence to advice, poor self-reported health, and lower healthcare utilization. Healthcare inequalities are complex and have been the topic of extensive investigation. Socioeconomic differences, structural factors (such as a lack of access to care), conflict of interest between clinicians and pharma manufacturers, accompanying financial incentives, and consumerism are significant drivers for declining trust in the health care system [26]. 4.4 Health Seeking Behavior and CAM The terms “complementary medicine” and “alternative medicine” refer to a wide range of healthcare methods that are not part of a country’s traditional or conventional medicine and are not fully incorporated into the dominant healthcare system. People who use complementary and alternative medicine (CAM) are usually looking for solutions to improve their health and well-being or alleviate symptoms associated with chronic, even terminal illnesses or the side effects of traditional treatments [27]. Having a holistic health philosophy or a transformational experience that affects one’s worldview and wanting better control over one’s health are further reasons for using CAM. 4.5 Intention to Use Chatbots Using the Technology Acceptance Model and Diffusion of Innovations Theory, various studies have identified factors for the intention of consumers to use chatbots. In addition, it was found that the intention to use chatbots stems from chatbot performance expectancy, social influence, hedonism, habit, anthropomorphism, and perceived innovativeness [28]. These factors positively affect chatbot usage intentions, while effort expectancy, inconveniences, and automation have negative effects. A study identified that reliability of the information provided, empathy in interaction, and tangibility have important roles in the intention of an individual to use chatbots [29]. It was found that perceived utility, ease of use, enjoyment, price consciousness, perceived risk, and personal innovativeness significantly impacted attitudes about chatbots [30]. However, trust, personal innovativeness, and attitude were the only factors that directly predicted intention to use. Age, gender, and prior experience with technology applications have a moderating effect on the intention to use chatbots.

5 Developing the Propositions Intelligence and emotions are the two factors that differentiate humans and machines, and the evolution of AI frightens and fascinates by mimicking human cognitive behavior. Trust is an important asset in the digital world, and this became evident during the pandemic when the awareness of the usage of health chatbots among people was on the rise [31]. Chatbots simulate the feeling of human presence and are more likely to be trusted [32]. Add-on features like 24*7 availability, human connection, scheduling

662

A. Nagarathinam et al.

appointments, locating clinics, and refilling medicines can encourage patients’ intention to use chatbots. The other side of the research says that chatbots are not mature enough to handle situations like medical professionals. Also, there are chances of a data breach [31], leading to mistrust of chatbots. Hence, we propose the proposition P1 as follows: P1: Trust/Mistrust on Chatbot mediates the impact of awareness on the intention to use chatbots. The potential use of Artificial Intelligence has improved the efficiency and effectiveness of healthcare. However, the perception of AI-based chatbot acceptance is still in its infancy. Past researchers have stated that the use of chatbots has a positive perception because of their responsiveness, availability, ease of use, improved efficiency, and reduced cost of health care [32, 33]. However, concerns like lack of trust, data privacy, patient safety, technological maturity, and automation might lead to negative perceptions among patients in the intention to use chatbots. Personal recommendation, linguistic ability, variability in responses, and customization may enhance the perception of healthcare chatbots. In this view, the authors put forward the proposition P2 as follows: P2: Trust/Mistrust on Chatbot mediates the impact of perception on the intention to use chatbots. Physicians presume that chatbots would reduce administrative tasks, redundant jobs, quick access to information, motivate patients, and surrogate patients. Physicians also believe that chatbots can’t give personal medications and comprehend human emotions, and algorithms can’t replace human intelligence and expertise [31]. Patients may trust and opine that initial screening by a chatbot from a physician may be a time-saving factor where doctors can spend quality time for later diagnosis with patients, which intends patients to use chatbots often. In this background, the authors propose: P3: Trust/Mistrust of Doctors moderates the impact of awareness and perception on the intention to use chatbots Irrespective of the healthcare system, whether ayurvedic, Unani, traditional, or complementary medicines, trust plays an important role as every individual’s health becomes a priority. Patients tend to lose trust in a particular health care system if the result is not intended. It might sometimes be a cultural influence in a country like India, which provokes them to shift to new forms of medicines where the chatbot platform might be fascinating. Having said these, the authors propose P4 as: P4: Trust/Mistrust in the healthcare system moderates the impact of awareness and perception on the intention to use chatbots Health-seeking behavior can be related to a country’s health status and economic development [34]. Health-seeking behavior can also be influenced by community norms, individual or household behavior, and varied expectations and characteristics of health care providers [35]. Women’s health is particularly concerned because of deep-rooted socio-cultural discrimination, which may inhibit them from availing best health care services [36]. The availability of healthcare chatbots may create awareness among these populations who may intend to use chatbots, and thus, the authors propose P5 as:

Modeling the Intention to Use AI Healthcare Chabot’s

663

P5: Health seeking behavior moderates the impact of awareness and perception on the intention to use chatbots India is a country known for its heritage and traditional practice. Apart from allopathic medicine, India has unique, recognized medicinal systems like Ayurveda, Yoga & Naturopathy, Unani, Siddha, and Homoeopathy, collectively known as AYUSH. Though varied medicinal systems are present, the acceptance of each system differs from rural to urban, and the demographic profile, disease to be treated, and perception and intention of individuals towards alternative medicinal systems [36]. Consumers in the current decade have become more proactive and tech-savvy toward self-care, which is evident with the rise of Fitness applications, Smartwatches, and apps such as step counter, HealthifyMe, and Zero calories. Before a health-related problem, healthcare customers constantly monitor their vitals to take preventive actions. This nature of health-seeking behavior persuades customers to try out the new avenue of healthcare, such as chatbots, to gain awareness and use it to clarify their doubts about their symptoms and general concern (Fig. 1). P6: Belief in traditional, complementary, and alternative medicine moderates the impact of awareness and perception on the intention to use chatbot

Fig. 1. Conceptual model of intention to use Chatbot (Indian Context). Source Author’s own

6 Conclusion The use of health chatbots among patients is growing at a faster rate. India is a large country with great potential and benefits from healthcare chatbots. However, unique

664

A. Nagarathinam et al.

challenges will impact the intention to use the healthcare chatbots by going through various literature and analyzing the theoretical frameworks of technology adoption. A conceptual model relating to the factors is presented in this study and is supported by the literature cases. The study proposes that the health-seeking behavior and the belief in traditional, complementary, and alternative medicine are factors unique to the Indian context that can dampen the use of chatbots. In contrast, the trust and mistrust in the healthcare system and the doctors reflect the Indian healthcare ecosystem and can promote the use of healthcare chatbots. Future studies can validate the propositions through empirical data from Indian patients.

References 1. Nadarzynski, T., Miles, O., Cowie, A., Ridge, D.: Acceptability of artificial intelligence (AI)-led chatbot services in healthcare: a mixed-methods study. Digital Health 5, 2055207619871808 (2019) 2. Central Bureau of Health Intelligence: ‘National Health Profile 2018’, Ministry of Health and Family Welfare (2018), http://www.cbhidghs.nic.in/WriteReadData/l892s/Before%20C hapter1.pdf. Accessed 19 March 2022 3. Paul, Y., Hickok, E., Sinha, A., Tiwari, U.: Artificial Intelligence in the healthcare industry in India, The Centre for Internet and Society (2018), https://cis-india.org/internet-governance/ files/ai-and-healtchare-report. Accessed 19 March 2022 4. Das, M., Angeli, F., Krumeich, A.J.S.M., van Schayck, O.C.P.: The gendered experience with respect to health-seeking behaviour in an urban slum of Kolkata, India. Int. J. Equity Health 17(1), 24 (2018). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5813424/#__ffn_ sectitle (accessed 19 Mar. 2022) 5. Ballabh, B., Chaurasia, O.P.: Traditional medicinal plants of cold desert Ladakh-Used in treatment of cold, cough and fever. J. Ethnopharmacol. 112(2), 341–349 (2007) 6. Pandey, S.R., Rawat, A.K.S.: Indian traditional ayurvedic system of medicine and nutritional supplementation. Evide. Based Compl. Alternat. Med. 2013, Article ID 376327, 12 p (2013). https://doi.org/10.1155/2013/376327 7. Davis, F.D.: Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Q. 13(3), 319–340 (1989) 8. Fishbein, M., Ajzen, I.: Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research. Addison-Wesley, Reading, MA (1975) 9. Rauniar, R., Rawski, G., Crumbly, G., Simms, J.: C2C online auction website performance: buyer’s perspective. J. Electron. Res. 10(2), 56–75 (2009) 10. Omar, A., Soar, J.: Technology Innovation Adoption Theories. Handbook of Research on Driving Competitive Advantage through Sustainable, Lean, and Disruptive Innovation. IGI Publishers (2016). https://doi.org/10.4018/978-1-5225-0135-0.ch001 11. Miller, R.L.: Rogers’ innovation diffusion theory (1962, 1995). In: Information Seeking Behavior and Technology Adoption: Theories and Trends, pp. 261–274. IGI Global (2015) 12. A. Khan, R., Qudrat-Ullah, H.: Technology Adoption Theories and Models. In: Adoption of LMS in Higher Educational Institutions of the Middle East. ASTI, pp. 27–48. Springer, Cham (2021). https://doi.org/10.1007/978-3-030-50112-9_5 13. Schuetzler, R.M., Grimes, G.M., Scott Giboney, J.: The impact of chatbot conversational skill on engagement and perceived humanness. J. Manag. Inf. Syst. 37(3), 875–900 (2020) 14. Blut, M., Wang, C., Wünderlich, N.V., Brock, C.: Understanding anthropomorphism in service provision: a meta-analysis of physical robots, chatbots, and other AI. J. Acad. Mark. Sci. 49(4), 632–658 (2021). https://doi.org/10.1007/s11747-020-00762-y

Modeling the Intention to Use AI Healthcare Chabot’s

665

15. Letheren, K., Jetten, J., Roberts, J., Donovan, J.: Robots should be seen and not heard…sometimes: Anthropomorphism and AI service robot interactions. Psychol. Mark. 38(12), 2393–2406 (2021) 16. Airenti, G.: The development of anthropomorphism in interaction: intersubjectivity, imagination, and theory of mind. Front. Psychol. 9, 2136 (2018) 17. Lowenthal, P.R.: The evolution and influence of social presence theory on online learning. In: Social Computing, pp. 113–128 (2010). https://doi.org/10.4018/978-1-60566-984-7.ch010 18. Miles, M.B., Huberman, A.M., Saldaña, J.: Qualitative Data Analysis: A Methods Sourcebook, 3rd edn. SAGE Publications, Thousand Oaks, CA (2014) 19. Maxwell, J.A.: Qualitative Research Design: An Interactive Approach, 3rd edn. SAGE Publications, Thousand Oaks, CA (2013) 20. Ravitch, S.M., Riggan, M.: Reason & Rigor: How Conceptual Frameworks Guide Research. Sage Publications (2016) 21. Adamopoulou, E., Moussiades, L.: Chatbots: History, technology, and applications. Mach. Learn. Appl. 2, 100006 (2020) 22. Fernandes, A.: NLP, NLU, NLG and how Chatbots work. Chatbots Life (2017). https://cha tbotslife.com/nlp-nlu-nlg-and-how-chatbots-work-dd7861dfc9df 23. Baker, D.W.: Trust in health care in the time of COVID-19. JAMA 324(23), 2373–2375 (2020). https://doi.org/10.1001/jama.2020.23343 24. Khullar, D.: Building trust in health care: why, where, and how. JAMA 322(6), 507–509. https://doi.org/10.1001/jama.2019.4892 25. Barnes, P.M., Bloom, B., Nahin, R.L.: Complementary and alternative medicine use among adults and children: United States, 2007. National health statistics reports; no 12. Hyattsville, MD: National Center for Health Statistics (2008) 26. Malik, Dr., Gautam, S., Srivastava, S.: A study on behaviour intention for using Chatbots, 332–338 (2020). https://doi.org/10.1109/ICRITO48877.2020.9197782 27. Kasilingam, D.L.: Understanding the attitude and intention to use smartphone chatbots for shopping. Technol. Soc. 62, 101280 (2020), ISSN: 0160-791X, https://doi.org/10.1016/j.tec hsoc.2020.101280 28. Almalki, M., Azeez, F.: Health Chatbots for fighting COVID-19: a scoping review. Acta Informatica Medica: AIM: J. Soc. Med. Inf. Bosnia Herzegovina: Casopis Drustva Za Medicinsku Informatiku BiH 28(4), 241–247 (2020) 29. Ogonowski, A., Montandon, A., Botha, E., Reyneke, M.: Should new online stores invest in social presence elements? The effect of social presence on initial trust formation. J. Retailing Consumer Serv. 21(4), 482–491 (2014). https://doi.org/10.1016/j.jretconser.2014.03.004 30. Palanica, A., Flaschner, P., Thommandram, A., Li, M., Fossat, Y.: Physicians’ perceptions of Chatbots in health care: cross-sectional web-based survey. J. Med. Int. Res. 21(4), e12887 (2019). https://doi.org/10.2196/12887 31. Lai, Y., Lioliou, E., Panagiotopoulos, P.: Understanding users’ switching intention to aipowered healthcare chatbots. ECIS 2021 Research Papers 51 (2021) 32. Chew, H.S.J., Achananuparp, P.: Perceptions and needs of artificial intelligence in health care to increase adoption: scoping review. J. Med. Internet Res. 24(1), e32939 (2022) 33. Latunji, O.O., Akinyemi, O.O.: Factors influencing health-seeking behaviour among civil servants in Ibadan. NIGERIA. Ann. Ibadan Postgraduate Med. 16(1), 52–60 (2018) 34. Oberoi, S., Chaudhary, N., Patnaik, S., Singh, A.: Understanding health seeking behavior. J. Family Med. Primary Care 5(2), 463–464 (2016)

666

A. Nagarathinam et al.

35. Reddy, P.M.C., Rineetha, T., Sreeharshika, D., Jothula, K.Y.: Health care seeking behaviour among rural women in Telangana: a cross sectional study. J. Family Med. Primary Care 9(9), 4778–4783 (2020) 36. Kumar, H., Kapinakadu, S., Anil, M.: Health seeking behaviour and its determinants among rural population- a cross sectional study in South India. Int. J. Commun. Med. Public Health 6(11), 4944 (2019)

Exploring the Technology Acceptance of Wearable Medical Devices Among the Younger Generation in Malaysia: The Role of Cognitive and Social Factors Way Zhe Yap1(B) , Bee Chuan Sia1

, Hong Lip Goh1

, and Tat Huei Cham2

1 Faculty of Accountancy and Management, Universiti Tunku Abdul Rahman, Kuala Lumpur,

Malaysia [email protected], {siabc,gohhl}@utar.edu.my 2 UCSI Graduate Business School, UCSI University, Kuala Lumpur, Malaysia

Abstract. The technological advancements in the contemporary world have made the offsite monitoring of patients possible, allowing individuals to keep track of their health by using wearable medical devices (WMDs). The ability of such devices to monitor and communicate health-related information between patients and caregivers can potentially help in combatting non-communicable diseases. Such devices are expected to grow exponentially in demand with the rapid proliferation of ICT and empowerment by both knowledgeable consumers and wider penetration of IoT, catalyzed by the younger generation. Hence, this quantitative research study was conducted in the Klang Valley (consisting of Kuala Lumpur and Selangor states) to explore and compare the factors influencing the intention to use WMDs among the younger generation in Malaysia by extending the technology acceptance model with word of mouth and electronic word of mouth. The results show that perceived usefulness, word of mouth, and electronic word of mouth are the significant factors influencing the intention to use WMDs among the younger generation. Furthermore, Gen Y and Gen Z have different factors influencing their intention to use WMDs. Keywords: Technology Acceptance Model · Wearable Medical Devices (WMD) · Younger Generation · Generation Y · Generation Z

1 Introduction The integration of digital technologies into medical applications has enabled Healthcare 4.0, where medical practices are no longer bound to clinical constraints and real-time personal monitoring of one’s medical condition is made possible [1]. Advancements in healthcare technology have also enabled wearable medical devices (WMDs) to be perceived as the fastest growing technology in the healthcare industry [2]. The global WMD market is expected to grow at a compound annual growth rate of 26.8% from 2021 to 2028 and reach a market value of USD 111.9 billion. Emerging markets, such as © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 667–679, 2023. https://doi.org/10.1007/978-3-031-20429-6_60

668

W. Z. Yap et al.

Malaysia, with a high acceptance rate of general wearable devices, have also provided a great market opportunity for WMD producers to expand their businesses. The smartwatch is a perfect example of WMD because: (1) most people are familiar with it; (2) it is available as a consumer product; (3) it allows real-time continuous monitoring of health and physical data (e.g., heart rate, body temperature, blood pressure, and respiration rate, as well as several life-threatening situations, such as worsening disease conditions, falls, and stroke [5–7]; (4) it enables patients, families, and physician to communicate; (5) it is equipped with tailored messages and reminders; and (6) it is able for in situ, mini-survey, and behavior verification based on the measurement of its sensors [6]. The ability to monitor and communicate health-related information is ideal for close monitoring of non-communicable diseases (NCDs), chronic obstructive pulmonary disease, diabetes, and other cardiovascular diseases [2, 9–13]. Driven by the potential benefits of WMDs, this study explores the acceptance of WMD in Malaysia, more specifically among the younger generation (Gen Y and Gen Z).

2 Problem Statement The rise of NCD among the younger generation of Malaysians has become a cause for major concern [13], partially due to the low frequency of medical check-ups. WMDs, which have garnered prominence recently, might be able to fill the gap and prevent the contraction of NCD. Therefore, the study on the demand of the younger Malaysian generations in adopting such technologies is crucial and beneficial to the nation’s healthcare. A vast plethora of literature focuses on the factors influencing the potential usage of WMD [5, 15–21]. Besides the study from Europe [15], studies on factors influencing the intention to use WMD (IU-WMD), particularly among the younger generations are lacking. Hence, this study is intended to fill the literature gap by investigating the factors influencing the IU WMD among the younger generation, specifically where the technology acceptance model (TAM) is adopted via IU-WMD among the younger generation. The moderating effects of Gen Y and Gen Z were used to examine the robustness of this study as well.

3 Conceptual Framework and Hypotheses Development Based on the theoretical foundation of TAM [21], this study has used perceived usefulness (PU), perceived ease of use (PEoU), perceived convenience (PC), word of mouth (WOM), and electronic word of mouth (EWOM) based on their origin and their relationship with IU WMD among the younger generation, as well as the hypotheses development of this study. Figure 1 shows the research model. 3.1 Perceived Usefulness and Intention to Use “Perceived usefulness” in this study is defined as the degree to which an individual believes that using WMD will enhance his or her performance in healthcare [21]. The

Exploring the Technology Acceptance

669

impact of PU on IU-WMD had been empirically proven in several previous studies [3, 14–16, 22–26]. As WMDs have been designed to help the users to improve and monitor their health conditions, this study suggests that when an individual believed that WMD is useful for their health, they will be more willing to adopt it. Along these lines, the first hypothesis was developed as follows: H1: Perceived usefulness positively influences the intention to use wearable medical devices 3.1.1 Perceived Ease of Use and Intention to Use “Perceived ease of use” in this study is defined as the extent to which an individual believes that WMDs can be used effortlessly to monitor their health [21]. The impact of PEoU on IU-WMD had been empirically tested in various studies [5, 15, 17, 23]. Therefore, this study suggests that when an individual believes that WMD is easy to use, they will be more willing to use the same. Hence, the second hypothesis was developed as follows: H2: Perceived ease of use positively influences the intention to use wearable medical devices. 3.1.2 Perceived Convenience and Intention to Use In this study, “perceived convenience” is defined as the degree to which an individual believes that WMD has provided them with the convenience to monitor their health with regard to time, place, and execution [25, 26]. As WMD is highly portable, this study suggests that when an individual believed that WMD has provided them with the convenience to monitor their health, they will be more willing to use WMDs. Hence, the third hypothesis was developed as follows: H3: Perceived convenience positively influences the intention to use wearable medical devices 3.1.3 Social Factors and Intention to Use The impact of social influence on the IU of new technology has been reported in a few studies [27, 28]. Social influence can be divided into WOM and EWOM, thanks to the ubiquity of the internet, where private communication independent of commercial may influence IU [29]. The information shared through the internet about WMDs such as reviews or testimonials had already been empirically tested in the study of the acceptance of WMDs [16, 21, 34]. This study suggests that the opinion from others either or both online and real time will positively influence the IU-WMD of an individual. Hence, the fourth and fifth hypotheses were developed as follows: H4: Word of mouth positively influences the intention to use wearable medical devices. H5: Electronic word of mouth positively influences the intention to use wearable medical devices

670

W. Z. Yap et al.

3.1.4 Generational Cohort Generation segmentation is a useful technique for market segmentation because individuals born into a different generation tend to have different experience with technology [36], which could affect their attitude, perception, and preferences toward any new technology. Hence, the sixth hypothesis was developed as follows: H6: Gen Y and Gen Z have a moderating effect on the acceptance of WMD.

Fig. 1. Research model

4 Research Methodology A quantitative survey was conducted to test the six hypotheses of this study. The measurement items used in the survey were adapted from previous studies; PU and PEoU were adapted from Refs. [14, 17]; PC was adapted from Ref. [17]; WOM was developed by adapting the measurement items of Ref. [37], and EWOM was developed by adapting the measurement items of “User Generated Content” of Ref. [37]. The measurement items of IU were essentially adapted from multiple studies. All the measurement items of the survey were measured with a 5-point Likert scale, where 1 corresponds to strongly disagree, while 5 corresponds to strongly agree. As Malaysia is a multicultural country, the survey instruments were also translated into the Malay language in order to collect the responses from the people who may not be so well versed in English. The translated version of the survey was proofread by native Malay speakers.

Exploring the Technology Acceptance

671

4.1 Participants and Data Collection This study used the G*Power software based on [38] to calculate the targeted sample. Specifically, this study used the F-test in G*Power with an effect size of 0.15, margin of error of 5%, confidence interval of 80%, and a total of five predictors. The results showed that the required sample size for this study was 92. However, as the group of interest is Gen Y and Gen Z, this study is targeting twice the suggested sample size. Klang Valley was selected for data collection due to its extensive usage of digital instruments [39] compared to other regions of Malaysia as the respondents were required to understand the health functionality of WMDs. The data collection of this study was originally planned to be conducted through social media with a convenient sampling method due to the COVID-19 pandemic. However, to increase the robustness of the data, the researcher also collected several data physically and randomly from various public places. However, 105 responses had been excluded due to incompletely filled questionnaires. In the end, 330 of the total responses (75.86%) were accepted and considered eligible for further data analysis (Fig. 2).

Fig. 2. Research methodology

4.2 Data Analysis Procedure Partial least square structural equation model (PLS-SEM) was applied with the use of SmartPLS 3.0 according to the guidelines mentioned in [40]. To examine the Rsquared value for the explanation of endogenous latent variables and the significance and relevance of the path coefficient, bootstrapping with 5000 iterations was conducted as well. Measurement invariance of composite model (MICOM) was carried to ensure that the standardized path coefficient is comparable between Gen Y and Gen Z. Next, partial least square multigroup assessment (PLS-MGA) was performed to examine the moderating effect of Gen Y and Gen Z.

672

W. Z. Yap et al.

4.3 Descriptive Analysis Table 1 indicates the demographic characteristics of the respondents. Overall, 330 responses were collected to conduct this study, where 190 are Gen Z and 140 are Gen Y. Besides that, 172 of the participants are female and 157 are male. Also, 216 respondents have experience with WMD (here, smartwatch) health features, while 113 respondents have no such experience. Furthermore, 170 of the 330 respondents own a WMD (here, smartwatch) and 160 do not. Lastly, only 123 of the total respondents are actively using WMD health features and 207 are not. Table 1. Demographic characteristics of the respondents Characteristic

Groups

Population

Percentage (%)

Generation

Gen Z (age: 18–27 years) Gen Y (age 28–45 years)

190 140

58 42

Gender

Female Male

172 158

52 48

Experience with smartwatch’s health features

Yes No

217 113

66 34

Ownership

Yes No

170 160

51 49

Actively using smartwatch health features

Yes No

123 207

37 63

4.4 Analysis of Measurement Items The convergent validity, internal consistency reliability, and discriminant validity tests were conducted to ensure the validity of the data. As evident from Table 2, the factor loadings of all measurement items were found to be above the minimum requirement of 0.70, except for PU1 and WOM1, which were noted to be just slightly lower than the minimum requirement. However, all AVE values were observed to be above the minimum requirement of 0.50; hence, there is no convergent validity issue within the measurement model. Besides, Cronbach’s alpha and composite reliability were found to be above the required value of 0.70, and thus, no internal consistency reliability issue was found in the measurement model. The data also showed no discriminant validity issue (see Table 3), where the crossloading of the AVE of each construct was noted to be greater than the cross-loading of it with the other construct, and every HTMT value in Table 4 is lower than the value of 0.85 [42]. 4.5 Structural Equation Modeling The variance inflation factor value in this study ranges from 1.0265 to 2.7522 (Table 5). Hence, there is no collinearity issue in the research model as suggested in Ref. [43]. This

Exploring the Technology Acceptance

673

Table 2. Reliability and validity analysis Construct

Measurement item

Factor loading

AVE

Cronbach’s Alpha

Composite reliability

EWOM

EWOM1

0.766

0.721

0.880

0.911

EWOM2

0.802

EWOM3

0.905 0.900

0.963

0.973

0.745

0.886

0.921

0.708

0.864

0.906

0.766

0.896

0.928

0.699

0.872

0.902

IU

PC

PEoU

PU

WOM

EWOM4

0.914

IU1

0.951

IU2

0.962

IU3

0.952

IU4

0.929

PC1

0.8

PC2

0.895

PC3

0.887

PC4

0.867

PeoU1

0.734

PeoU2

0.859

PeoU3

0.891

PeoU4

0.873

PU1

0.669

PU2

0.926

PU3

0.949

PU4

0.926

WOM1

0.673

WOM2

0.803

WOM3

0.928

WOM4

0.916

study further examined the R-squared value for the explanation of endogenous variables, where the R-squared value of IU is 0.545 which is greater than the minimum value of 0.20 [40]. As shown in Table 5, the path coefficients of PU → IU (β = 0.503; p-value ≤ 0.01), WOM → IU (β = 0.103; p-value ≤ 0.05), and EWOM → IU (β = 0.206; pvalue ≤ 0.01) are significant. These results indicate that PU, WOM, and EWOM have a significant positive impact on IU of WMD among the younger generation.

674

W. Z. Yap et al. Table 3. Fornell–Larcker criterion EWOM

IU

PC

PeoU

EWOM

0.849

IU

0.569

0.949

PC

0.432

0.447

0.863

PeoU

0.524

0.433

0.611

0.841

PU

WOM

PU

0.576

0.704

0.533

0.531

0.875

WOM

0.55

0.52

0.365

0.382

0.573

0.836

PEoU

PU

WOM

Table 4. HTMT EWOM

IU

PC

EWOM IU

0.562

PC

0.507

0.474

PEoU

0.605

0.454

0.695

PU

0.628

0.734

0.615

0.609

WOM

0.607

0.49

0.449

0.448

0.602

Table 5. PLS-SEM Path coefficient

P-values

Corresponding hypothesis

PU → IU

0.503

0.000**

H1: Supported

PEoU → IU

0.022

0.366

H2: Not Supported

PC → IU

0.066

0.165

H3: Not Supported

WOM → IU

0.103

0.041*

H4: Supported

EWOM → IU

0.206

0.000**

H5: Supported

Note: Significance levels are denoted by * (5%) and ** (1%)

4.6 Measurement Invariance of Composite Model (MICOM) As this study used SmartPLS 3.0 to assess the MICOM as suggested in Ref. [41]. The results of MICOM confirm that the compositional invariance of the research model is established among Gen Y and Gen Z. The results further show that full measurement invariance is established in WOM, and another construct is established with partial measurement invariance. After fulfilling the criteria of the MICOM assessment, the study has proceeded with the PLS-MGA assessment.

Exploring the Technology Acceptance

675

4.7 Multigroup Analysis The PLS-MGA results show that the generational cohort does have a significant moderating effect on the path of PEoU → IU (p-value ≤ 0.01) but not in the other path coefficient within the research model. Nevertheless, the results confirm that H6 is supported (Table 6). Table 6. PLS-MGA

EWOM → IU PC → IU PEoU → IU PU → IU WOM → IU

Difference in path coefficient (Gen Y − Z)

p-value (Gen Y vs Z)

−0.01

0.467

0.018

0.439

−0.331

0.003

0.224

0.066

−0.061

0.313

Note: Significance levels are denoted by * (5%) and ** (1%)

After separating the sample population according to their generational cohort to run the bootstrapping result, the R-squared value had improved from 0.545 in total population to 0.563 and 0.562 for Gen Y and Gen Z, indicating that the explanation of endogenous variables had improved while analyzing each of the generational cohort separately. The results have shown that the path coefficient of PU → IU (β = 0.624; p-value ≤ 0.01), EWOM → IU (β = 0.175; p-value ≤ 0.01), and PEoU → IU (β = − 0.148; p-value ≤ 0.05) are significant in Gen Y, whereas the path coefficient of PU → IU (β = 0.399; p-value ≤ 0.01), WOM → IU (β = 0.131; p-value ≤ 0.05), EWOM → IU (β = 0.184; p-value ≤ 0.01), and PEoU → IU (β = 0.183; p-value ≤ 0.05) are significant in Gen Z (Table . 7). Table 7. Bootstrapping result by generational cohort Path coefficient (Gen Y)

Path coefficient (Gen Z)

P-value (Gen Y)

P-value (Gen Z)

EWOM → IU

0.175

0.184

0.006**

0.009**

PC → IU

0.054

0.025

0.32

0.391

−0.148

0.183

0.039*

0.015**

PEoU → IU PU → IU

0.624

0.399

0.000**

0.000**

WOM → IU

0.07

0.131

0.262

0.026*

Note: Significance levels are denoted by * (5%) and ** (1%)

676

W. Z. Yap et al.

5 Discussion The results of this study have shown that PU, WOM, and EWOM significantly influence IU-WMD among the younger generation in Malaysia. The findings are consistent with certain previous studies [5, 15, 18, 23, 24] and suggest that PU is one of the most robust factors in the study of technology acceptance of WMD among the younger generation. On the one hand, WOM and EWOM as social factors are equally crucial in influencing the IU-WMD of the younger generation. On the other hand, PEoU is found to have an insignificant impact on IU-WMD among the younger generation in Malaysia, which is consistent with previous studies [15, 19]. This result could be ascribable to the high WMD penetration and good ICT literacy in Malaysia [19]. The impact of PC was empirically investigated in this study, and it was found to have no significant impact on IU-WMD. Perhaps, the degree of convenience provided might have little impact on the acceptance of WMD. The results showed that Gen Y and Gen Z are able to moderate the PEoU toward IU-WMD. Gen Y’s PEoU has a significant negative impact on IU-WMD, while Gen Z’s PEoU has a significant positive impact on IU-WMD. The results indicated that the more the Gen Y individuals perceive the ease of use of WMD, the less likely they are to be willing to use WMDs to monitor their health. This could be due to Gen Y’s skepticism about the reliability of WMDs when these devices are too easy to use. However, the impact of EWOM on IU-WMD was similar for both Gen Y and Gen Z. Lastly, although the path of WOM does not have a significant moderating effect from the generational cohort, the results show that WOM does have a significant impact on Gen Z, but not in the case of Gen Y. Looking into the practical implications, the results have shown that PU and EWOM are the most impactful determinants of IU-WMD of all overall younger generations, and between Gen Y and Gen Z. Regardless, WMD providers are targeting the overall younger generation, but they need to emphasize the important health features that are perceived useful by their target consumers. Besides, the results of WOM and EWOM suggest that WMD providers could highlight the online key opinion leaders to encourage Gen Y or Gen Z to use WMD in Malaysia.

6 Conclusion and Limitations Driven by the potential benefit of the NCD issue and the market opportunities of WMDs in Malaysia among the younger population, this study has provided insights on the determinants of the IU-WMD among the younger generation in Malaysia, as well as the moderating effect of Gen Y and Gen Z. Despite the study’s meaningful findings and implications, it has a few limitations, which can be overcome by future studies. First, due to the complex challenges to include every factor into one unified model [44], future studies can investigate the impact of other factors such as cultural factors on IU-WMD among the younger generation in Malaysia. The results of this study are focused on the younger generation, whereas future studies may explore the factors influencing the older generation such as Gen X and baby boomers on the acceptance of WMD as they are more vulnerable in health as compared to the younger generation.

Exploring the Technology Acceptance

677

References 1. Li, J., Carayon, P.: Health care 4.0: a vision for smart and connected health care. IISE Trans. Healthcare Syst. Eng. (2021). https://doi.org/10.1080/24725579.2021.1884627 2. Voss, T.J., Subbian, V., Beyette, F.R.: Feasibility of energy harvesting techniques for wearable medical devices. 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2014, pp. 626–629 (2014https://doi.org/10.1109/ EMBC.2014.6943669 3. Lee, S.M., Lee, D.H.: Healthcare wearable devices: an analysis of key factors for continuous use intention. Serv. Bus. 14(4), 503–531 (2020). https://doi.org/10.1007/s11628-020-00428-3 4. O’Donovan, T., O’Donoghue, J., Sreenan, C., Sammon, D., O’Reilly, P., O’Connor, K.A.: A context aware wireless Body Area Network (BAN). 2009 3rd International Conference on Pervasive Computing Technologies for Healthcare - Pervasive Health 2009, PCTHealth 2009 (2009). https://doi.org/10.4108/ICST.PERVASIVEHEALTH2009.5987 5. Pataranutaporn, P., Jain, A., Johnson, C.M., Shah, P., Maes, P.: Wearable lab on body: combining sensing of biochemical and digital markers in a wearable device. In: 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBS), pp. 3327–3332 (2019). https://doi.org/10.1109/EMBC.2019.8857479 6. Reeder, B., David, A.: Journal of biomedical informatics. J. Biomed. Inform. 63, 269–276 (2016). https://doi.org/10.1016/j.jbi.2021.103845 7. Greiwe, J., Nyenhuis, S.M.: Wearable technology and how this can be implemented into clinical practice. Curr. Allergy Asthma Rep. 20(8), 1 (2020). https://doi.org/10.1007/s11882020-00927-3 8. Tison, G.H. et al.: Passive detection of atrial fibrillation using a commercially available smartwatch. JAMA Cardiol. 3(5) (2018). https://doi.org/10.1001/jamacardio.2018.0136 9. Zhou, J., Salvendy, G. (eds.): DUXU 2015. LNCS, vol. 9194. Springer, Cham (2015). https:// doi.org/10.1007/978-3-319-20913-5 10. Castaneda, D., Esparza, A., Ghamari, M., Soltanpur, C., Nazeran, H.: A review on wearable photoplethysmography sensors and their potential future application in hDealthcare. HHS Public Access 4(4), 195–202 (2019). https://doi.org/10.15406/ijbsbe.2018.04.00125.A 11. Isakadze, N., Martin, S.S.: How useful is the smartwatch ECG? Trends Cardiovasc. Med. 30(7), 442–448 (2020). https://doi.org/10.1016/j.tcm.2019.10.010 12. Hsu, H., Chang, Y.: Extended TAM model: impacts of convenience on acceptance and use of moodle. US-China Educ. Rev. A 3(4), 211–218 (2013) 13. National Institutes of Health, “Non-Communicable Diseases: Risk Factors and other Health Problems” (2019) 14. Kim, T., Chiu, W.: Consumer acceptance of sports wearable technology: the role of technology readiness. Int. J. Sports Mark. Spons. 20(1), 109–126 (2019). https://doi.org/10.1108/IJSMS06-2017-0050 15. Pfeiffer, J., Von Entress-Fürsteneck, M., Urbach, N., Buchwald, A.: Quantify-ME: consumer acceptance of wearable self-tracking devices. 24th European Conference on Information Systems, ECIS 2016, pp. 0–15 (2016) 16. Asadi, S., Safaei, M., Yadegaridehkordi, E., Nilashi, M.: Antecedents of consumers’ intention to adopt wearable healthcare devices. J. Soft Comput. Decision Support Syst. 6(2), 6–11 (2019) 17. Zhang, M., Luo, M., Nie, R., Zhang, Y.: Technical attributes, health attribute, consumer attributes and their roles in adoption intention of healthcare wearable technology. Int. J. Med. Informatics 108, 97–109 (2017). https://doi.org/10.1016/j.ijmedinf.2017.09.016 18. Al-Maroof, R.S., Alhumaid, K., Alhamad, A.Q., Aburayya, A., Salloum, S.: User acceptance of smart watch for medical purposes: an empirical study. Future Internet 13(5), 1–19 (2021). https://doi.org/10.3390/fi13050127

678

W. Z. Yap et al.

19. Loo, C.W.: “Modelling Malaysia intention to use smartwatch_ Smartwatch.pdf,” University of Wales Trinity Saint David (2022) 20. Beh, P.K., Ganesan, Y., Iranmanesh, M., Foroughi, B.: Using smartwatches for fitness and health monitoring: the UTAUT2 combined with threat appraisal as moderators. Behav. Inf. Technol. (2019). https://doi.org/10.1080/0144929X.2019.1685597 21. Davis, F.D.: Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quart. Manag. Inf. Syst. 13(3), 318–340 (1989) 22. Dutot, V., Bhatiasevi, V., Bellallahom, N.: Applying the technology acceptance model in a three-countries study of smartwatch adoption. J. High Technol. Manag. Res. 30, 1–14 (2019). https://doi.org/10.1016/j.hitech.2019.02.001 23. Yap, W.Z., Get, L.J., Sia, B.C., Goh, H.L.: The 2nd Conference on Management , Business , Innovation , Education , and Social Science (CoMBInES) The 2nd Conference on Management, Business, Innovation, Education, and Social Science (CoMBInES ),” The 2nd Conference on Management, Business, Innovation, Education, and Social Science (CoMBInES), pp. 164–177 (2022) 24. AlAjmi, Q., Al-Sharafi, M.A., Yassin, A.A.: Behavioral Intention of Students in Higher Education Institutions Towards Online Learning During COVID-19. In: Arpaci, I., Al-Emran, M., A. Al-Sharafi, M., Marques, G. (eds.) Emerging Technologies During the Era of COVID-19 Pandemic. SSDC, vol. 348, pp. 259–274. Springer, Cham (2021). https://doi.org/10.1007/ 978-3-030-67716-9_16 25. Arpaci, I., Al-Emran, M., Al-Sharafi, M.A.: The impact of knowledge management practices on the acceptance of Massive Open Online Courses (MOOCs) by engineering students: a cross-cultural comparison. Telematics Inf. 54 (2020). https://doi.org/10.1016/j.tele.2020. 101468 26. Al-Tahitah, A.N., Al-Sharafi, M.A., Abdulrab, M.: How COVID-19 Pandemic Is Accelerating the Transformation of Higher Education Institutes: A Health Belief Model View. In: Arpaci, I., Al-Emran, M., A. Al-Sharafi, M., Marques, G. (eds.) Emerging Technologies During the Era of COVID-19 Pandemic. SSDC, vol. 348, pp. 333–347. Springer, Cham (2021). https:// doi.org/10.1007/978-3-030-67716-9_21 27. Venkatesh, V., Morris, M.G., Davis, G.B., Davis, F.D.: User acceptance of information technology: toward a unified view. Inorg. Chem. Commun. 27(3), 425–478 (2003). https://doi. org/10.1016/j.inoche.2016.03.015 28. Venkatesh, V., Thong, J.Y.L., Xu, X.: Consumer acceptance and use of information technology: extending the unified theory of acceptance and use of technology. MIS Quart. Manag. Inf. Syst. 36(1), 157–178 (2012). https://doi.org/10.2307/41410412 29. Huete-Alcocer, N.: A literature review of word of mouth and electronic word of mouth: implications for consumer behavior. Front. Physiol. 8(1256), 1–4 (2017). https://doi.org/10. 3389/fpsyg.2017.01256 30. Hsiao, K.L.: What drives smartwatch adoption intention? Comparing Apple and non-Apple watches. Library Hi Tech 35(1), 186–206 (2017). https://doi.org/10.1108/LHT-09-2016-0105 31. Ruangkanjanases, A., Wongprasopchai, S.: Adoption of mobile banking services: an empirical examiniation between generation y and generation z in Thailand 1(1), 1–12 (2018) 32. Cham, T.H., Cheng, B.L., Low, M.P., Cheok, J.B.C.: Brand image as the competitive edge for hospitals in medical tourism. Europ. Bus. Rev. 33(1) (2021). https://doi.org/10.1108/EBR10-2019-0269 33. Memon, M.A., Ting, H., Cheah, J.-H., Thurasamy, R., Chuah, F., Cham, T.H.: Sample size for survey research: review and recommendations. J. Appl. Struct. Eq. Model. 4(2), i–xx (2020). https://doi.org/10.47263/jasem.4(2)01 34. “Contribution of Digital Economy was 18.5 per cent to National Economy,” Department of Statistics Malaysia, Oct. 16, 2019. https://dosm.gov.my/v1/index.php?r=column/cth

Exploring the Technology Acceptance

35. 36.

37.

38.

679

eme&menu_id=b0pIV1E3RW40VWRTUkZocEhyZ1pLUT09&bul_id=UERrcjExbnRK d3NwQ0Y3V011RngyQT09. Accessed 12 July 2021 Hair, J.F., Hult, G.T.M., Ringle, C.M., Sarstedt, M.: A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM). Thousand Oaks (2017) Henseler, J., Ringle, C.M., Sarstedt, M.: A new criterion for assessing discriminant validity in variance-based structural equation modeling. J. Acad. Mark. Sci. 43(1), 115–135 (2014). https://doi.org/10.1007/s11747-014-0403-8 Cheah, J.H., Thurasamy, R., Memon, M.A., Chuah, F., Ting, H.: Multigroup analysis using smartpls: Step-by-step guidelines for business research. Asian J. Bus. Res. 10(3) (2020). https://doi.org/10.14707/ajbr.200087 Niknejad, N., Hussin, A.R.C., Ghani, I., Ganjouei, F.A.: A confirmatory factor analysis of the behavioral intention to use smart wellness wearables in Malaysia. Univ. Access Inf. Soc. 19(3), 633–653 (2019). https://doi.org/10.1007/s10209-019-00663-0

Can LSTM Model Predict the Moroccan GDP Growth Using Health Expenditure Features? Ismail Ouaadi1(B)

and Aomar Ibourk2

1 Management Sciences, FSJES Agdal, Mohammed V University of Rabat, Rabat, Morocco

[email protected] 2 Microeconometrics, LARESSGD University of Marrakech, Marrakech, Morocco

Abstract. In this work we attempt to introduce recurrent neural network algorithms that can be used as means to treat healthcare expenditures and their impact on economic growth, and through this to help in making better decisions for forecasting the Gross Domestic Product (GDP). As methodology, we gather data related to healthcare expenditures and the Moroccan GDP from 1980 till 2020 and then we utilize these data to train and validate a Long Short-Term Memory (LSTM), which enable the development and the use of algorithms and statistical models to analyze and draw inferences from pattern and creating both short-term and long-term memory components to efficiently study and learn sequential data. In this study, we build three LSTM models to show the performance of these models in forecasting the GDP growth based on healthcare expenditure in the case of Morocco. The results showed that LSTM models produce forecasts that have different level of accuracy according to the categories of health expenditure, and according to the RMSE and MAE accuracy metrics we can assume that the LSTM model with pure investment in health is more accurate than the other models. Keywords: GDP forecasting · Health expenditure · LSTM algorithm

1 Introduction The nexus between GDP growth and health expenditure gets more importance nowadays, given the emergence of some critical health issues like SARS, COVID’19 and Monkeypox virus. According to World Health Organization (WHO) [7], over the last two decades, global health spending has more than doubled, hitting 8.5 trillion in 2019, or 9.8% of global Gross domestic product (GDP). While, in this WHO report, during the COVID’19 pandemic, early estimates from high-income economies indicate that health care spending has increased significantly in 2020, more than in previous years. Given that, investments in health are part of the objectives of sustainable development, which in turn aims to improve the economic, social and ecological conditions of nations. These investments can be divided as pure investments, which devoted to acquire new machines, equipments, hiring employees, and to build new infrastructures (hospitals, centres...), and as operational investments, which are devoted the wage payments, maintenance and utilization of machines, equipments and infrastructures. Health investments in Morocco has increased from 1% to 1.5% of GDP. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 680–689, 2023. https://doi.org/10.1007/978-3-031-20429-6_61

Can LSTM Model Predict the Moroccan GDP Growth

681

Moreover, the digitization era allows the collection and storage of vast amounts of data, which presents significant obstacles for decision-makers due to the need for abilities to process, analyze, and make decisions based on this data. Furthermore, the introduction of sophisticated artificial intelligence algorithms that have proven to be quite efficient in forecasting and generating predictions in a variety of sectors has advocated for more and more dependable studies. Taking these factors into account, the goal of our research is to use artificial intelligence techniques to offer some methods for dealing with health investments variables in order to forecast GDP growth. Artificial intelligence, in general, refers to a computer’s ability to do activities that resemble human intelligence. It refers to a group of approaches that work together to attain this goal, specifically machine learning techniques. Deep learning (DL) algorithms, which enable the construction and application of algorithms and statistical models to evaluate and make inferences from patterns in data, are a class of these advanced and complex methods. To demonstrate the performance of these models in predicting GDP growth, we introduce a specific method from the Recurrent Neural Network class called LSTM (for Long short-term memory) as a function of health investments variables in the case of Morocco. And in a view to highlight the usefulness of DL algorithms in macroeconomic fields, and to show the appropriateness of these algorithms to deal with timeseries data. Our work is structured as follows, according to this introduction: The next part provides some theoretical foundation and discusses how machine learning techniques can be used to forecast GDP growth. The methodology part, in which we detail the data we used and how we developed our model, is the third section. The remain of this paper is organized as follow: the findings in section four, which is followed by a summary of our article, limitations, and future work in section five.

2 Literature Review Given the importance of GDP growth and health investment, the literature on these two fields got more and more attention. While many works address the relationship between these fields, others attempt to study each field separately. Some of these works implement econometrics models to investigate the issues related to the subject that examine the relationship between GDP growth and climate change, such as a recent one [9], which has used an extended ordinary least squares estimation to analyze the dynamic impacts of economic growth, renewable energy use, and agricultural land expansion on CO2 emissions in Peru. Moreover, the study in [6] seeks to demonstrate the health expenditure-economic growth relationship in China over the period 1980–2017 using an econometric and Machine Learning technique. New research, such as [11], have looked at the application of machine learning algorithms to forecast GDP growth. The author employed gradient boosting (GB) and random forest (RF) models to anticipate real GDP growth in Japan. He forecasted annual real GDP growth using data from the Bank of Japan and the International Monetary Fund (IMF) from 1981 to 2018. For the period 2001–2018, he discovered that GB and RF are more accurate than the benchmark forecast model. Similarly, the study described in [4] aims to conduct economic prediction using machine learning approaches (Elastic Net,

682

I. Ouaadi and A. Ibourk

Recurrent Neural Network (RNN), and Super Learner). Seven countries’ GDP growth are forecasted using these methodologies (Germany, Mexico, Philippines, Spain, United Kingdom, United States, and Vietnam) using quarterly and annually data, where the data is taken from the World Economic Outlook (WEO) and the database of the International Monetary Fund. The results reveal that the machine learning method outperforms the IMF’s WEO forecast performance. Machine learning algorithms have employed also in many fields like education [3] where authors attempt to predict grade repetition based on some features, and in [2] authors try to predict teacher effectiveness of more than 800 teacher in preparation cycle in Morocco. Reference [5] used a more advanced machine learning technology called Feedforward Multilayer Perceptron, which is a particular algorithm of artificial neural network, to construct a forecasting model to predict GDP volatility. The author discovered that machine learning algorithms can be useful in public and financial management exploiting data from the UK Office for National Statistics from 1971 to 2013. Reference [10] is another study that has applied such complex models. The authors used two DL algorithms, recurrent neural network and Long Short-Term Memory (LSTM), to anticipate Indonesia’s GDP fluctuation using GDP and inflation percentages as predictors using data from the World Bank, Macrotrend, and Financial Trade Data from 1990 to 2020. They discovered that the LSTM and RNN algorithms can accurately estimate GDP growth variation by 90%. In [8] authors show the performance of an LSTM algorithm in predicting GDP growth based on drought indexes.

3 Methodology A set of stages should be carefully followed in order to develop a strong model capable of making proper and accurate predictions. (i) data preprocessing, which includes cleaning and handling data; (ii) model training and validation; and (iii) accuracy evaluation, are the most significant steps. This section is organized as follows based on these phases. A brief overview of the dataset and the features employed is given first. Second, the proposed data preparation approach is provided. Finally, the predictive models that were chosen for execution are described. 3.1 Data Description and Preprocessing The monthly GDP data from the Moroccan Financial Studies and Forecasting Department Database (available at https://manar.finan-ces.gov.ma) was used because our goal is to anticipate GDP growth based on health expenditures. Health investments data related to the Morocco are gathered from the same database, where health expenditure is categorized according to the categories of investment discussed in the introduction section. Our research spans the years 1980 to 2020 (Table 1). GDP growth variable ( GDP) is computed by taking the variation between two consecutive years (Eq. 1): GDP (n) =

GDP (n) − GDP (n−1) GDP (n−1)

(1)

Can LSTM Model Predict the Moroccan GDP Growth

683

Table 1. Descriptive statistics of the data. GDP

Pure investment in health

Operational investment in health

Total health expenditure

count

41

41

41

41

mean

515084.15

852.49

4846.54

5699.02

std

328767.25

737.15

4022.95

4743.70

min

85537.48

72.00

645.82

748.82

2 5%

248753.51

189.00

1451.29

1640.29

5 0%

412897.15

615.00

2991.00

3639.00

7 5%

784624.00

1382.00

7845.00

9174.00

max

1152806.00

2735.00

13421.00

16156.00

Where: GDP(n) = GDP of year n GDP(n–1) = GDP of year n–1 Our dataset is composed of 41 observations, taking into account that the first year is dropped because it is used to compute the next year’s GDP growth rates (here 1981). Moreover, this dataset encompasses four variables, the first (GDP GROWTH) of which is the forecasted variable, and the other three variables (pure investment in health, operational investment in health and total health expenditure) are designed as features. Figure 1 illustrates the distribution of all variables (pure investment in health, operational investment in health, total health expenditure and GDP GROWTH). The analysis of the subplots allows the determination of the stationarity (seasonality and trend) of these variables. Here we assume that the data are stationary and hence we perform our model. 3.2 Model The current problem is nonlinear, as we mentioned in the previous subsection. As a result, standard artificial neural networks (ANN) are unable to provide us with appropriate results. As a result, we’ve used a type of neural network known as a recurrent neural network (RNN) to deal with the nonlinearity issue. Whereas ANNs are widely used in time-series data prediction, which they present some lack due to their architectures that do not take sequence data into account, RNNs, on the other hand, provide an alternative way to deal with this type of data, allowing for consistent and good observation forecasting [10]. While having hidden states, RNN allows previous outputs to be used as inputs, and it admits current hidden layers to be updated depending on previous knowledge. In this work, we have a timeseries, which is constituted, usually, by a huge amount of historical data and need powerful computing resources to process. For this reasons RNN techniques presents a sort of algorithm that can hold only relevant data and capable to solve the long-term dependency problem. This algorithm, called LSTM (for Long-Short

684

I. Ouaadi and A. Ibourk

Fig. 1. Data analysis plot

Term Memory), is formed by three main gates: Forget, Input and Output. These gates allow LSTM model to keep previous information or delete irrelevant knowledge. The LSTM architecture is given in Fig. 2, and its formulas according to each gate in Eqs. 3, 4 and 5.

Fig. 2. LSTM architecture

Can LSTM Model Predict the Moroccan GDP Growth

685

Forget gate unit f i (t) (for time step t and cell i), which sets weights to a value between 0 and 1 via a sigmoid unit [1]: ⎞ ⎛  f (t)  f (t−1) f (t) ⎠ fi = σ ⎝bi + (2) Ui,j xj + Wi,j hj j

j

The last gate of the LSTM cell is the output qi (t) , which, like the other gates, uses a sigmoid unit for gating: ⎛ ⎞   g g g ⎠ gi(t) = σ ⎝bi + Ui,j xj(t) + Wi,j h(t−1) (3) j j

j

The last gate of the LSTM cell is the output qi (t) , which, like the other gates, uses a sigmoid unit for gating: ⎛ ⎞   (t) o (t) o (t−1) ⎠ qi = σ ⎝boi + Ui,j xj + Wi,j hj (4) j

j

where x (t) is the current input vector and h(t) is the current hidden layer vector, containing the outputs of all the LSTM cells, b(f) , U (f) and W (f) are biases, input weights and recurrent weights for the forget gates, respectively.

4 Result This section is reserved to describe the experiment methodology, to report the findings of the training and testing the algorithms, and to discuss the results lastly. 4.1 Experiment Fitting Models. To fit our model, we have used Keras from the Python packages. Once the dataset is loaded, it is splited into training, test and validation dataset. LSTM algorithm is implemented with sigmoid activation function and 200 epochs. Figure 3 shows model fitted with pure investment in health as feature to predict annual GDP growth. Similarly, the second Fig. 4 shows model fitted with operational investment in health as feature, the last one (Fig. 5) shows model fitted with total health expenditure for predicting quarterly GDP growth. As observed, in Figs. 3, 4 and 5, these models tend to converge, while the number of epoch increase. Consequently, the gap between train losses and test losses is getting smaller and smaller. In addition, these models may perform very well if we change the activation function or add some hidden layers or adjust some other parameters. Forecasting and simulation. Figures 6, 7 and 8, illustrate the results of the predictions made by these two models. Graphicly, we can infer the success of these models, where the trend of predicted values follows the real values. In other words, the LSTM model

686

I. Ouaadi and A. Ibourk

Fig. 3. Model fitted with pure investment in health.

Fig. 4. Model fitted with operational investment in health.

Fig. 5. Model fitted with total health expenditure.

can predict the outputs very well. To perform a rigorous accuracy assessment of the models, the root mean square error (RMSE) and the mean absolute error (MAE) metrics are computed based on the following formulas:    n

 yˆ i − yi 2 (5) RMSE = n i=1

and n MAE =

i=1



yˆ i − yi n

Can LSTM Model Predict the Moroccan GDP Growth 



687



where, y1 , y2 , …, yn are the predicted values; y1 , y2 , …, yn are the observed values. n is the number of observations. Table 2 presents the RMSE and MAE computed for each model:

Fig. 6. Model fitted with pure investment in health.

Fig. 7. Model fitted with operational investment in health.

Fig. 8. Model fitted with total health expenditure.

4.2 Results and Discussion Based on the results shown in Figs. 6, 7, and 8, by means of the LSTM algorithms the Moroccan GDP growth can be predictable by the healthcare expenditure, and hence it can be predicted also by other healthcare variables. It can be inferred that some indexes perform better than other as features. From the Table 2, we can say with some degree of

688

I. Ouaadi and A. Ibourk Table 2. Accuracy metrics. First model (pure investment)

Second model (operational investment)

Third model (total expenditure)

RMSE

0.026

0.030

0.029

MAE

0.020

0.024

0.023

confidence that the LSTM model with pure investment in health is more accurate than the other models. Our results are consistent with those found in the literature, which means that machine learning algorithms provide powerful tools to make predictions than classical tools. Most of studies have compared their findings with some benchmark models of forecasting like those produced by the IMF and the Central Banks. This strategy will be covered in further research to better understand the relationship among Moroccan GDP growth and other healthcare variables. Moreover, we have implemented an LSTM model with a limited number of parameters and hidden layers, but some kind of models that use more parameters and more data can provide better results. In addition, we have used two accuracy metrics to evaluate the performance of our models, namely RMSE and MAE, which are the most popular metrics. It is, however, possible to perform more metrics to evaluate the accuracy and robustness of machine learning models.

5 Conclusion The finding of this study advocates the use of machine learning techniques in forecasting macroeconomic data. The DL method employed in this study, which creates LSTM models for the 1980–2020 period, produces forecasts that have different level of accuracy given the type of features. While traditional econometric models focus on explanations of the relationships, machine learning models focus on predictions. Precisely, DL models are considered as black boxes, which means that they are not a good choice for determining the impact of independent variables on the dependent variable or analyzing a causal relationship. However, as we can find in multiple previous studies and as results of this study, DL models often show sufficient forecasting accuracy. In further work, we will try to model GDP growth prediction with different environmental indexes as new features and by introducing new techniques of DL, in intention to get a powerful model that allows to have more accurate and robust results.

References 1. Goodfellow, I., Bengio, Y., Courville, A.: Deep Learning. MIT press (2016) 2. Ibourk, A., Hnini, K., Ouaadi, I.: Analysis of the pedagogical effectiveness of teacher qualification cycle in morocco: a machine learning model approach. In: Studies in Computational Intelligence. Springer, (in press)

Can LSTM Model Predict the Moroccan GDP Growth

689

3. Ibourk, A., Ouaadi, I.: An exploration of student grade retention prediction using machine learning algorithms. In: International Conference on Business Intelligence, pp. 94–106. Springer (2022) 4. Jung, J.K., Patnam, M., Ter-Martirosyan, A.: An algorithmic crystal ball: forecasts-based on machine learning. Int. Monetary Fund (2018) 5. Kouziokas, G.N.: Machine learning technique in time series prediction of gross domestic product. In: Proceedings of the 21st Pan-Hellenic Conference on Informatics, pp. 1–2 (2017) 6. Mele, M., Randazzo, L.: On the chinese’ health expenditure: from toda-yamamoto to machine learning approach. J. Chinese Econ. Bus. Stud. 18(4), 289–309 (2020) 7. Organization, W.H. et al.: Global expenditure on health: public spending on the rise? (2021) 8. Ouaadi, I., Ibourk, A.: The contribution of deep learning models: application of lstm to predict the moroccan gdp growth using drought index. In: Studies in Computational Intelligence. Springer, (in press) 9. Raihan, A., Tuspekova, A.: The nexus between economic growth, renewable energy use, agricultural land expansion, and carbon emissions: new insights from peru. Energy Nexus, p. 100067 (2022) 10. Sa’adah, S., Wibowo, M.S.: Prediction of gross domestic product (gdp) in indonesia using deep learning algorithm. In: 2020 3rd International Seminar on Research of Information Technology and Intelligent Systems (ISRITI), pp. 32–36. IEEE (2020) 11. Yoon, J.: Forecasting of real gdp growth using machine learning models: Gradient boosting and random forest approach. Comput. Econ. 57(1), 247–265 (2021)

Author Index

MISC Özel, Selma Ay¸se, 95 A Abdallah, Sheriff, 613 Abdelmaksoud, Sherif I., 246 Abdelrahim, Misbah, 287 Abdulbaqi, Azmi Shawkat, 22, 66, 575 Abdulkadir, Said Jadid, 118 Abdulkhaleq, Mohammed, 546 Abdullah, Mohd Hafizul Afifi, 108, 118 Abdul-Qawy, Antar Shaddad Hamed, 181 Abdulsada, Mohammed Abdulla, 225 Abro, Rizwan Ali, 501 Agrawal, Reeya, 235 Ahmad, Azrina, 426 Ahmed, Mohammed A., 53 Akhiate, Aziz, 459, 491, 526 Akhir, Emelia Akashah Patah, 108, 118 Al-Abbas, Amjed R., 95 Alasbali, Asma Abdullah, 367 Al-Askari, Mohanad A., 66 Al-Awlaqi, Mohammed Ali, 392 Al-Barzinji, Shokhan M., 66 AL-Dhief, Fahad Taha, 270 Alduais, Nayef, 181 Alghenaim, Mohammed Fahad, 298 Alhadawi, Hussam S., 270 Al-Hamdani, Khalid I., 451 Ali, Haider Abdullah, 3 Ali, Mohammed A. H., 181 Ali, Yaseen Hadi, 3 Aljaberi, Musheer Abdulwahid, 546 Alkahtani, Ammar Ahmed, 287

Alkawsi, Gamal, 287 Almannaee, Shamma, 150 Al-Saffar, Bashar, 3, 95 Al-Salihi, Anam R., 647 Al-Samhi, Nezar, 392 Al-shakri, Hamid S., 451 Alshehhi, Shamma, 150 Althewaynee, Hasan Badir, 169 Alzuabidi, Ola Hussein Abd Ali, 225 Amur, Zaira Hassan, 501 Anmary, Antony Sheela, 546 Anwarsha, A., 76 Aw, Eugene Cheng-Xi, 426 Aziz, Norshakirah, 108, 118 Aziz, Omar Sabraldeen, 319 B Baashar, Yahia, 287 Badrani, Morad, 361 Baharum, Harmi Izzuan, 367 Bakar, Noor Akma Abu, 128 Bakar, Nur Azaliah Abu, 298 Baruah, Amlan Jyoti, 436 Baruah, Siddhartha, 436 Belkebir, Malak, 32 Benrahal, Mohamed, 459, 491, 526 Bilquise, Ghazala, 42 Bourhim, El Mostafa, 459, 481, 491, 526 C Cham, Tat Huei, 84, 667 Cham, Tat-Huei, 536, 562 Cheah, Phaik Kin, 536 Chellasamy, Aarthy, 657

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 M. A. Al-Sharafi et al. (Eds.): ICETIS 2022, LNNS 573, pp. 691–693, 2023. https://doi.org/10.1007/978-3-031-20429-6

692 Chetouani, Abdelaziz, 361 Cob, Zaihisma Che, 382 D Dahane, Ali, 459, 491, 526 Dauwed, Mohammed, 216, 235 Dorsey, Colby, 638 E Ebrahimi, Mansoureh, 515 Elangovan, N., 657 Elsersy, Mohamed, 638 Eng, Poh Hwa, 562 Er, Pek Hoon, 536 F Fitrianah, Devi, 138 Foo, Wen Hui, 84 Fuji, S. Ayu Eka, 138 G Ghaleb, Sanaa A. A., 181 Ghanem, Waheed, 181 Gilal, Abdul Rehman, 278, 501 Gilal, Ruqaya, 278 Goh, Hong Lip, 84, 667 H Hamed, Radwan Nazar, 575 Hammad, Anfal Hamid, 22 Hamood, Maytham M., 169 Hasim, Nurulhaida, 469 Hassan, Ahmad Muhyuddin, 515 Helwan, Abdulkader, 627 Hilaluddin, Thaharah, 382 Hing, Tang H., 246 Husain, Nik Rosmawati Nik, 12 Hussain, Nor Haliza Che, 426 Hussain, Zena, 319 Hussein, Harith A., 162, 169 Hussein, Mohammed Wajeeh, 225 I Ibourk, Aomar, 680 Ibrahim, Emad Mahmood, 257 Ibrahim, Samar, 585, 613 J Jaafar, Jafreezal, 278, 501 K Kader, Nur Izzati Ab, 12 Kannouf, Nabil, 361 Khalid, Mohd Nor Akmal, 12 Khamiss, Nasser N., 647

Author Index Kharwar, Saurabh, 207, 216 Kobbaey, Thaeer, 42 Kumar, Balivada Yashwant, 216 Kumari, Arti, 207 L Labti, Oumayma, 459, 481, 491, 526 Lee, Chee Heong, 536 Lim, Siew Mooi, 128, 138 M Ma’aitah, Mohamad Khaleel Sallam, 627 Maarouk, Toufik Messaoud, 32 Maasum, Tg Nor Rizan Tg Mohamad, 337 Mahdi, Ahmed Salih, 270 Mailah, Musa, 246 Majid, Mazlina Abdul, 128 Marouan, Adil, 361 Mehdy, Mehdy Mwaffeq, 647 Mejdoub, Mahmoud, 257 Mohammed, Ahmed Y., 162 Mohammed, Lubna Ali, 546 Mohammed, Mustafa K. A., 207, 216, 235 Mohammed, Nur, 603 Mohammed, Sarah Raad, 647 Momen, Abdul, 515 Moosavi, Nazanin, 308 Moses, Priscilla, 536 Musa, Muhamad Nabil, 108 Muslim, Ali M., 3 N Nagarathinam, Aishwarya, 657 Narendiranath Babu, T., 76 Nasser, Abdullah B., 181 Nini, Brahim, 32 O Osman, Nurul Aida, 108 Ouaadi, Ismail, 680 P Palli, Abdul Sattar, 278 Pereira, Jerito, 402, 416 Polley, Timothy, 638 Prihandini, Rafiantika Megahnia, 416 R Rahim, Fitriya Abdul, 426 Rahim, Fiza Abdul, 298 Ratnasari, Anita, 138 Razak, Norizan Abdul, 337 Rengasamy, Sangeetha, 657 Rennier, Daniel, 638 S Saad, Abdul-Malik, 181

Author Index Saad, Nor Hasliza Binti Md, 392 Salleh, Narishah Mohamed, 194 Salman, Yasir Dawood, 270 Shahril Khuzairi, Nur Maisarah, 382 Shatnawi, Maad, 150 Sheng, Zhao Ming, 562 Sherif, Ahmed, 638 Shuwandy, Moceheb Lazam, 162 Sia, Bee Chuan, 84, 667 Singh, Sangeeta, 207, 216, 235 Soares, Bento, 416 T Taherdoost, Hamed, 308 Talpur, Bandeh Ali, 278, 501 Talpur, Noureen, 118 Tan, Shiwei, 402 Tang, Jianlan, 402, 416 Tang, Tiong Yew, 194 Taqi, Ahmed Mohammed, 392 Tariq, Zeena, 319 Tat-Huei, Cham, 426

693 Tey, Tiny Chiu Yuen, 536 Tiong, Sieh Kiong, 287 W Waqas, Ahmad, 278, 501 Watkins, Matthew, 638 Wijaya, Tommy Tanu, 402, 416 Wong, Mikkay Ei Leen, 194 Y Yap, Way Zhe, 84, 667 Yassin, Amr Abdullatif, 337 Yusof, Junaidah, 469 Yusof, Umi Kalsom, 12 Z Zaghden, Nizar, 257 Zahidur Rahman, Md., 603 Zaki, Salim M., 207 Zin, Zuhana Mohamed, 367 Zulfakar, Zufara Arneeda, 426