Intelligent Systems Design and Applications: 18th International Conference on Intelligent Systems Design and Applications (ISDA 2018) held in Vellore, India, December 6-8, 2018, Volume 1 [1st ed.] 978-3-030-16656-4;978-3-030-16657-1

Table of contents :
Front Matter ....Pages i-xxii
A Study of Multi-space Search Optimization (Derrick Beckedahl, Andreas Nel, Nelishia Pillay)....Pages 1-9
Clinical Decision Support System for Neuro-Degenerative Disorders: An Optimal Feature Selective Classifier and Identification of Predictor Markers (Lokeswari Venkataramana, Shomona Gracia Jacob, S. Saraswathi, R. Athilakshmi)....Pages 10-20
Favoring the k-Means Algorithm with Initialization Methods (Anderson Francisco de Oliveira, Maria do Carmo Nicoletti)....Pages 21-31
A Novel Design and Implementation of 8-Bit and 16-Bit Hybrid ALU (Suhas B. Shirol, S. Ramakrishna, Rajashekar B. Shettar)....Pages 32-42
Authentication Scheme Using Sparse Matrix in Cloud Computing (Sunita Meena, Shivani Kapur, Vipin C. Dhobal, Subhrat Kr. Sethi)....Pages 43-52
A Thermal Imaging Based Classification of Affective States Using Multiclass SVM (C. M. Naveen Kumar, G. Shivakumar)....Pages 53-63
Multidimensional Crime Dataset Analysis (Prerna Kapoor, Prem Kumar Singh)....Pages 64-72
AKCSS: An Asymmetric Key Cryptography Based on Secret Sharing in Mobile Ad Hoc Network (R. Preethi, M. Sughasiny)....Pages 73-86
API Call Based Malware Detection Approach Using Recurrent Neural Network—LSTM (J. Mathew, M. A. Ajay Kumara)....Pages 87-99
Review and Analysis of Zero, One and Few Shot Learning Approaches (Suvarna Kadam, Vinay Vaidya)....Pages 100-112
Simulation of Friction Stir Welding of Aluminium Alloy AA5052 – Tailor Welded Blanks (M. Arun Siddharth, R. Padmanaban, R. Vaira Vignesh)....Pages 113-122
Information Systems Success: Extending the Theoretical Model from IT Business Value Perspective (Thanh D. Nguyen)....Pages 123-137
Towards an Automatic Detection of Sensitive Information in Mongo Database (Houyem Heni, Faiez Gargouri)....Pages 138-146
Business Growth Using Open Source e-Commerce and ERP in Small Business (Valtteri Kujala, Raija Halonen)....Pages 147-158
Directional Multiscale Feature Extraction for Biomedical Image Indexing and Retrieval Using Contourlet Transform (Amita A. Shinde, Amol D. Rahulkar, Chetankumar Y. Patil)....Pages 159-169
Link Quality and QoE Aware Predictive Vertical Handoff Mechanism for Video Streaming in Urban VANET (Emna Bouzid Smida, Sonia Gaied Fantar, Habib Youssef)....Pages 170-181
Performance Comparison of PID and ANFIS Controller for Stabilization of x and x-y Inverted Pendulums (Ishan Chawla, Vikram Chopra, Ashish Singla)....Pages 182-192
Modeling Hybrid Indicators for Stock Index Prediction (R. Arjun, K. R. Suprabha)....Pages 193-202
XOR Encryption Techniques of Video Steganography: A Comparative Analysis (Namrata Singh)....Pages 203-214
Intention to Use M–Banking: The Role of E–WOM (Thanh D. Nguyen, Thy Q. L. Nguyen, Thi V. Nguyen, Tung D. Tran)....Pages 215-229
Activity Gesture Recognition on Kinect Sensor Using Convolutional Neural Networks and FastDTW for the MSRC-12 Dataset (Miguel Pfitscher, Daniel Welfer, Marco Antonio de Souza Leite Cuadros, Daniel Fernando Tello Gamarra)....Pages 230-239
Plug in Electric Vehicle-Wind Integrated Multi-area Automatic Generation Control Tuned by Intelligent Water Drops Algorithm (Subhranshu Sekhar Pati, Aurobindo Behera, Tapas Kumar Panigrahi)....Pages 240-250
Design of Time-Frequency Localized Filter Bank Using Modified Particle Swarm Optimization (Swati P. Madhe, Amol D. Rahulkar, Raghunath S. Holambe)....Pages 251-261
Development of Low-Cost Real-Time Driver Drowsiness Detection System Using Eye Centre Tracking and Dynamic Thresholding (Fuzail Khan, Sandeep Sharma)....Pages 262-271
A Hybrid Entropy Based Method Using Gaussian Kernel for Retinal Blood Vessel Segmentation (N. K. Adhish, R. Rajesh, T. M. Thasleema)....Pages 272-279
Precision Crop Protection Using Wireless Sensor Network (R. Radha, Amit Kumar Tyagi, K. Kathiravan, G. Staflin Betzy)....Pages 280-290
Deep Learning Based Approach for Classification and Detection of Papaya Leaf Diseases (Rathan Kumar Veeraballi, Muni Sankar Nagugari, Chandra Sekhara Rao Annavarapu, Eswar Varma Gownipuram)....Pages 291-302
Three-Materials Image Recover from Value Range Projection Data (Chuanlin Liu, Amit Yadav, Asif Khan, Jing Zou, Weizhen Hu)....Pages 303-314
Multiple Criteria Fake Reviews Detection Using Belief Function Theory (Malika Ben Khalifa, Zied Elouedi, Eric Lefèvre)....Pages 315-324
Improved Logistic Regression Approach in Feature Selection for EHR (Shreyal Gajare, Shilpa Sonawani)....Pages 325-334
Background Modeling Using Deep-Variational Autoencoder (Midhula Vijayan, R. Mohan)....Pages 335-344
Sewage Sludge Removal Method Through Arm-Axis by Machine Robot (M. Gobinath, S. Malathi)....Pages 345-353
K-Nearest Neighbors Under Possibility Framework with Optimizing Parameters (Sarra Saied, Zied Elouedi)....Pages 354-364
A Visual Spelling System Using SSVEP Based Hybrid Brain Computer Interface with Video-Oculography (D. Saravanakumar, M. Ramasubba Reddy)....Pages 365-375
QBEECH: Multi-hop Clustering of Cognitive Based Sensor Nodes in the Administration of Queen Nodes (Souvik Kundu, Srividhya Karthikeyan, A. Karthikeyan)....Pages 376-385
Perceive Core Logical Blocks of a C Program Automatically for Source Code Transformations (Pallavi Ahire, Jibi Abraham)....Pages 386-400
Asymmetric Key Cryptosystem and Digital Signature Algorithm Built on Discrete Logarithm Problem (DLP) (Ashish Kumar, Jagadeesh Kakarla, Muzzammil Hussain)....Pages 401-410
A Study on Big Cancer Data (Sabuzima Nayak, Ripon Patgiri)....Pages 411-423
Food Monitoring Using Adaptive Naïve Bayes Prediction in IoT (Pramod D. Ganjewar, Selvaraj Barani, Sanjeev J. Wagh, Santosh S. Sonavane)....Pages 424-434
Mixed Credit Scoring Model of Logistic Regression and Evidence Weight in the Background of Big Data (Keqin Chen, Kun Zhu, Yixin Meng, Amit Yadav, Asif Khan)....Pages 435-443
A Model for Identifying Historical Landmarks of Bangladesh from Image Content Using a Depth-Wise Convolutional Neural Network (Afsana Ahsan Jeny, Masum Shah Junayed, Syeda Tanjila Atik, Sazzad Mahamd)....Pages 444-454
M2U2: Multifactor Mobile Based Unique User Authentication Mechanism (Rachit Bhalla, N. Jeyanthi)....Pages 455-464
Generation of Image Caption Using CNN-LSTM Based Approach (S. Aravindkumar, P. Varalakshmi, M. Hemalatha)....Pages 465-474
ADABA: An Algorithm to Improve the Parallel Search in Competitive Agents (Lídia Bononi Paiva Tomaz, Rita Maria Silva Julia)....Pages 475-485
A Novel Approach to Solve Class Imbalance Problem Using Noise Filter Method (Gillala Rekha, Amit Kumar Tyagi, V. Krishna Reddy)....Pages 486-496
Mobility Aware Routing Protocol Based on DIO Message for Low Power and Lossy Networks (Shridhar Sanshi, C. D. Jaidhar)....Pages 497-508
Boosting Convolutional Neural Networks Performance Based on FPGA Accelerator (Omran Al-Shamma, Mohammed Abdulraheem Fadhel, Rabab Alaa Hameed, Laith Alzubaidi, Jinglan Zhang)....Pages 509-517
Real-Time PCG Diagnosis Using FPGA (Mohammed Abdulraheem Fadhel, Omran Al-Shamma, Sameer Razzaq Oleiwi, Bahaa Hussein Taher, Laith Alzubaidi)....Pages 518-529
Cluster Center Initialization and Outlier Detection Based on Distance and Density for the K-Means Algorithm (Qi He, Zhenxiang Chen, Ke Ji, Lin Wang, Kun Ma, Chuan Zhao et al.)....Pages 530-539
A Novel Method for Retrieval of Remote Sensing Image Using Wavelet Transform and HOG (Minakshi N. Vharkte, Vijaya B. Musande)....Pages 540-549
Classification of Red Blood Cells in Sickle Cell Anemia Using Deep Convolutional Neural Network (Laith Alzubaidi, Omran Al-Shamma, Mohammed A. Fadhel, Laith Farhan, Jinglan Zhang)....Pages 550-559
Robust and Efficient Approach to Diagnose Sickle Cell Anemia in Blood (Laith Alzubaidi, Mohammed A. Fadhel, Omran Al-Shamma, Jinglan Zhang)....Pages 560-570
An Improved Classifier Based on Entropy and Deep Learning for Bug Priority Prediction (Madhu Kumari, V. B. Singh)....Pages 571-580
Turbo Coded STBC MIMO OFDM with DWT Based I/Q Balancing System (K. Sundar Srinivas, M. N. L. Kalyani, N. Mounika, Ch. Aruna Kumari)....Pages 581-589
A UML/MARTE Based Design Pattern for a Wireless Sensor Node (Raoudha Saida, Yessine Hadj Kacem, M. S. BenSaleh, Mohamed Abid)....Pages 590-599
ECC Based Encryption Algorithm for Lightweight Cryptography (Soumi Banerjee, Anita Patil)....Pages 600-609
Reduced Complexity Affine Projection Algorithm Based on Variable Projection Order and Multiple Sub Filter Approach (S. Radhika, A. Chandrasekar)....Pages 610-619
A Prototype Model of Hand Assistive System Useful for Hearing Impaired (J. Divya Udayan, Anupama K. Ingale, R. Hemalatha)....Pages 620-628
Towards Micro-expression Recognition Through Pyramid of Uniform Temporal Local Binary Pattern Features (Taoufik Ben Abdallah, Radhouane Guermazi, Mohamed Hammami)....Pages 629-640
Misbehavior Detection in C-ITS Using Deep Learning Approach (Pranav Kumar Singh, Manish Kumar Dash, Paritosh Mittal, Sunit Kumar Nandi, Sukumar Nandi)....Pages 641-652
Authorship Identification with Multi Sequence Word Selection Method (Mubin Shoukat Tamboli, Rajesh S. Prasad)....Pages 653-661
A Single Program Multiple Data Algorithm for Feature Selection (Bhabesh Chanduka, Tushaar Gangavarapu, C. D. Jaidhar)....Pages 662-672
Prosodic Feature Selection of Personality Traits for Job Interview Performance (Rohit Mishra, Santosh Kumar Barnwal, Shrikant Malviya, Prasoon Mishra, Uma Shanker Tiwary)....Pages 673-682
Hybrid Association Rule Miner Using Probabilistic Context-Free Grammar and Ant Colony Optimization for Rainfall Prediction (S. Saranyadevi, R. Murugeswari, S. Bathrinath, M. S. Sabitha)....Pages 683-695
Design of an Intelligent Cooperative Road Hazard Detection Persistent System (Islam Elleuch, Achraf Makni, Rafik Bouaziz)....Pages 696-707
Clustering Time-Series Data Generated by Smart Devices for Human Activity Recognition (R. Jothi)....Pages 708-716
A Priority-Based Ranking Approach for Maximizing the Earned Benefit in an Incentivized Social Network (Suman Banerjee, Mamata Jenamani, Dilip Kumar Pratihar, Abhinav Sirohi)....Pages 717-726
Analysis of Basic-SegNet Architecture with Variations in Training Options (Ganesh R. Padalkar, Madhuri B. Khambete)....Pages 727-735
CRIST900: A Fully-Labeled Natural Image Dataset for Multi-Operator Content Aware Image Retargeting (M. Abhayadev, T. Santha)....Pages 736-748
A Data Mining Approach to Predict Academic Performance of Students Using Ensemble Techniques (Samuel-Soma M. Ajibade, Nor Bahiah Ahmad, Siti Mariyam Shamsuddin)....Pages 749-760
A Late Acceptance Hill-Climbing Heuristic Algorithm for the Double Vehicle Routing Problem with Multiple Stacks and Heterogeneous Demand (André L. S. Souza, Jonatas B. C. Chagas, Puca H. V. Penna, Marcone J. F. Souza)....Pages 761-771
Evaluation of Advanced Analysis Method for Human Relationship Using Fuzzy Theory (Toshihiro Yoshizumi, Tomoo Sumida, Yasunori Shiono, Mitsuhiro Namekawa, Kensei Tsuchida)....Pages 772-782
Analysis of Overhead View Images at Intersection Using Machine Learning (Taisuke Hori, Mitsuhiro Namekawa, Syuya Kanagawa)....Pages 783-791
A New Design Prospective for User Specific Intelligent Control of Devices in a Smart Environment (Vaskar Deka, Shikhar Kumar Sarma)....Pages 792-802
Crime Information Improvement for Situation Awareness Based on Data Mining (Lucas Zanco Ladeira, Valdir Amancio Pereira Junior, Raphael Zanon Rodrigues, Leonardo Castro Botega)....Pages 803-812
A Novel Approach Towards Enhancing the Performance of Trust Based RPL Protocol in Internet of Things (Jayaram Hariharakrishnan, N. Bhalaji)....Pages 813-822
Fractional Order Extended Kalman Filter for Attitude Estimation (Nimmi Sharma, Elizabeth Rufus, Vinod Karar, Shashi Poddar)....Pages 823-832
Implementation of Robust Solid State Drive Controller Using LZ77 Compression and SHA-1 Encryption Technique (Amanda Kelly D’costa, K. P. Raksha, D. R. Vasanthi)....Pages 833-843
A Convolution Neural Network Based Classification Approach for Recognizing Traditional Foods of Bangladesh from Food Images (Nishat Tasnim, Md. Romyull Islam, Shaon Bhatta Shuvo)....Pages 844-852
An Efficient Outlier Detection Mechanism for RFID-Sensor Integrated MANET (Adarsh Kumar, Alok Aggarwal)....Pages 853-863
Design of Low Power SAR ADC with Two Different DAC Structure and Two Different SAR Logic Designs and Their Comparisons (Aruna Kumari Chirapangi, G. M. G. Madhuri, Praveen Kitti Burri, Naga Lakshmi Kalyani Movva)....Pages 864-874
Efficient Decision Support System on Agrometeorological Data (Abhishek Teli, A. Amith, K. Bhanu Kaushik, K. Gopala Krishna Vasanth, B. J. Sowmya, S. Seema)....Pages 875-890
Distributed Mining of Significant Frequent Colossal Closed Itemsets from Long Biological Dataset (Manjunath K. Vanahalli, Nagamma Patil)....Pages 891-902
Intelligent System for Weather Prediction (Vyom Unadkat, Sneh Gajiwala, Prachi Doshi, Mitchell D’silva)....Pages 903-911
A GPU-Based jDE Algorithm Applied to Continuous Unconstrained Optimization (Mateus Boiani, Gabriel Dominico, Rafael Stubs Parpinelli)....Pages 912-922
OP3DBFT: A Power and Performance Optimal 3D BFT NoC Architecture (Bheemappa Halavar, Basavaraj Talawar)....Pages 923-933
Comparative Analysis of Elliptic Curve Cryptography Based Lightweight Authentication Protocols for RFID-Sensor Integrated MANETs (Adarsh Kumar, Alok Aggarwal)....Pages 934-944
An FPGA Based Hardware Accelerator for Classification of Handwritten Digits (R. Gautham Sundar Ram, Nitin Chaturvedi, Sumeet Saurav, Sanjay Singh)....Pages 945-954
Selection of Optimal Game Engine by Using AHP Approach for Virtual Reality Fire Safety Training (El Mostafa Bourhim, Abdelghani Cherkaoui)....Pages 955-966
Feature Selection Using Fast Ensemble Learning for Network Intrusion Detection (Ujjwal Pasupulety, C. D. Adwaith, Suraj Hegde, Nagamma Patil)....Pages 967-977
An Embedded System for Watershed Based Hard Exudate Extraction (Vasanthi Satyananda, K. V. Narayanaswamy, Karibasappa)....Pages 978-987
Detection of Exudates from Fundus Images (Vasanthi Satyananda, K. V. Narayanaswamy, Karibasappa)....Pages 988-997
Intuitionistic Fuzzy Soft Aggregation Operator Based on Einstein Norms and Its Applications in Decision-Making (Rishu Arora)....Pages 998-1008
Parametric Similarity Measures on Linguistic Single-Valued Neutrosophic Sets with Application to Decision-Making Problems ( Nancy)....Pages 1009-1019
An SOA Design Patterns Recommendation System Based on Ontology (Karama Abdelhedi, Nadia Bouassidar)....Pages 1020-1030
Framework for Intelligent Software Defined Networking for Wired and Wireless Networks (Rakesh Kumar Ambhati, G. Selva Kumar, Y. Shashikant Chaudhari, Valluri Sarimela)....Pages 1031-1039
Devanagari Character Classification Using Capsule Network (Jeel Sukhadiya, Yashi Suba, Mitchell D’silva)....Pages 1040-1049
Lightweight Cipher Using GRP Bit Permutation and Tweak (Aruna Gawade, Narendra Shekokar)....Pages 1050-1059
A State-of-Art Review on Automatic Video Annotation Techniques (Krunal Randive, R. Mohan)....Pages 1060-1069
A Robust Speech Encryption System Based on DNA Addition and Chaotic Maps (R. Nagakrishnan, A. Revathi)....Pages 1070-1080
Efficient Energy Attentive and Fault Recognition Mechanism in Distributed Wireless Sensor Networks: A Review (Roshani Talmale, M. Nirupama Bhat, Nita Thakare)....Pages 1081-1092
Digital Color Documents Authentication Using QR Code Based on Digital Watermarking (Zinah Mohsin Arkah, Laith Alzubaidi, Ammar A. Ali, Ahmed Talib Abdulameer)....Pages 1093-1101
Comparative Study of Regression Models and Deep Learning Models for Insurance Cost Prediction (Aditya Shinde, Purva Raut)....Pages 1102-1111
Extending Borda Rule Under q-rung Orthopair Fuzzy Set for Multi-attribute Group Decision-Making (R. Krishankumar, S. Shyam, R. P. Nethra, S. Srivatsa, K. S. Ravichandran)....Pages 1112-1122
A Novel Approach for Operational Performance Based Mail Sorting Facility Layout Selection Using Grey Relational Analysis: A Case on India Speed Post Service Industry (S. M. Vadivel, A. H. Sequeira)....Pages 1123-1132
Recognition of Handwritten Meitei Mayek and English Alphabets Using Combination of Spatial Features (Sanasam Chanu Inunganbi, Prakash Choudhary)....Pages 1133-1142
A Self-adaptive Differential Evolution with Local Search Applied to Multimodal Optimization (Gabriel Dominico, Mateus Boiani, Rafael Stubs Parpinelli)....Pages 1143-1153
Back Matter ....Pages 1155-1158

Citation preview

Advances in Intelligent Systems and Computing 940

Ajith Abraham Aswani Kumar Cherukuri Patricia Melin Niketa Gandhi Editors

Intelligent Systems Design and Applications 18th International Conference on Intelligent Systems Design and Applications (ISDA 2018) held in Vellore, India, December 6–8, 2018, Volume 1

Advances in Intelligent Systems and Computing Volume 940

Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Nikhil R. Pal, Indian Statistical Institute, Kolkata, India Rafael Bello Perez, Faculty of Mathematics, Physics and Computing, Universidad Central de Las Villas, Santa Clara, Cuba Emilio S. Corchado, University of Salamanca, Salamanca, Spain Hani Hagras, Electronic Engineering, University of Essex, Colchester, UK László T. Kóczy, Department of Automation, Széchenyi István University, Gyor, Hungary Vladik Kreinovich, Department of Computer Science, University of Texas at El Paso, El Paso, TX, USA Chin-Teng Lin, Department of Electrical Engineering, National Chiao Tung University, Hsinchu, Taiwan Jie Lu, Faculty of Engineering and Information Technology, University of Technology Sydney, Sydney, NSW, Australia Patricia Melin, Graduate Program of Computer Science, Tijuana Institute of Technology, Tijuana, Mexico Nadia Nedjah, Department of Electronics Engineering, University of Rio de Janeiro, Rio de Janeiro, Brazil Ngoc Thanh Nguyen, Faculty of Computer Science and Management, Wrocław University of Technology, Wrocław, Poland Jun Wang, Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong

The series “Advances in Intelligent Systems and Computing” contains publications on theory, applications, and design methods of Intelligent Systems and Intelligent Computing. Virtually all disciplines such as engineering, natural sciences, computer and information science, ICT, economics, business, e-commerce, environment, healthcare, life science are covered. The list of topics spans all the areas of modern intelligent systems and computing such as: computational intelligence, soft computing including neural networks, fuzzy systems, evolutionary computing and the fusion of these paradigms, social intelligence, ambient intelligence, computational neuroscience, artificial life, virtual worlds and society, cognitive science and systems, Perception and Vision, DNA and immune based systems, self-organizing and adaptive systems, e-Learning and teaching, human-centered and human-centric computing, recommender systems, intelligent control, robotics and mechatronics including human-machine teaming, knowledge-based paradigms, learning paradigms, machine ethics, intelligent data analysis, knowledge management, intelligent agents, intelligent decision making and support, intelligent network security, trust management, interactive entertainment, Web intelligence and multimedia. The publications within “Advances in Intelligent Systems and Computing” are primarily proceedings of important conferences, symposia and congresses. They cover significant recent developments in the field, both of a foundational and applicable character. An important characteristic feature of the series is the short publication time and world-wide distribution. This permits a rapid and broad dissemination of research results. ** Indexing: The books of this series are submitted to ISI Proceedings, EI-Compendex, DBLP, SCOPUS, Google Scholar and Springerlink ** More information about this series at http://www.springer.com/series/11156

Ajith Abraham Aswani Kumar Cherukuri Patricia Melin Niketa Gandhi •

•

•

Editors

Intelligent Systems Design and Applications 18th International Conference on Intelligent Systems Design and Applications (ISDA 2018) held in Vellore, India, December 6–8, 2018, Volume 1

123

Editors Ajith Abraham Machine Intelligence Research Labs Auburn, WA, USA Patricia Melin Tijuana Institute of Technology Tijuana, Mexico

Aswani Kumar Cherukuri School of Information Technology and Engineering Vellore Institute of Technology Vellore, Tamil Nadu, India Niketa Gandhi Machine Intelligence Research Labs Auburn, WA, USA

ISSN 2194-5357 ISSN 2194-5365 (electronic) Advances in Intelligent Systems and Computing ISBN 978-3-030-16656-4 ISBN 978-3-030-16657-1 (eBook) https://doi.org/10.1007/978-3-030-16657-1 Library of Congress Control Number: 2019936140 © Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved 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

Welcome to the Proceedings of the Joint Conferences on 18th International Conference on Intelligent Systems Design and Applications (ISDA) and 10th World Congress on Nature and Biologically Inspired Computing (NaBIC), which is held in VIT University, India, during December 6–8, 2018. ISDA - NaBIC 2018 is jointly organized by the VIT University, India, and Machine Intelligence Research Labs (MIR Labs), USA. ISDA - NaBIC 2018 brings together researchers, engineers, developers, and practitioners from academia and industry working in all interdisciplinary areas of intelligent systems, nature-inspired computing, big data analytics, real-world applications and to exchange and cross-fertilize their ideas. The themes of the contributions and scientific sessions range from theories to applications, reflecting a wide spectrum of the coverage of intelligent systems and computational intelligence areas. ISDA 2018 received submissions from 30 countries, and each paper was reviewed by at least five reviewers in a standard peer-review process. Based on the recommendation by five independent referees, finally 189 papers were accepted for ISDA 2018 (acceptance rate of 48%). NaBIC 2018 received submissions from 11 countries, and each paper was reviewed by at least five reviewers in a standard peer-review process. Based on the recommendation by five independent referees, finally about 23 papers were accepted for NaBIC 2018 (acceptance rate of 37%). Conference proceedings are published by Springer Verlag, Advances in Intelligent Systems and Computing Series. Many people have collaborated and worked hard to produce the successful ISDA - NaBIC 2018 conference. First, we would like to thank all the authors for submitting their papers to the conference, for their presentations and discussions during the conference. Our thanks go to program committee members and reviewers, who carried out the most difficult work by carefully evaluating the submitted papers. Our special thanks to Raija Halonen, University of Oulu, Finland, Junzo Watada, Universiti Teknologi Petronas, Malaysia, and Nelishia Pillay, University of Pretoria, South Africa, for the exciting plenary talks. We express our

v

vi

Preface

sincere thanks to the session chairs and organizing committee chairs for helping us to formulate a rich technical program. Enjoy reading the articles! Ajith Abraham Aswani Kumar Cherukuri General Chairs Patricia Melin Emilio Corchado Florin Popentiu Vladicescu Ana Maria Madureira Program Chairs

Organization

Chief Patron G. Viswanathan (Chancellor)

Vellore Institute of Technology

Patrons Sanakar Viswanathan (Vice President) Sekar Viswanathan (Vice President) G. V. Selvam (Vice President)

Vellore Institute of Technology, Vellore Vellore Institute of Technology, Vellore Vellore Institute of Technology, Vellore

Advisors Anand A. Samuel (Vice Chancellor) S. Narayanan (Pro-vice Chancellor)

Vellore Institute of Technology, Vellore Vellore Institute of Technology, Vellore

General Chairs Ajith Abraham Aswani Kumar Cherukuri

Machine Intelligence Research Labs (MIR Labs), USA Vellore Institute of Technology, India

Program Chairs Patricia Melin Emilio Corchado Florin Popenţiu Vlădicescu Ana Maria Madureira

Tijuana Institute of Technology, Mexico University of Salamanca, Spain University Politehnica of Bucharest, Romania Instituto Superior de Engenharia do Porto, Portugal

vii

viii

Organization

International Advisory Board Albert Zomaya Bruno Apolloni Imre J. Rudas Janusz Kacprzyk Marina Gavrilova Patrick Siarry Ronald Yager Sebastian Ventura Vincenzo Piuri Francisco Herrera Sankar Kumar Pal

University of Sydney, Australia University of Milano, Italy Óbuda University, Hungary Polish Academy of Sciences, Poland University of Calgary, Canada Université Paris-Est Créteil, France Iona College, USA University of Cordoba, Spain Universita’ degli Studi di Milano, Italy University of Granada, Spain ISI, Kolkata, India

Publication Chair Niketa Gandhi

Machine Intelligence Research Labs (MIR Labs), USA

Web Master Kun Ma

Jinan University, China

Publicity Committee Mayur Rahul Sanju Tiwari

C.S.J.M. University, Kanpur, India National Institute of Technology, Kurukshetra, Haryana, India

Local Organizing Committee Advisory Committee R. Saravanan Jasmine Norman Sree Dharinya T. Ramkumar Agilandeswari Lakshmi Priya Ajit Kumar Santra Bimal Kumar Ray K. Ganesan Hari Ram Vishwakarma

SCOPE Department Department Department Department Department SITE SITE SITE SITE

of of of of of

IT, SITE SSE, SITE CACM, SITE DC, SITE MM, SITE

Organization

Daphne Lopez Shantharajah Subha Organizing Chair E. Sathiyamoorthy Finance Committee S. Prasanna (In-charge) J. Karthikeyan Event Management Committee R. Srinivasa Perumal (In-charge) R. Sujatha S. L. Arthy S. Siva Ramakrishnan S. Jayakumar G. Kavitha M. Priya Hospitality and Guest Care Committee G. Jagadeesh (In-charge) P. Thanapal J. Prabu Vijaya Anand J. Gitanjali Printing Committee N. Deepa (In-charge) U. Rahamathunnisa P. Jayalakshmi L. B. Krithika Registration Committee N. Mythili (In-charge) Divya Udayan K. Brindha S. Sudha M. Deepa K. Shanthi

ix

SITE SITE SITE

x

Organization

Technical Program Committee Abdelkrim Haqiq Alberto Cano Ali Yakoob Amit Kumar Shukla Amparo Fuster-Sabater Antonio J. Tallón-Ballesteros Angel Garcia-Baños Anjali Chandavale Aswani Cherukuri Atta Rahman Azah Kamilah-Muda Aurora Ramírez Cesar Hervas Daniela Zaharie Denis Felipe Durai Raj Vincent Elizabeth Goldbarg Francisco Chicano Frantisek Zboril Giovanna Castellano Givanaldo Rocha de Souza Gregorio Sainz-Palmero Igor Medeiros Isaac Chairez Isabel S. Jesus Ishwarya Srinivasan Jagadeesh Kakarla Jaroslav Rozman Javier Ferrer Jerry Chun-Wei Lin

GREENTIC, FST, Hassan First University, Settat, Morocco University of Córdoba, Spain University of Babylon, Iraq South Asian University, Delhi, India Institute of Physical and Information Technologies (CSIC), Spain University of Seville, Spain Universidad del Valle/Cali, Colombia Dr. Vishwanath Karad MIT World Peace University, India Vellore Institute of Technology, India University of Dammam, Dammam, Saudi Arabia UTeM, Malaysia University of Córdoba, Spain Martínez, University of Córdoba, Spain West University of Timisoara, Romania Federal University of Rio Grande do Norte, Brazil Vellore Institute of Technology, India Universidade Federal do Rio Grande do Norte, Brazil Universidad de Málaga, Spain Brno University of Technology, Czechia Università degli Studi di Bari Aldo Moro, Italy Federal University of Rio Grande do Norte, Brazil Universidad de Valladolid, Spain Federal University of Rio Grande do Norte, Brazil Instituto Politécnico Nacional, Mexico Instituto Superior de Engenharia do Porto, Portugal Vellore Institute of Technology, India IIITDM Kancheepuram, Chennai, India Brno University of Technology, Czech Republic University of Málaga, Spain Western Norway University of Applied Sciences (HVL), Bergen, Norway

Organization

Jesus Alcala-Fdez José Everardo Bessa Maia José Raúl Romero Jose Tenreiro Machado Kathiravan Srinivasan Kaushik Das Sharma Lin Wang Mario Giovanni C. A. Cimino Martin Hruby Matheus Menezes Mohammad Shojafar Nadesh Rk Niketa Gandhi Oscar Castillo Ozgur Koray Sahingoz P. E. S. N. Krishna Prasad Paolo Buono Patrick Siarry Prabukumar Manoharan Pranab Muhuri Radu-Emil Precup Raghavendra Kumar Chunduri Ricardo Tanscheit Saravanakumar Kandasamy Simone Ludwig Sustek Martin Thatiana C. N. Souza Thomas Hanne

Varun Ojha

xi

University of Granada, Spain State University of Ceará, Brazil University of Córdoba, Spain ISEP, Portugal Vellore Institute of Technology, India University of Calcutta, India Jinan University, China University of Pisa, Italy Brno University of Technology, Czech Republic Universidade Federal Rural do Semi-Árido, Brazil Sapienza University of Rome, Italy Vellore Institute of Technology, India Machine Intelligence Research Labs (MIR Labs), USA Tijuana Institute of Technology, Tijuana Istanbul Kultur University, Turkey S V College of Engineering, Tirupati, India Università degli Studi di Bari Aldo Moro, Italy Université Paris-Est Créteil, France Vellore Institute of Technology, India South Asian University, Delhi, India Politehnica University of Timisoara, Romania Vellore Institute of Technology, India PUC-Rio, Brazil Vellore Institute of Technology, India North Dakota State University, USA Brno University of Technology, Czechia Federal University Rural Semi-Arid, Brazil University of Applied Sciences Northwestern Switzerland, Switzerland Swiss Federal Institute of Technology, Switzerland

Contents

A Study of Multi-space Search Optimization . . . . . . . . . . . . . . . . . . . . . Derrick Beckedahl, Andreas Nel, and Nelishia Pillay Clinical Decision Support System for Neuro-Degenerative Disorders: An Optimal Feature Selective Classifier and Identification of Predictor Markers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lokeswari Venkataramana, Shomona Gracia Jacob, S. Saraswathi, and R. Athilakshmi Favoring the k-Means Algorithm with Initialization Methods . . . . . . . . Anderson Francisco de Oliveira and Maria do Carmo Nicoletti A Novel Design and Implementation of 8-Bit and 16-Bit Hybrid ALU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Suhas B. Shirol, S. Ramakrishna, and Rajashekar B. Shettar Authentication Scheme Using Sparse Matrix in Cloud Computing . . . . Sunita Meena, Shivani Kapur, Vipin C. Dhobal, and Subhrat Kr. Sethi A Thermal Imaging Based Classification of Affective States Using Multiclass SVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. M. Naveen Kumar and G. Shivakumar Multidimensional Crime Dataset Analysis . . . . . . . . . . . . . . . . . . . . . . . Prerna Kapoor and Prem Kumar Singh

1

10

21

32 43

53 64

AKCSS: An Asymmetric Key Cryptography Based on Secret Sharing in Mobile Ad Hoc Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R. Preethi and M. Sughasiny

73

API Call Based Malware Detection Approach Using Recurrent Neural Network—LSTM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . J. Mathew and M. A. Ajay Kumara

87

xiii

xiv

Contents

Review and Analysis of Zero, One and Few Shot Learning Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 Suvarna Kadam and Vinay Vaidya Simulation of Friction Stir Welding of Aluminium Alloy AA5052 – Tailor Welded Blanks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 M. Arun Siddharth, R. Padmanaban, and R. Vaira Vignesh Information Systems Success: Extending the Theoretical Model from IT Business Value Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Thanh D. Nguyen Towards an Automatic Detection of Sensitive Information in Mongo Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 Houyem Heni and Faiez Gargouri Business Growth Using Open Source e-Commerce and ERP in Small Business . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 Valtteri Kujala and Raija Halonen Directional Multiscale Feature Extraction for Biomedical Image Indexing and Retrieval Using Contourlet Transform . . . . . . . . . . . . . . . 159 Amita A. Shinde, Amol D. Rahulkar, and Chetankumar Y. Patil Link Quality and QoE Aware Predictive Vertical Handoff Mechanism for Video Streaming in Urban VANET . . . . . . . . . . . . . . . . . . . . . . . . . 170 Emna Bouzid Smida, Sonia Gaied Fantar, and Habib Youssef Performance Comparison of PID and ANFIS Controller for Stabilization of x and x-y Inverted Pendulums . . . . . . . . . . . . . . . . . 182 Ishan Chawla, Vikram Chopra, and Ashish Singla Modeling Hybrid Indicators for Stock Index Prediction . . . . . . . . . . . . . 193 R. Arjun and K. R. Suprabha XOR Encryption Techniques of Video Steganography: A Comparative Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 Namrata Singh Intention to Use M–Banking: The Role of E–WOM . . . . . . . . . . . . . . . 215 Thanh D. Nguyen, Thy Q. L. Nguyen, Thi V. Nguyen, and Tung D. Tran Activity Gesture Recognition on Kinect Sensor Using Convolutional Neural Networks and FastDTW for the MSRC-12 Dataset . . . . . . . . . . 230 Miguel Pfitscher, Daniel Welfer, Marco Antonio de Souza Leite Cuadros, and Daniel Fernando Tello Gamarra Plug in Electric Vehicle-Wind Integrated Multi-area Automatic Generation Control Tuned by Intelligent Water Drops Algorithm . . . . . 240 Subhranshu Sekhar Pati, Aurobindo Behera, and Tapas Kumar Panigrahi

Contents

xv

Design of Time-Frequency Localized Filter Bank Using Modified Particle Swarm Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 Swati P. Madhe, Amol D. Rahulkar, and Raghunath S. Holambe Development of Low-Cost Real-Time Driver Drowsiness Detection System Using Eye Centre Tracking and Dynamic Thresholding . . . . . . 262 Fuzail Khan and Sandeep Sharma A Hybrid Entropy Based Method Using Gaussian Kernel for Retinal Blood Vessel Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . 272 N. K. Adhish, R. Rajesh, and T. M. Thasleema Precision Crop Protection Using Wireless Sensor Network . . . . . . . . . . 280 R. Radha, Amit Kumar Tyagi, K. Kathiravan, and G. Staflin Betzy Deep Learning Based Approach for Classification and Detection of Papaya Leaf Diseases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 Rathan Kumar Veeraballi, Muni Sankar Nagugari, Chandra Sekhara Rao Annavarapu, and Eswar Varma Gownipuram Three-Materials Image Recover from Value Range Projection Data . . . 303 Chuanlin Liu, Amit Yadav, Asif Khan, Jing Zou, and Weizhen Hu Multiple Criteria Fake Reviews Detection Using Belief Function Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Malika Ben Khalifa, Zied Elouedi, and Eric Lefèvre Improved Logistic Regression Approach in Feature Selection for EHR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325 Shreyal Gajare and Shilpa Sonawani Background Modeling Using Deep-Variational Autoencoder . . . . . . . . . 335 Midhula Vijayan and R. Mohan Sewage Sludge Removal Method Through Arm-Axis by Machine Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 M. Gobinath and S. Malathi K-Nearest Neighbors Under Possibility Framework with Optimizing Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354 Sarra Saied and Zied Elouedi A Visual Spelling System Using SSVEP Based Hybrid Brain Computer Interface with Video-Oculography . . . . . . . . . . . . . . . . . . . . . 365 D. Saravanakumar and M. Ramasubba Reddy QBEECH: Multi-hop Clustering of Cognitive Based Sensor Nodes in the Administration of Queen Nodes . . . . . . . . . . . . . . . . . . . . . . . . . . 376 Souvik Kundu, Srividhya Karthikeyan, and A. Karthikeyan

xvi

Contents

Perceive Core Logical Blocks of a C Program Automatically for Source Code Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386 Pallavi Ahire and Jibi Abraham Asymmetric Key Cryptosystem and Digital Signature Algorithm Built on Discrete Logarithm Problem (DLP) . . . . . . . . . . . . . . . . . . . . . . . . . . 401 Ashish Kumar, Jagadeesh Kakarla, and Muzzammil Hussain A Study on Big Cancer Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411 Sabuzima Nayak and Ripon Patgiri Food Monitoring Using Adaptive Naïve Bayes Prediction in IoT . . . . . . 424 Pramod D. Ganjewar, Selvaraj Barani, Sanjeev J. Wagh, and Santosh S. Sonavane Mixed Credit Scoring Model of Logistic Regression and Evidence Weight in the Background of Big Data . . . . . . . . . . . . . . . . . . . . . . . . . 435 Keqin Chen, Kun Zhu, Yixin Meng, Amit Yadav, and Asif Khan A Model for Identifying Historical Landmarks of Bangladesh from Image Content Using a Depth-Wise Convolutional Neural Network . . . 444 Afsana Ahsan Jeny, Masum Shah Junayed, Syeda Tanjila Atik, and Sazzad Mahamd M2U2: Multifactor Mobile Based Unique User Authentication Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455 Rachit Bhalla and N. Jeyanthi Generation of Image Caption Using CNN-LSTM Based Approach . . . . 465 S. Aravindkumar, P. Varalakshmi, and M. Hemalatha ADABA: An Algorithm to Improve the Parallel Search in Competitive Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475 Lídia Bononi Paiva Tomaz and Rita Maria Silva Julia A Novel Approach to Solve Class Imbalance Problem Using Noise Filter Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 486 Gillala Rekha, Amit Kumar Tyagi, and V. Krishna Reddy Mobility Aware Routing Protocol Based on DIO Message for Low Power and Lossy Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 497 Shridhar Sanshi and C. D. Jaidhar Boosting Convolutional Neural Networks Performance Based on FPGA Accelerator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509 Omran Al-Shamma, Mohammed Abdulraheem Fadhel, Rabab Alaa Hameed, Laith Alzubaidi, and Jinglan Zhang

Contents

xvii

Real-Time PCG Diagnosis Using FPGA . . . . . . . . . . . . . . . . . . . . . . . . . 518 Mohammed Abdulraheem Fadhel, Omran Al-Shamma, Sameer Razzaq Oleiwi, Bahaa Hussein Taher, and Laith Alzubaidi Cluster Center Initialization and Outlier Detection Based on Distance and Density for the K-Means Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 530 Qi He, Zhenxiang Chen, Ke Ji, Lin Wang, Kun Ma, Chuan Zhao, and Yuliang Shi A Novel Method for Retrieval of Remote Sensing Image Using Wavelet Transform and HOG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 540 Minakshi N. Vharkte and Vijaya B. Musande Classification of Red Blood Cells in Sickle Cell Anemia Using Deep Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550 Laith Alzubaidi, Omran Al-Shamma, Mohammed A. Fadhel, Laith Farhan, and Jinglan Zhang Robust and Efficient Approach to Diagnose Sickle Cell Anemia in Blood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560 Laith Alzubaidi, Mohammed A. Fadhel, Omran Al-Shamma, and Jinglan Zhang An Improved Classifier Based on Entropy and Deep Learning for Bug Priority Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571 Madhu Kumari and V. B. Singh Turbo Coded STBC MIMO OFDM with DWT Based I/Q Balancing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581 K. Sundar Srinivas, M. N. L. Kalyani, N. Mounika, and Ch. Aruna Kumari A UML/MARTE Based Design Pattern for a Wireless Sensor Node . . . 590 Raoudha Saida, Yessine Hadj Kacem, M. S. BenSaleh, and Mohamed Abid ECC Based Encryption Algorithm for Lightweight Cryptography . . . . . 600 Soumi Banerjee and Anita Patil Reduced Complexity Affine Projection Algorithm Based on Variable Projection Order and Multiple Sub Filter Approach . . . . . . . . . . . . . . . 610 S. Radhika and A. Chandrasekar A Prototype Model of Hand Assistive System Useful for Hearing Impaired . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 620 J. Divya Udayan, Anupama K. Ingale, and R. Hemalatha Towards Micro-expression Recognition Through Pyramid of Uniform Temporal Local Binary Pattern Features . . . . . . . . . . . . . . . . . . . . . . . . 629 Taoufik Ben Abdallah, Radhouane Guermazi, and Mohamed Hammami

xviii

Contents

Misbehavior Detection in C-ITS Using Deep Learning Approach . . . . . 641 Pranav Kumar Singh, Manish Kumar Dash, Paritosh Mittal, Sunit Kumar Nandi, and Sukumar Nandi Authorship Identification with Multi Sequence Word Selection Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653 Mubin Shoukat Tamboli and Rajesh S. Prasad A Single Program Multiple Data Algorithm for Feature Selection . . . . . 662 Bhabesh Chanduka, Tushaar Gangavarapu, and C. D. Jaidhar Prosodic Feature Selection of Personality Traits for Job Interview Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673 Rohit Mishra, Santosh Kumar Barnwal, Shrikant Malviya, Prasoon Mishra, and Uma Shanker Tiwary Hybrid Association Rule Miner Using Probabilistic Context-Free Grammar and Ant Colony Optimization for Rainfall Prediction . . . . . . 683 S. Saranyadevi, R. Murugeswari, S. Bathrinath, and M. S. Sabitha Design of an Intelligent Cooperative Road Hazard Detection Persistent System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696 Islam Elleuch, Achraf Makni, and Rafik Bouaziz Clustering Time-Series Data Generated by Smart Devices for Human Activity Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 708 R. Jothi A Priority-Based Ranking Approach for Maximizing the Earned Benefit in an Incentivized Social Network . . . . . . . . . . . . . . . . . . . . . . . 717 Suman Banerjee, Mamata Jenamani, Dilip Kumar Pratihar, and Abhinav Sirohi Analysis of Basic-SegNet Architecture with Variations in Training Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 727 Ganesh R. Padalkar and Madhuri B. Khambete CRIST900: A Fully-Labeled Natural Image Dataset for Multi-Operator Content Aware Image Retargeting . . . . . . . . . . . . . 736 M. Abhayadev and T. Santha A Data Mining Approach to Predict Academic Performance of Students Using Ensemble Techniques . . . . . . . . . . . . . . . . . . . . . . . . . 749 Samuel-Soma M. Ajibade, Nor Bahiah Ahmad, and Siti Mariyam Shamsuddin

Contents

xix

A Late Acceptance Hill-Climbing Heuristic Algorithm for the Double Vehicle Routing Problem with Multiple Stacks and Heterogeneous Demand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 761 André L. S. Souza, Jonatas B. C. Chagas, Puca H. V. Penna, and Marcone J. F. Souza Evaluation of Advanced Analysis Method for Human Relationship Using Fuzzy Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 772 Toshihiro Yoshizumi, Tomoo Sumida, Yasunori Shiono, Mitsuhiro Namekawa, and Kensei Tsuchida Analysis of Overhead View Images at Intersection Using Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783 Taisuke Hori, Mitsuhiro Namekawa, and Syuya Kanagawa A New Design Prospective for User Specific Intelligent Control of Devices in a Smart Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 792 Vaskar Deka and Shikhar Kumar Sarma Crime Information Improvement for Situation Awareness Based on Data Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 803 Lucas Zanco Ladeira, Valdir Amancio Pereira Junior, Raphael Zanon Rodrigues, and Leonardo Castro Botega A Novel Approach Towards Enhancing the Performance of Trust Based RPL Protocol in Internet of Things . . . . . . . . . . . . . . . . 813 Jayaram Hariharakrishnan and N. Bhalaji Fractional Order Extended Kalman Filter for Attitude Estimation . . . . 823 Nimmi Sharma, Elizabeth Rufus, Vinod Karar, and Shashi Poddar Implementation of Robust Solid State Drive Controller Using LZ77 Compression and SHA-1 Encryption Technique . . . . . . . . . . . . . . . . . . 833 Amanda Kelly D’costa, K. P. Raksha, and D. R. Vasanthi A Convolution Neural Network Based Classification Approach for Recognizing Traditional Foods of Bangladesh from Food Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 844 Nishat Tasnim, Md. Romyull Islam, and Shaon Bhatta Shuvo An Efficient Outlier Detection Mechanism for RFID-Sensor Integrated MANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 853 Adarsh Kumar and Alok Aggarwal Design of Low Power SAR ADC with Two Different DAC Structure and Two Different SAR Logic Designs and Their Comparisons . . . . . . . 864 Aruna Kumari Chirapangi, G. M. G. Madhuri, Praveen Kitti Burri, and Naga Lakshmi Kalyani Movva

xx

Contents

Efficient Decision Support System on Agrometeorological Data . . . . . . . 875 Abhishek Teli, A. Amith, K. Bhanu Kaushik, K. Gopala Krishna Vasanth, B. J. Sowmya, and S. Seema Distributed Mining of Significant Frequent Colossal Closed Itemsets from Long Biological Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 891 Manjunath K. Vanahalli and Nagamma Patil Intelligent System for Weather Prediction . . . . . . . . . . . . . . . . . . . . . . . 903 Vyom Unadkat, Sneh Gajiwala, Prachi Doshi, and Mitchell D’silva A GPU-Based jDE Algorithm Applied to Continuous Unconstrained Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 912 Mateus Boiani, Gabriel Dominico, and Rafael Stubs Parpinelli OP3DBFT: A Power and Performance Optimal 3D BFT NoC Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 923 Bheemappa Halavar and Basavaraj Talawar Comparative Analysis of Elliptic Curve Cryptography Based Lightweight Authentication Protocols for RFID-Sensor Integrated MANETs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 934 Adarsh Kumar and Alok Aggarwal An FPGA Based Hardware Accelerator for Classification of Handwritten Digits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 945 R. Gautham Sundar Ram, Nitin Chaturvedi, Sumeet Saurav, and Sanjay Singh Selection of Optimal Game Engine by Using AHP Approach for Virtual Reality Fire Safety Training . . . . . . . . . . . . . . . . . . . . . . . . . 955 El Mostafa Bourhim and Abdelghani Cherkaoui Feature Selection Using Fast Ensemble Learning for Network Intrusion Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 967 Ujjwal Pasupulety, C. D. Adwaith, Suraj Hegde, and Nagamma Patil An Embedded System for Watershed Based Hard Exudate Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 978 Vasanthi Satyananda, K. V. Narayanaswamy, and Karibasappa Detection of Exudates from Fundus Images . . . . . . . . . . . . . . . . . . . . . . 988 Vasanthi Satyananda, K. V. Narayanaswamy, and Karibasappa Intuitionistic Fuzzy Soft Aggregation Operator Based on Einstein Norms and Its Applications in Decision-Making . . . . . . . . . . . . . . . . . . 998 Rishu Arora

Contents

xxi

Parametric Similarity Measures on Linguistic Single-Valued Neutrosophic Sets with Application to Decision-Making Problems . . . . . 1009 Nancy An SOA Design Patterns Recommendation System Based on Ontology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1020 Karama Abdelhedi and Nadia Bouassidar Framework for Intelligent Software Defined Networking for Wired and Wireless Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1031 Rakesh Kumar Ambhati, G. Selva Kumar, Y. Shashikant Chaudhari, and Valluri Sarimela Devanagari Character Classification Using Capsule Network . . . . . . . . 1040 Jeel Sukhadiya, Yashi Suba, and Mitchell D’silva Lightweight Cipher Using GRP Bit Permutation and Tweak . . . . . . . . . 1050 Aruna Gawade and Narendra Shekokar A State-of-Art Review on Automatic Video Annotation Techniques . . . . 1060 Krunal Randive and R. Mohan A Robust Speech Encryption System Based on DNA Addition and Chaotic Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1070 R. Nagakrishnan and A. Revathi Efficient Energy Attentive and Fault Recognition Mechanism in Distributed Wireless Sensor Networks: A Review . . . . . . . . . . . . . . . 1081 Roshani Talmale, M. Nirupama Bhat, and Nita Thakare Digital Color Documents Authentication Using QR Code Based on Digital Watermarking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1093 Zinah Mohsin Arkah, Laith Alzubaidi, Ammar A. Ali, and Ahmed Talib Abdulameer Comparative Study of Regression Models and Deep Learning Models for Insurance Cost Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1102 Aditya Shinde and Purva Raut Extending Borda Rule Under q-rung Orthopair Fuzzy Set for Multi-attribute Group Decision-Making . . . . . . . . . . . . . . . . . . . . . . 1112 R. Krishankumar, S. Shyam, R. P. Nethra, S. Srivatsa, and K. S. Ravichandran A Novel Approach for Operational Performance Based Mail Sorting Facility Layout Selection Using Grey Relational Analysis: A Case on India Speed Post Service Industry . . . . . . . . . . . . . . . . . . . . 1123 S. M. Vadivel and A. H. Sequeira

xxii

Contents

Recognition of Handwritten Meitei Mayek and English Alphabets Using Combination of Spatial Features . . . . . . . . . . . . . . . . . . . . . . . . . 1133 Sanasam Chanu Inunganbi and Prakash Choudhary A Self-adaptive Differential Evolution with Local Search Applied to Multimodal Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1143 Gabriel Dominico, Mateus Boiani, and Rafael Stubs Parpinelli Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1155

A Study of Multi-space Search Optimization Derrick Beckedahl, Andreas Nel, and Nelishia Pillay(B) Department of Computer Science, University of Pretoria, Pretoria, South Africa [email protected], [email protected], [email protected]

Abstract. Traditionally search algorithms have explored a solution space to solve optimization problems. However, as the field has advanced, in order to overcome the challenges posed by searching the solution space directly such as premature convergence, search has been applied to different spaces. These include genetic programming which explores the program space, cultural learning and semantic genetic programming which work in the belief and behavioural spaces respectively and hyperheuristics which explores a heuristic space. In solving a problem one of these spaces is usually explored. The research presented in this paper forms part of an initiative aimed at deriving n-space search algorithms that explore more than one space to solve a problem. As a starting point this paper focuses on 2-space search algorithms. The paper presents two models for exploring search across spaces, namely, the concurrent model and the sequential model. The application of these models is then illustrated, the concurrent model is applied to solving the one-dimensional packing problem and the sequential model is applied to classification, namely, network instrusion detection and weather prediction. The models have produced good results for both problem domains, illustrating the potential of multi-space search. Future work will examine extending these algorithms to n-space search algorithms with n greater than 2 as well as additional models for combining the search across spaces.

Keywords: Multi-space search Program space

1

· Solution space · Heuristic space ·

Introduction

Search techniques such as metaheuristics traditionally work in the solution to space to find a solution to a problem. For example, when applying a genetic algorithm to solving the travelling salesman problem this multipoint search is applied to the space of potential routes that can be traversed by the salesman to identify the optimial route. However, solution space search is not without challenges, the most prominent being premature convergence which results in the search technique getting stuck at a local optimum. To alleviate such challenges alternative spaces have been explored. Genetic programming explores the c Springer Nature Switzerland AG 2020  A. Abraham et al. (Eds.): ISDA 2018, AISC 940, pp. 1–9, 2020. https://doi.org/10.1007/978-3-030-16657-1_1

2

D. Beckedahl et al.

program space rather than a solution space to identify a program that produces the optimal solution [2]. Hyper-heuristics explore the heuristic space rather than a solution space, with the heuristic space mapping to the solution space to produce a solution [7]. Similarly, cultural algorithms work in the belief space [9] and semantic genetic programming in the behavioural space [3]. Other spaces have also emerged in an attempt to improve the application of search techniques, for example the process of determining the parameters of search algorithms is in itself an optimization problem. To solve this problem the design space is searched. Generally, in solving a problem a single space is searched, for example either the solution or the heuristic space. In this paper we examine applying a search algorithm over more than one space. The study illustrates the potential of 2space multi-space search and future work will extend this idea to n spaces. The study conducted by Qu et al. [8] is the only study to the knowledge of the authors’ that explores more than one space. In this study search is conducted in the solution and heuristic space to solve university timetabling problems. Local search is applied to the heuristic and solution spaces. The study compares interleaving the search between both spaces and the sequential search of both spaces in solving this problem. Interleaving was found to be more effective. In the study presented in this paper two models, namely, the concurrent model and sequential model, are presented to perform search over two search spaces. Evolutionary algorithms are used to explore both the spaces. The application of these models is illustrated by applying them to solving the one dimensional bin packing problem and two classification problems, namely, network intrusion detection and weather prediction. The following section describes the multi-space search approach and the two models. Details of the multi-space search applied to the one dimensional bin packing problem and classification are given in Sects. 3 and 4 respectively. Section 5 presents the experimental setup for evaluating the multi-space search. The performance of the multi-space search for both domains is discussed in Sect. 6. The paper concludes by summarizing the findings of the study and providing an overview of future extensions of the work given the potential of multi-space search illustrated in this study.

2

Multi-space Search Approach

The study presented in this paper proposes the idea of n-space search with this study looking specifically at 2-space search as a starting point. Two models for combining the search across spaces are proposed in this study, namely, the concurrent model and the sequential model. The concurrent model explores more than one space simultaneously. In this study the solution and heuristic spaces are explored. The sequential model explores two spaces sequentially by hybridizing search techniques. In this study the sequential model is illustrated by combining search across the program and heuristic spaces. Different search models are applicable to different problem domains. The concurrent model is presented in Sect. 2.1 and the sequential model in Sect. 2.2.

A Study of Multi-space Search Optimization

2.1

3

Concurrent Model

The heuristic and solution space are explored simultaneously. A genetic algorithm is used to search across both spaces. Each chromosome is comprised of low-level construction heuristics [7] for a creating solution to the problem as well as move operators that are applied to the partial solutions created by the lowlevel construction heuristics. Each heuristic and move operator is represented by a character in the chromosome. Each chromosome is created by randomly selecting characters representing the heuristics and move operators. The length of each chromosome is maintained within a specified maximum length. The fitness function used is problem dependent. Offspring for each generation are created by applying crossover and mutation to parents selected using tournament selection. The genetic algorithm terminates after a set number of generations. 2.2

Sequential Model

In the sequential model search techniques are hybridized to explore the different spaces sequentially. In this study two spaces, namely, the program space and heuristic space, are explored sequentially. Genetic programming is applied to search the program space and genetic algorithms the heuristic space. The program space is explored using a generational genetic programming algorithm. Each element of the population is a parse tree representing a program. The tournament selection method is used to choose parents. One of two versions of tournament selection can be used for a particular implementation. The first is the standard tournament selection introduced by Koza [2]. The second tournament selection is the double tournament selector introduced by Luke and Panait [5]. This version of tournament selection performs two rounds of the tournament. In the first round parents are selected according to fitness. The selected parents are then subject to a second round which choses a parent based on the size of the parse tree with preference given to smaller trees. The fitness function used is problem dependent. Crossover and mutation are used to create offspring of each successive generation. The algorithm terminates after a set number of generations. The best performing individuals identified by the program space search then form input into the heuristic space search. The genetic programming algorithm is run n times with a different random number generator seed on each run. These individuals represent heuristics in the heuristic space. A genetic algorithm is used to explore the heuristic space. Each chromosome is composed of characters representing the best programs evolved from the program space search. The characters are randomly selected to create chromosomes of variable length within a specified maximum length. The fitness function is problem dependent. Tournament selection is used to select parents which the mutation and crossover is applied to in order to produce the offspring of the next generation. The population is evolved for a set number of generations.

4

3

D. Beckedahl et al.

Multi-space Search for One Dimensional Bin Packing

The one dimensional bin packing problem involves packing items into bins so as to minimize the number of bins used [1,10]. The version of the problem studied in this paper is the offline version in which the sizes of the items are known before hand. The concurrent multi-space model is applied to solving this problem. This model simultaneously searches the heuristic and solution space. The low-level construction heuristics used for the one dimensional bin packing problem are: – First-fit decreasing - This heuristic allocates the next item in the list of items, sorted in decreasing order according to size, to the first bin the item fits into. If the item does not fit into any of the existing bins the item is placed in a new bin. – Best-fit decreasing - The next item in the list of items, sorted in decreasing order according to size, is placed in the bin that the item fits into leaving the least amount of residual space. If the item does not fit into any of the existing bins the item is placed in a new bin. – Next-fit decreasing - The next item in the list of items, sorted in decreasing order according to size, is placed in the next bin that the item fits in. If the item does not fit into the next bin the item is placed in a new bin. – Worse-fit decreasing - The next item in the list of items, sorted in decreasing order according to size, is placed in the bin that the item fits into leaving the most residual space. If the item does not fit into any of the existing bins the item is placed in a new bin. A single move operator is used to explore the solution space, namely, that defined by Levine and Ducatelle [4] which aims to reshuffle the items in the bins so as to ultimately reduce the number of bins used. An overview of the operator is depicted in Algorithm 1: Algorithm 1. Move operator to explore the solution space 1: Empty the least full bin and store the items in the list free 2: For each remaining bin: 1. Try to swap two items from the bin with two items in free 2. Try to swap two items from the bin with one item in free 3. Try to swap one item from the bin with one item in free 3: Allocate the items remaining in free using the first-fit decreasing heuristic

The fitness of each chromosome is calculated by applying the chromosome to create a solution and calculating the objective value of this solution. The fitness of the chromosome is the objective value of the solution produced by the chromosome. The objective function is that proposed by Faulkenauer [1]: N f itness = 1 −

Fi 2 i=1 ( C )

N

(1)

A Study of Multi-space Search Optimization

4

5

Multi-space Search for Classification

The multi-space search for classification employs the sequential model hybridizing genetic programming and genetic algorithms to explore across the program and heuristic spaces sequentially. Each program in the program space is a classifier. The classifier can be an arithmetic tree or a production rule. The fitness of each program is the accuracy attained by the classifier. The best n classifiers are used to create the initial population of the genetic algorithm that explores the heuristic space with each classifier a low-level heuristic. The genetic algorithm exploring the heuristic space also uses accuracy as the fitness function. The sequential model multi-space search is applied to two problem domains, namely, network intrusion detection and weather prediction. Network intrusion detection involves determining whether there is an intrusion or not or the type/class of intrusion [6]. The weather prediction problem involves determining whether there will be heavy rainfall or not. The multi-space search exploring the program and heuristic space uses a training set to create classifiers which are evaluated on a test set for both classification problems.

5

Experimental Setup

This section provides an overview of the experimental setup employed to evaluate the multi-space search for one dimensional bin-packing and classification in terms of the data sets used, parameter values and technical specifications. 5.1

Data Sets

The Scholl benchmark set is used for the one dimensional bin packing problem [10]. This benchmark set is comprised of 720 easy, 480 medium and 10 hard problem instances. The NSL-KDD benchmark set [11] is used for network intrusion detection. The data set consists of instances comprised of attributes for the network traffic and the type/class of intrusion, including no intrusion. The data is divided into training and test sets. The weather data set is a real world data1 set to predict the intensity of rainfall in Seattle. 5.2

Parameter Values

The parameter values for the evolutionary algorithms for both the concurrent model multi-space search and the sequential model multi-space search were determined empirically by performing trial runs. The parameters for the genetic algorithm employed by the concurrent model is illustrated in Table 1. The parameters for the genetic programming algorithm and the genetic algorithm for the sequential model are listed in Tables 2 and 3 respectively. 1

https://www.kaggle.com/rtatman/did-it-rain-in-seattle-19482017.

6

D. Beckedahl et al. Table 1. Concurrent model GA parameter values Parameter

Value

Population size

500

Tournament size

5

No. of generations

75

Mutation rate

0.15

Crossover rate

0.85

Offspring maximum length 10 Mutation length

5

Table 2. Sequential model GP parameter values Parameter

Network intrusion detection

Weather prediction

Population size

250

600

Tournament size

50

5

No. of generations

17

75

Mutation rate

Randomly sampled

0.2

Crossover rate

Randomly sampled

0.8

Initial population maximum length 10

15

Offspring maximum length

10

-

Mutation depth

2

5

Table 3. Sequential model GA parameter values Parameter

Network intrusion detection

Weather prediction

Population size

250

750

Tournament size

10

5

No. of generations

30

75

Mutation rate

randomly selected

0.2

Crossover rate

randomly selected

0.8

Initial population maximum length 10

15

Offspring maximum length

10

-

Mutation length

-

5

The double tournament selector was employed for the GP algorithm for the network intrusion detection problem and standard tournament selection for the GA and weather predication. For the network intrusion network detection problem 5 runs were performed and the best classifiers on these runs formed the

A Study of Multi-space Search Optimization

7

heuristic set for the heuristic space search. For the weather classification problem 5 runs were performed to produce arithmetic tree classifiers and 5 producing rule-based classifiers. The heuristic set for the heuristic space search was comprised of these 10 classifiers. 5.3

Technical Specifications

Simulations were run on machines with the following technical specifications: – One dimensional bin packing experiments - CHPC Lengau cluster – Network intrusion detection experiments - Windows 10, i5-7500 CPU, 3.40 GHz – Weather prediction - CHPC Lengau cluster Due to the stochastic nature of evolutionary algorithms 30 runs were performed for each problem/data instance for the one dimensional bin packing problem and the weather data set and 20 runs for the network intrusion detection data instances.

6

Results and Discussion

Table 4 displays the results obtained by the concurrent model multi-space search (CMSS) for the one dimensional bin packing problem. To get some idea of the contribution made by search the results are compared to a selection constructive hyper-heuristic (SCHH) that works in just the heuristic space. The table lists the number of optimal solutions found by CMSS and SCHH for the easy set of 720 problem instances, the medium set of 480 problem instances and the hard set of 10 problem instances. Table 4. Concurrent model performance for one dimensional bin packing Problem instances CMSS SCHH Easy (720)

663

563

Medium (480)

423

238

6

0

Hard (10)

As can be seen from Table 4 the CMSS performs much better than SCHH which works in just one space. The Friedman test with Holm’s post-hoc procedure confirmed that this result is statistically significant. Table 5 lists the results of the sequential model multi-space search (SMSS) for the network intrusion detection data instance and the weather prediction data instance. As can be seen from Table 5 in the domain of classification the multi-space search has not had as big an impact as it has had for the one dimensional

8

D. Beckedahl et al. Table 5. Concurrent model performance for classification Data instances

SMSS Training Test

GP Training Test

Network intrusion detection 92.8%

92.49% 92.86%

92.3%

Weather prediction

88%

88%

92%

92%

bin packing problem. The multi-space search has performed slightly better than searching in just the program space for the network intrusion detection problem for testing and the performance is on par for weather prediction. It could be the model for the multi-space search is not appropriate. Future work will examine other options for models as well as the possibility of automating the process of deriving models. The average runtimes for the multi-space search is as follows: – CMSS for one dimensional bin packing: 11 min 3 s – SMSS for network intrusion detection: 31 min – SMSS for weather prediction: 8 h

7

Conclusion

The aim of the research presented in this paper is to introduce the concept of nspace search where search technologies explore more than one space. As a proof of concept we illustrate this for 2-space search. The paper presents two models for performing search over two spaces, namely, the concurrent model and the sequential model. The concurrent model was used to simultaneously search the heuristic and solution space to solve the one-dimensional bin packing problem. The sequential model combined exploration of the program space and heuristic space for classification and was evaluated for network intrusion detection and weather prediction. The concurrent model was found to perform much better than exploring just one search space in solving the one dimensional bin packing problem. A significant difference in performance was not found for the classifiers produced by the sequential model multi-space search and those generated by examining the program space only. It is hypothesised that if a different model for combining search across spaces would improve performance. This will be investigated as part of future work. This study has introduced the idea of n-space search and has illustrated the potential of such search for two spaces. Future work will investigate extending the search to more than two spaces. Two models for multi-space search were presented, namely, the concurrent model and the sequential model. Future work will as look at other options for models. Furthermore, this study has revealed the most appropriate model to use to hybridize search over more than one space to solve the problem at hand is challenging and is an optimization problem in itself. Hence, future work will also examine automating the process of deciding

A Study of Multi-space Search Optimization

9

which spaces to search and how to search across the spaces. Reusability of the automatically generated models will also be investigated. Acknowledgments. This work was funded as part of the Multichoice Research Chair in Machine Learning at the University of Pretoria, South Africa. The authors would like to acknowledge the Centre for High Performance Computing (CHPC) in South Africa for the provision of resources to run the simulations for the experiments in this study.

References 1. Falkenauer, E.: A hybrid grouping genetic algorithm for bin packing. J. Heuristics 2(1), 5–30 (1996) 2. Koza, J.R.: Genetic Programming: On the Programming of Computers by Means of Natural Selection. MIT Press, Cambridge (1992) 3. Krawiec, K.: Behavioral Program Synthesis with Genetic Programming. Springer Nature. Springer, Switzerland (2016) 4. Levine, J., Ducatelle, F.: Ant colony optimization and local search for bin packing and cutting stock problems. J. Oper. Res. Soc. 55, 705–716 (2004) 5. Luke, S., Panait, L.: Fighting bloat with nonparametric parsimony pressure. In: Proceedings of the International Conference on Parallel Problem Solving from Nature (PPSN 2002), pp. 411–421 (2002) 6. Mukherjee, B., Heberlein, L., Levitt, K.N.: Network intrusion detection. IEEE Network 8(3), 26–41 (1994) 7. Pillay, N., Qu, R.: Hyper-Heuristics: Theory and Applications. Natural Computing Series. Springer, Cham (2018) 8. Qu, R., Burke, E.K.: Hybridizations within a graph-based hyper-heuristic frame work for university timetabling problems. J. Oper. Res. Soc. 60, 1273–1285 (2009) 9. Reynolds, R.: An introduction to cultural algorithms. In: Proceedings of the 3rd Annual Conference on Evolutionary Programming, pp. 131–139 (1994) 10. Scholl, A., Klein, R., Jurgens, C.: Bison: a fast hybrid procedure for exactly solving the one-dimensional bin packing problem. Comput. Oper. Res. 24(7), 5–30 (1997) 11. Tavallaee, M., Bagheri, E., Lu, W., Ghorbani, A.A.: A detailed analysis of the KDD cup 99 data set. In: Proceedings of the IEEE Symposium on Computational Intelligence for Security and Defense Applications, pp. 1–6 (2009)

Clinical Decision Support System for Neuro-Degenerative Disorders: An Optimal Feature Selective Classifier and Identification of Predictor Markers Lokeswari Venkataramana(&), Shomona Gracia Jacob, S. Saraswathi, and R. Athilakshmi Sri Sivasubramaniya Nadar College of Engineering, Chennai 603110, Tamil Nadu, India [email protected], [email protected], {saraswathis,athilakshmir}@ssn.edu.in

Abstract. Detecting divergence between Neuro-degenerative diseases is essential for right treatment. This intelligent system is implemented through computational methods to predict the class of Neuro-degenerative disease (Alzheimer’s, Parkinson’s or common) from the structural and physicochemical properties (1437 attributes respectively) of protein sequences extracted from genes. The Gene Set Enrichment Analysis database (GSEA db) was utilized to obtain the gene sets that contributed to the development of Alzheimer’s and Parkinson’s disease. Optimal features for classification were obtained by applying Gain Ratio followed by Correlation-based Feature Selection (CFS) and Decremental Feature Selection (DFS) on extracted properties from Kyoto Encyclopedia of Genes and Genomes (KEGG) dataset for the GSEA database. The selected features are evaluated using Random Forest model. The Clinical Decision Support System (CDSS) was build which extract rules from the least sized Decision tree automatically and predict the type of Neuro-degenerative disorder as Alzheimer’s disease, Parkinson’s disease or common to both diseases. The CDSS predicts the disease with classification accuracy as 79.7% and Mathew’s Correlation Coefficient as 0.689. Keywords: Neuro-degenerative disorder
Alzheimer’s disease
Parkinson’s disease
Prediction system
Clinical Decision Support System

1 Introduction Neuro-degenerative diseases are caused by genetic mutation, mostly located in completely unrelated genes. One common feature is repetition of CAG (Cytosine Adenine Guanine) nucleotide triplet. Affected people and few family members of affected patients have expansion of this repeat number (60–70) compared to normal people (20– 30). Brain disorders emerge as leading contributors to global disease burden; for example, according to the most recent (2004) World Health Organization (WHO) survey, unipolar depressive disorders are the single biggest source of lost © Springer Nature Switzerland AG 2020 A. Abraham et al. (Eds.): ISDA 2018, AISC 940, pp. 10–20, 2020. https://doi.org/10.1007/978-3-030-16657-1_2

Clinical Decision Support System for Neuro-Degenerative Disorders

11

DALY’s (Disability Adjusted Life Year) in the high income countries, and the third biggest worldwide. In Europe it has been estimated that 35% of all disease burden is attributable to brain disorders. Alzheimer’s disease (AD) and Parkinson’s disease (PD) are found only in later stages or older ages. In 2012 annual report, the Alzheimer’s Association (AA) estimates that 5.4 million people in the US have Alzheimer’s disease (AD). The risk of AD increases with age, so unless new treatments are discovered this number will grow sharply as the baby boomer generation reaches old age. By 2050, the Alzheimer’s Association estimates that close to 16 million Americans will have the disease, with one new case appearing every 33 s. The US National Institute for Neurological Disorders and Stroke (NINDS) estimated in a 2006 report that about 50,000 new cases of Parkinson’s disease are diagnosed in the US each year, and the total number of cases in the US is at least 500,000. The true prevalence (total number of cases) of Parkinson’s disease is difficult to assess, because the disease is typically not diagnosed until the disease process is already far advanced. Therefore, the actual number of Americans with the disease is almost certainly higher than the diagnostic numbers would suggest [1]. Hence determination of potential and informative markers (diagnostic and prognostic) from both the biological and molecular perspective is highly essential to come up with diagnosis. Also prediction markers before the symptoms shown down the line will determine the difference between life and death. This will give a sea of options for clinical treatments and study that the person can participate to neutralize or avoid cell degeneration by using the derived data from the system. These diseases have grown so much that they are currently the sixth leading cause of death all over the world. So it has become immensely important to identify these diseases at a very early stage. The proposed system uses the gene sets of those patients who already have Alzheimer’s disease and Parkinson’s disease. From these gene sets the protein sequence and the structural properties of them are extracted by the processes of transcription followed by translation. The extracted data is prepared for feature selection using the feature evaluators. This is the training dataset for the machine learning system. The user of the system will then provide the system with a test data set to predict whether the new patient have either of the two disease or not. The system reduces the test data to prominent features which contribute for classification, compares them with the classifier obtained from training data set and predicts the results. The symptoms of Alzheimer’s diseases are poor decision making and judgment, misplacing things, impairments of movements, verbal communication, abnormal moods, and complete loss of memory [2–4]. If the disease is not diagnosed at the initial stage, the severity of the disease increases. Parkinson’s disease is a neurological disorder based on dopamine receptors. It affects the mobility of the subject. It is a progressive condition characterized by both motor and non-motor symptoms. People with Parkinson’s disease are presented with the symptoms and signs associated with Parkinsonism, namely, hypokinesia, rigidity, bradykinesia and rest tremor [5, 6].

12

L. Venkataramana et al.

2 Related Work The research work outlined in the work [7], focused on predicting the class of lung cancer tumors. This is done by designing a computational strategy defined by microarray analysis. Structural and physiochemical properties (1497 attributes) of protein sequences are used for this prediction. The proposed methodology uses hybrid feature selection techniques which consist of gain ratio and correlation based subset evaluators with Incremental Feature Selection. Bayesian Network prediction [8] is then used to discriminate lung cancer tumors as Small Cell Lung Cancer (SCLC), NonSmall Cell Lung Cancer (NSCLC) and common classes. This approach has advantages such as extensive data cleansing strategies on protein properties is not required. This also leads to lung cancer tumor classification with an improved accuracy using optimal and minimal set of features. The drawback of this approach is that cases that fell under common class cannot be further used as an informative source. The work in [9] discussed about an anonymized Electronic Medical Records (EMR) used to create a first risk classification model for Cyclic Redundancy Check (CRC) using data mining techniques. Data include general characteristics of the patient (age, gender, observation time) structured and coded information on consults (dates, description of the findings such as symptoms and diagnoses in the form of International Classification of Primary Care (ICPC) codes, a structured coding approach used worldwide, medication (prescription data and ATC (Anatomical Therapeutic Chemical) code of the type of medicine), and information about referral to specialists. The algorithm deployed is the CHAID (Chi-Square Automatic Interactive Detector) decision tree learning algorithm as the specialists need to be able to understand the resulting predictive models. The advantage of this method is, it provides studies and results on symptoms and studies of similar cases fed to the system. The drawback is that the system doesn’t learn new cases it just presents inference from existing cases according to age distribution and gender. Having surveyed, the recent work in Neuro-degenerative disorder and applying computational methods on clinical data for predicting the disease, this research focuses on applying a suitable feature selection methods and designing a Clinical Decision Support System (CDSS) which extracts rules from the Decision Tree and improved the prediction accuracy and MCC in diagnosing Alzheimer, Parkinson diseases. Thus the objectives of this research are stated as follows: 1. To select optimal biomarkers from gene and protein sequences of Neurodegenerative disorder data set. 2. To predict Neuro-degenerative disorder at earlier stage using physiochemical properties of protein sequences. 3. To design an expert system that extract rules from tree models for predicting the class of disease.

Clinical Decision Support System for Neuro-Degenerative Disorders

13

3 Materials and Methods The gene sets for Neuro-degenerative diseases are obtained from Kyoto Encyclopedia of Genes and Genomes (KEGG) [10]. From these gene sets, proteins are obtained using GENECARD database and for these proteins their protein sequences are obtained using UNIPROT knowledge database [11]. Structural and physio-chemical properties of these proteins are obtained using PROFEAT web server [12]. This resulted in generating dataset having 199 instances each having 1437 features and 3 target classes namely Alzheimer’s Disease (AD), Parkinson’s Disease (PD) and common to both diseases. The description of each disorder type is tabulated in Table 1. Table 1. Dataset Description of Neuro-degenerative disease Dataset Neurodegenerative disorder

No. of genes 1437

Total samples 199

Target class 3

Class-wise samples 74 37 88

Disorder types Alzheimer Disease (AD) Parkinson Disease (PD) Common to AD & PD

The following methods were explored for classifying Neuro-degenerative disorder. 3.1

Dataset Generation

Dataset generation begins with collecting the genes related to Alzheimer’s and Parkinson’s disease from Kyoto Encyclopedia of Genes and Genomes [10] (KEGG) database. A total of 112 genes are collected. There are 74 genes uniquely pertaining to Alzheimer’s disease and 38 genes uniquely pertaining to Parkinson’s disease. There are 95 genes common to both the diseases. The gene sequences for every gene were obtained in the next step from the UniProt database. A gene related to Parkinson’s disease – LOC729317 – had no authorized gene sequence and hence was disregarded and not included in further investigations. The output at this phase gives the gene sequence for 111 genes. The next phase deals with the extraction of structural and physicochemical properties of the 111 genes from the PROFEAT server. There are 1437 features or protein properties for every gene. Thus the dataset consists of 111 rows of genes with 1437 columns representing the protein properties. The class label that identifies the disease takes the total of to1438 columns in the final dataset [19]. 3.2

Feature Selection

As the number of features is many in number, all features may not contribute to both the diseases. Few genes contribute to diagnose either AD or PD or common to both diseases. Three feature selection methods are applied in pipeline fashion to select very important and optimal features. Gain Ratio followed by Correlation Feature Subset Selection and Decremental Feature Selection methods are applied on to protein properties.

14

L. Venkataramana et al.

3.2.1 Gain Ratio Protein properties from PROFEAT are obtained for 1437 protein sequences. Features are ranked based on Gain ratio. Gain Ratio uses an extension to information gain that attempts to overcome the bias on attributes selected by the information gain criterion. It applies a kind of normalization to information gain using a split information value defined analogously with InfoA(D) as stated by Han and Kamber [13]. The gain ratio is obtained from the Eq. 2. SplitInfoA ðDÞ ¼ 

V X jDj j j¼1

jDj

X log2

jDj j jDj

ð1Þ

This value represents the potential information generated by splitting the training data set, D, into v partitions, corresponding to the v outcomes of a test on attribute A. The number of tuples having a certain outcome with respect to the total number of tuples in D alone was considered. It differs from information gain, which measures the information with respect to classification that is acquired based on the same partitioning [14]. The gain ratio is defined by Eq. 2. The attribute with the maximum gain ratio is selected as the splitting attribute. GainRatioA ¼

GainA SplitInfoA

ð2Þ

3.2.2 Correlation Feature Subset Selection (CFS) Attribute Evaluator Method The removal of irrelevant and redundant information often improves the performance of machine learning algorithms [15]. Feature subset selection is the process of identifying and removing as much of the irrelevant and redundant information as possible. The following equation dictated the merit of a feature subset S that consisted of ‘k’ features [16, 17]. k
rcf MeritSk ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi k þ kðk  1Þ
rff

ð3Þ

where
rcf was the average value of all feature-classification correlations, and
rff was the average value of all feature-feature correlations. The CFS criterion was defined as follows. "

rcf 1 þ rcf 2 þ . . . þ rcfk CFS ¼ MAX pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Sk k þ 2ðrf 1f 2 þ . . . þ rfifj þ . . . þ rfkf 1 Þ

# ð4Þ

where rcfi and rfifj variables are referred to as correlations. The attributes that portrayed a high correlation to the target class and least relevance to each other were chosen as the best subset of attributes [16, 17]. The attributes filtered by the CFS subset evaluator method were given to classifier algorithms.

Clinical Decision Support System for Neuro-Degenerative Disorders

15

3.2.3 Decremental Feature Selection Decremental Feature Selection method was applied on optimal feature subset with filter combinations GR-CFS. Gain Ratio are CFS are done to improve classifier accuracy. DFS is the method in which, it will start with all the features and remove one feature at each iteration, which improves the performance of the model. We repeat this until no improvement is observed on removal of features. This method also reduced the number of attributes required for the accurate prediction of the class. The best performing classification algorithm was again executed on the new subset of features and the accuracy was noted. It was observed that the accuracy increased steadily as features were removed before becoming stable. The subset of attributes which recorded the maximum accuracy was recognized as the final set of optimized attributes for which classifier model was developed.

4 Framework for Predicting Neuro-Degenerative Disorder The proposed system for identifying biological markers is shown in Fig. 1 and it comprises of four processes: 1. Extraction of protein sequences from gene set using Uniprot database 2. Extraction of physio-chemical properties (features) using PROFEAT web server. 3. Application of novel algorithms for optimal feature selection (Gain ratio, Correlation based Feature Selection (CFS) and Decremental feature selection) and training the predictor model using classification algorithm. (Random Forest algorithm). Initially, Gain Ration with ranker method is used to select important features. Features with score less than 0.2 are removed. At least half of the datasets lie with gain ratio close to and equal to 0.2. Therefore best features will have threshold close to this value. Ranker method is used and thus the resulting features are arranged in descending order of their ranks. 4. Secondly, CFS with best first search method is applied. Optimal feature set is constructed using CFS feature selection. The input of this operation is the entire 1437 features of the dataset. After the CFS algorithm is applied, the number of features gets reduced to a mere 85 thus making data handling easy and the process much faster. These 85 features contribute more towards the disease than the rest. 5. Finally Decremental Feature Selection (DFS) is applied. Accuracy of classifier model is checked on removal of every single feature. The expected classifier model should have highest accuracy and MCC compared to other classifier models computed. Feature removal is done iteratively till accuracy increases after stable values. 6. Construct Random Forest model and rules are extracted from least sized Decision Tree [18] by the Clinical Decision Support System (CDSS). Test data is supplied to evaluate the build predictor model. 7. The predicted result (Alzheimer’s disease, Parkinson’s disease or common to both diseases) is displayed.

16

L. Venkataramana et al.

Fig. 1. Framework for predicting neuro-degenerative disease

4.1

Test Data Generation

The test data is generated using stratified random sampling method. Figure 2 represents test data generation. Resampling method is used to generate test data for prediction system. A stratified sample is one that ensures that subgroups (strata) of a given population are each adequately represented within the whole sample population. 1. The first step is to get the test data in the required format. 2. Additionally, every attribute in each category is averaged to produce a new dataset. This resulted in three samples each one for AD, PD and common to both. Twenty instances from Resampling method and three from synthesized data by average of attribute values for each feature. This gave totally 23 instances as test data.

Fig. 2. Framework for Test data Generation

Clinical Decision Support System for Neuro-Degenerative Disorders

17

Algorithm: Expert system for predicting Neuro-degenerative disorder Input: Protein sequences of Neuro-degenerative disorder from KEGG. Output: Predicting type of Neuro-degenerative disorder. 1. Obtain the structural and physio-chemical properties of the protein sequence of genes from KEGG. 2. Apply Gain ratio on the obtained dataset. 3. On the resultant dataset, apply Correlation Feature Selection. 4. Finally, apply Decremental Feature Selection to obtain the optimal features of the dataset. 5. Construct Random Forest from the optimal features obtained in step 4. 6. Extract rules from the Decision Tree with least size in the Random Forest. 7. Evaluate random forest using 10-fold Cross-Validation. 8. Generate test dataset using Stratified Resampling method and synthesized data from training dataset. 9. Predict the class of test dataset. 10. Return Prediction accuracy

4.2

Prediction System

Decision rules are constructed using Random Forest for various properties of the protein sequence such that accuracy is high. As Decision Tree and Random Forest performs better than any other classification model for extracting rules, this research focussed on Decision Tree model [2]. These rules are then applied on the test data which helps classify this data as either Alzheimer’s disease or Parkinson’s disease or common to both diseases. For example as depicted in Fig. 3, if G5.2.2.22 is greater than or equal to 0.04 AND G4.3.2.4 is less than 28.81 then the test data belongs to both Alzheimer’s and Parkinson’s disease. Java with Weka was used as a programming language [20].

5 Results and Discussion The Clinical Decision Support System constructs the Random Forest [18]. The Random Forest algorithm which prints the all possible ‘n’ trees, where ‘n’ is the number of predefined trees that is to be considered for the voting process that the Random Forest algorithm uses. Here the value of n is 10. Since any decision tree from the generated trees can be chosen, the tree with the least number of nodes is chosen for faster performance and for convenience. The selected tree is then modelled into a programming language for rule extraction. When Gain Ratio is applied at least half of the datasets lie with gain ratio close to and equal to 0.2 which resulted in 718 features. CFS is applied which yielded 85 features as important features. To these 85 features, DFS is applied, which resulted in optimal and best feature set having 55 features. These 55 features were evaluated using

18

L. Venkataramana et al.

Fig. 3. Decision tree model on Neuro-degenerative dataset

Random Forest and test data is supplied. The Clinical Decision Support System extracted a Decision Tree which has least size with 41 nodes which is shown in Fig. 3. Rules are extracted from the CDSS gave a prediction accuracy as 79.9% with Mathew’s Correlation Coefficient (MCC) as 0.689. When the next feature was removed (with 54 features) it reduced prediction accuracy to 76.4% and MCC to 0.632. When the number of features was reduced to 70, it gave prediction accuracy as 79%, later further removal of features reduced prediction accuracy to 76% and attained a stable accuracy value when there are 55 features. Figure 3 represents the Decision Tree model obtained from Neuro-degenerative disorder data obtained from KEGG database. The test data is fed as input and the system classifies it using the random forest classifier model that is obtained after DFS. Since the test data is obtained after resampling the training data, new data has been synthesized from the training data by averaging every attribute in separate categories to produce a new dataset with 3 instances and 1437 attributes. This is fed as input to the system and the classified results are displayed. The accuracy obtained using 10-fold cross validation is 79.9% and 100% with stratified resampling method. Table 2 depicts the results obtained from CDSS. When DFS method selected 55 features, accuracy value becomes stable and iterative DFS method is halted.

Clinical Decision Support System for Neuro-Degenerative Disorders

19

Table 2. Performance of clinical decision support system S. No. Feature selection No. of features 1 Gain ratio 718 2 CFS 85 3 DFS 70 4 DFS 64 5 DFS 62 6 DFS 55 7 DFS 54

Accuracy 68 73 79 77 78 79 76

MCC 0.653 0.643 0.521 0.633 0.653 0.689 0.632

6 Conclusion A training dataset for Neuro-degenerative diseases was initially supplied to the Clinical Decision Support System. The physio-chemical properties of the protein sequence of the input data set act as features to build the prediction system. Gain Ratio, Correlationbased feature selection and Decremental Feature Selection were applied on features to compute optimal feature set. The CDSS was constructed with Random Forest model which extracted rules from the least sized Decision Tree. Random forest algorithm was used to build the classifier model. The Random Forest consists of a set of 10 Decision Trees. Thus, a dynamic prediction model was the resultant system. When test data set was provided to this model, the system will predict which Neuro-degenerative disease the sequence falls under: Alzheimer’s disease, Parkinson’s disease or common to both diseases. This test data is a resampled dataset and data synthesized from training dataset. The prediction was based on which category the test dataset receives maximum votes from the Random Forest generated previously. 10-fold cross validation yielded classification accuracy of 79.9% and MCC of 0.689 with 55 optimal features. 100% accuracy was obtained when stratified remove folded test data was provided as input. This system can further be extended to a generalized system for all diseases which will select the best feature selection algorithm based on the dataset. Acknowledgements. This research work is part of project work funded by Science and Engineering Research Board (SERB), Department of Science and Technology (DST) funded project under Young Scientist Scheme – Early Start-up Research Grant- titled “Investigation on the effect of Gene and Protein Mutants in the onset of Neuro-Degenerative Brain Disorders (Alzheimer’s and Parkinson’s disease): A Computational Study” with Reference No- SERB – YSS/2015/000737.

References 1. Brain Disorders by Numbers (for brain research at MIT) (2018). https://mcgovern.mit.edu/ brain-disorders/by-the-numbers. Accessed 03 June 2018 2. Jacob, S.G., Athilakshmi, R.: Extraction of protein sequence features for prediction of neurodegenerative brain disorders: Pioneering the CGAP database. In: Proceedings of the International Conference on Informatics and Analytics, p. 30. ACM, August 2016

20

L. Venkataramana et al.

3. Shree, S.B., Sheshadri, H.S.: Diagnosis of Alzheimer’s disease using rule based approach. Indian J. Sci. Technol. 9(13) (2016) 4. Shree, S.B., Sheshadri, H.S.: An initial investigation in the diagnosis of Alzheimer’s disease using various classification techniques. In: 2014 IEEE International Conference on Computational Intelligence and Computing Research (ICCIC), pp. 1–5. IEEE, December 2014 5. Bind, S., Tiwari, A.K., Sahani, A.K., Koulibaly, P.M., Nobili, F., Pagani, M., Sabri, O., Borght, T.V., Laere, K.V., Tatsch, K.: A survey of machine learning based approaches for Parkinson disease prediction. Int. J. Comput. Sci. Inform. Technol. 6(2), 1648–1655 (2015) 6. Ramani, R.G., Sivagami, G.: Parkinson disease classification using data mining algorithms. Int. J. Comput. Appl. 32(9), 17–22 (2011) 7. Ramani, R.G., Jacob, S.G.: Improved classification of lung cancer tumors based on structural and physicochemical properties of proteins using data mining models. PLoS ONE 8(3), e58772 (2013) 8. Kaladhar, D.S.V.G.K., Pottumuthu, B.K., Rao, P.V.N., Vadlamudi, V., Chaitanya, A.K., Reddy, R.H.: The elements of statistical learning in colon cancer datasets: data mining, inference and prediction. Algorithms Res. 2(1), 8–17 (1926) 9. Hoogendoorn, M., Moons, L.M., Numans, M.E., Sips, R.J.: Utilizing data mining for predictive modeling of colorectal cancer using electronic medical records. In: International Conference on Brain Informatics and Health, pp. 132–141. Springer, Cham (2014) 10. KEGG dataset for neuro-degenerative diseases (2018). http://www.genome.jp/kegg-bin/ gethtext?br08402.keg. Accessed 6 June 2018 11. Uniprot database Server, 22 June 2017. http://www.uniprot.org. Accessed 6 June 2018 12. PROFEAT web server (2018). http://bidd2.nus.edu.sg/cgi-bin/prof2015/prof_home.cgi. Accessed 05 June 2018 13. Han, J., Pei, J., Kamber, M.: Data Mining: Concepts and Techniques. Elsevier, New York (2011) 14. Karegowda, A.G., Manjunath, A.S., Jayaram, M.A.: Comparative study of attribute selection using gain ratio and correlation based feature selection. Int. J. Inf. Technol. Knowl. Manag. 2(2), 271–277 (2010) 15. Jacob, S.G., Ramani, R.G., Nancy, P.: Feature selection and classification in breast cancer datasets through data mining algorithms. In Proceedings of the IEEE International Conference on Computational Intelligence and Computing Research (ICCIC 2011), Kanyakumari, India, IEEE Catalog Number: CFP1120 J-PRT, December 2011, ISBN 978-1 16. Hall, M.A.: Correlation-based feature selection for machine learning (1999) 17. Doshi, M.: Correlation based feature selection (Cfs) technique to predict student perfromance. Int. J. Comput. Networks Commun. 6(3), 197 (2014) 18. Richards, J.W., Eads, D., Bloom, J.S., Brink, H., Starr, D.: WiseRFTM: a fast and scalable Random Forest (2013) 19. Tejeswinee, K., Jacob, S.G.: Binary classification of cognitive disorders: investigation on the effects of protein sequence properties in Alzheimer’s and Parkinson’s disease. In: IAENGIMECS 2017, 166–170 (2017) 20. Implementing WEKA in java, 03 June 2018. https://weka.wikispaces.com/Use+WEKA+in +your+Java+code

Favoring the k-Means Algorithm with Initialization Methods Anderson Francisco de Oliveira1 and Maria do Carmo Nicoletti1,2(&) 1

2

Centro Universitário C. Limpo Paulista (UNIFACCAMP), Campo Limpo Paulista, SP, Brazil [email protected], [email protected] Universidade Federal de S. Carlos (UFSCar), São Carlos, SP, Brazil

Abstract. Clustering algorithms are non-supervised algorithms and, among the many available, the k-Means can be considered one of the most popular and successful. The performance of the k-Means, however, is highly dependent on a ‘good’ initialization of the k group centers (centroids) as well as of the value assigned to the number (k) of groups the final clustering should have. This chapter addresses experiments using five initialization algorithms available in the literature namely, the Method1, the k-Means++, the CCIA, the Maedeh&Suresh and the SPSS algorithms, to empirically evaluate their contribution to improving k-Means performance. Keywords: Unsupervised learning


k-Means
Initialization algorithms

1 Introduction and Motivations The solution to a clustering problem can be addressed in several ways; one of them is known as a partitional approach, in which the clustering problem is seen as a partitioning problem of the initial set of data instances. Among the several partitional algorithms available in the literature, the k-Means [1] is considered the most successful and has been used in a large number of applications. It is known that the k-Means suffers from a problem identified as initialization problem, related to the initial set of group centers (or centroids), from which the iterative process conducted by the algorithm begins. It can be found in the literature several algorithms that attempt to solve the problem of centroid initialization. The research work described in this chapter aimed at the investigation of five initialization algorithms, with the goal of identifying their real contributions to the problem, to evaluate how feasible they are and to exam their actual originality. The remainder of this chapter is organized as follows. Section 2 reviews a few general characteristics of clustering algorithms in general and briefly revisits the k-Means algorithm. Section 3 presents and discusses the main aspects of the five initialization algorithms used to initialize the k-Means, in the experiments described in Sect. 4 of this chapter: (1) the Method1 [2], (2) the k-Means++ [3], (3) the Maedeh&Suresh [4], the CCIA [5] and finally, the SPSS [6] algorithms. Section 5 resumes the work done and concludes presenting a few directions in which the work may proceed. © Springer Nature Switzerland AG 2020 A. Abraham et al. (Eds.): ISDA 2018, AISC 940, pp. 21–31, 2020. https://doi.org/10.1007/978-3-030-16657-1_3

22

A. F. de Oliveira and M. do Carmo Nicoletti

2 Revisiting the k-Means Algorithm The process of grouping data instances based on measures of similarity (or dissimilarity) between them can be trivially performed by humans, but designing an algorithm for performing the task is not trivial. An algorithm for this purpose should identify groups of instances based only on their descriptions. As pointed out in [7], efficient clustering techniques are considered a great challenge, mainly due to the fact that they do not have external supervision; this somehow implies they should work under the constraint of a total lack of prior knowledge about the internal structure of the data (such as spatial distribution, volume, density, geometric shapes of groups, etc.). In this scenario, automatic learning becomes an exploratory activity, aiming at identifying statistically separable data groups, detecting the most evident groups and their relation to what one wishes to discriminate, in an attempt to highlight the underlying structure of the data set, having as information only the data instance descriptions, each of them represented by a vector of attribute values. Figure 1 describes the pseudocode of the k-Means algorithm based on the description found in [8]. Many works reported in the literature mention that both, the convergence of the iterative process and the performance of the clustering induced by the k-Means depend on the initial set of centroids [9]. Both, the number of groups (k) and the initialization of the centroids of groups are relevant aspects that affect the performance of the algorithm. procedure k-means(SDI,k,AG) Input: SDI = {p1, p2, ..., pN} % set of data instances to be clustered k % number of groups to be created Output: AG = {G1,G2,...Gk} % clustering with k groups of data instances begin % Initialization phase (1) randomly choose k data instances ∈ SDI, as the initial k centroids, C1,C2,...Ck , of groups G1,G2,...Gk % at this point each Gi only has one element, the Ci (i=1, ...,k) % Iterative phase (2) repeat (3) (re)assign each pi ∈ SDI to the group whose centroid is the closest one to pi. (4) update each Ci (i=1, ...,k) as the average of the data instances that belong to Gi (5) until no further modification happens. end. return AG = {G1,G2,...Gk} end procedure Fig. 1. Simplified pseudocode of the k-Means, based on the description found in [8].

As shows Fig. 1, in the initialization process carried out by the original k-Means, k data instances from the given data set are randomly selected as the initial centroids of k groups, which can be considered a deficiency, since there are more convenient ways of selecting a more appropriate set of k centroids, rather than randomly choosing them. Section 3 addresses five algorithms found in the literature, which are based on a diverse set of mathematical formalisms and aim at providing the k-Means with a ‘good’ set of k initial centroids.

Favoring the k-Means Algorithm with Initialization Methods

23

3 A Brief Introduction to the Five Initialization Algorithms Used The five algorithms focusing on the initialization phase i.e., choosing an appropriate initial set of centroids, are briefly approached next in the following sequence: the Method1 [2], the k-Means++ [3], the Maedeh&Suresh algorithm [4] named in this chapter by the names of its creators, the Cluster Center Initialization Algorithm (CCIA) [5] and the Single Pass Seed Selection (SPSS) [6]. In [2] two k-Means initialization algorithms are proposed; the Method1 is one of them. The Method1 can be characterized as grid-based algorithm, since it considers the space defined by the data divided into a certain number of cells, all of them with the same dimensions. The process of selecting the initial set of centroids is done all at once and, according to the authors, there is no need for further attempts to define them. Also, as the authors point out, the algorithm does not require a lower limit in relation to the number of data instances. The algorithm is based on the idea of selecting the initial centroids of groups according to the distribution of data instances at a macro level, leaving the grouping task itself, in charge of the clustering algorithm (the k-Means, in this case) that is expected to improve and refine the initial solution provided by Method1. The algorithm distributes the centroids directly, driven by the density of the data instances. The k-Means variant known as k-Means++ [3] is the original k-Means itself, with a change in its initialization step. The k-Means++ initialization process still randomly chooses the initial k centroids, but ponders them according to the square of their distances to the centroid that is closest to them, among those already chosen. The authors empirically show that the k-Means++ performs better than the original k-Means in both, accuracy and speed and, generally, by a substantial margin. The Maedeh&Suresh algorithm [4] can be considered a modified k-Means, having the two phases of the original k-Means been modified; the modified phases, however, have the same purpose of the original phases. In the initialization phase, k centroids of groups are chosen and in the iterative phase, the algorithm assigns data instances to the k groups, depending on their distance to the corresponding group centroids. In the initialization phase the algorithm calculates de distance from each data instance x to the Cartesian origin and then, the instances are sorted in ascending order, based on their corresponding distance from the origin. The first instance and the last instances are then chosen as first and second group centroids, respectively. All the remaining data instances are then assigned to their closest group centroid. For a data set with N data instances, two N-dimensional vectors are then created: Cluster_Id and Nearest_Dist. Each position i (1  i  N) in Cluster_Id has the cluster number assigned to data instance xi and, in Nearest_Dist, has its distance from the group centroid. Next the algorithm identifies the position (g) in the Nearest_Dist vector having the highest value and chooses xg as the third group centroid. The process is repeated until k centroids are identified. So, in the experiments described in Sect. 4, the first phase of the Maedeh&Suresh algorithm described above was used as the initialization process of the k-Means.

24

A. F. de Oliveira and M. do Carmo Nicoletti

According to Khan and Ahmad [5], the CCIA algorithm was proposed with the intent of obtaining a good startup of centroids, to be used by the k-Means. The algorithm was based on two observations associated with clustering processes under k-Means: (1) some data instances are very similar to each other and, due to that, they end up belonging to the same group, regardless of the way the initial choice of centroids is made; (2) attributes, individually, may also provide some information regarding the initial centroids. CCIA exclusively aimed at initializing centroids for the k-Means and applies only to numerical data. Authors in [6] point out that, considering the fact that the k-Means++ [3] starts the initialization process by randomly choosing the first centroid, this can affect the number of iterations at each execution, eventually giving rise to different results, for the same set of data instances to be clustered. The authors also state that for the k-Means++ to reach a good result, it has to be run a certain number of times. Their proposed SPSS algorithm can be considered a slightly modified k-Means++ that selects, as the first centroid, the highest density data instance (note that the first centroid is a deterministic choice) and the choice of the maximum distance separating centroids is based on the instance that is closest to other instances in the data set. According to the authors the initial set of centroids selected by the SPSS promotes the quality of the groups induced by the k-Means, the number of iterations performed by k-Means as well as the number of times distance calculations have to be conducted.

4 Experiments, Results and Analysis To evaluate the five algorithms, four synthetic data sets and four real data sets were used. The characteristics of the eight data sets are presented in Table 1 and the plottings of the four synthetic data sets are in Fig. 2. Out of the four real data sets, three were downloaded from the UCI Repository [10] namely the Iris, Wine and Seeds and one, the Fossil data set, was available in [11]. The synthetic data sets used in the experiments were: MSD (referring to the data employed by the authors Maedeh&Suresh in [4], to evaluate their algorithm named in this chapter as Maedeh&Suresh or simply MS), LongSquare [12], Aggregation [13] and 3MC [14]. In what follows the five initialization algorithms are referred to by their abbreviations: k-Means++ (++), CCIA (CCIA), Method1 (M1), Maedeh-Suresh (MS) and SPSS (SPSS). The random initialization, which is the default initialization of the original k-Means is referred to as k-M and is presented for comparative purposes. Tables 2, 3, 4, 5, 6, 7, 8 and 9 show the results of the k-Means algorithm, represented by the values of two validation indices, in the 8 data sets. For each data set the k-Means was initialized by each of the five initialization algorithms considered, as well as randomly (k-M), as the original k-Means. In the eight tables the quality of an induced clustering by the k-Means was measured by the validation indices Silhouette (S) [15] and Rand (R) [16]. Let P be a set of points to be clustered and consider two different clusterings of P, namely, A ¼ fA1 ; A2 ; . . .; ANA g and the other B ¼ fB1 ; B2 ; . . .; BNB g. The Rand index is defined as R = (a + b) / (a + b + c + d) where a: is the number of pairs of points in P that are in the same set in A and in the same set in B; b: is the number of pairs of points in P that are in different sets in A and in different sets in

Favoring the k-Means Algorithm with Initialization Methods

25

Table 1. Data sets characteristics where ID: data set identification, #NI: no. of data instances, #NA: no. of attributes, #NG: no. of groups and G_Id = #NI: no. of instances per group where G_id represents the group identification. The first four lines in the table refer to synthetic data and the last four, to real data. ID MSD LongSquare

#NI 30 900

#NA 2 2

#NG 5 6

Aggregation

788

2

7

3MC Iris Fossil Wine Seeds

400 150 87 178 210

2 4 6 13 7

3 3 3 3 3

G_id = #NI 1 = 4, 2 = 7, 3 = 10, 4 = 4, 5 = 5 1 = 147, 2 = 155, 3 = 150, 4 = 148, 5 = 150, 6 = 150 1 = 45, 2 = 170, 3 = 102, 4 = 273, 5 = 34, 6 = 130, 7 = 34 1 = 120, 2 = 170, 3 = 170 1 = 50, 2 = 50, 3 = 50 1 = 40, 2 = 24, 3 = 13 1 = 59, 2 = 71, 3 = 48 1 = 70, 2 = 70, 3 = 70

MSD

LongSquare

Aggregation

3MC

Fig. 2. Plotting of the four synthetic data sets used in the experiments.

26

A. F. de Oliveira and M. do Carmo Nicoletti

B; c: is the number of pairs of points in P that are in the same set in A and in different sets in B and d: is the number of pairs of points in P that are in different sets in A and in the same set in B. A detailed definition of the Silhouette index can be found in [15]. In the tables µ/r represents the average/standard deviation of each index value in 20 runs, taking into account the induced clusterings. The methodology for comparing the performances of the five k-Means initialization algorithms was implemented by sequentially executing the steps described below, for each data set (generically identified as X) described in Table 1 and for each algorithm, generically identified by Y. As mentioned before, the original k-Means using the default random initialization was also used for comparative purposes. The methodology approached the 6 algorithms (random choice (k-M) included) split into two sets: {k-M, ++, M1} and {CCIA, SPSS, MS}. The justification for that was the fact that all the three algorithms, k-M, ++ and M1, involve some random choice while the other three, CCIA, SPSS and MS, do not. The random choices occur when the original k-M chooses the k centroids in a random manner, when the ++ chooses the first centroid, and when the M1 makes a random selection of instances within a cell, as well as at its end, when certain conditions are not satisfied. For the k-M, ++ and M1 algorithms it was decided that steps (2.1), (2.2) and (2.3) (described next) should be performed 20 times; such number was chosen based on the experiments described in [3] and [6]. For the CCIA, SPSS and MS algorithms, which always select the same centroids for a given data set when executed more than once, only one execution was conducted for each of the three algorithms. The methodology adopted for the experiments has the following steps: (1) assign to k the number of visually identifiable groups (or classes) in X; (2) for algorithms {k-M, ++, M1} perform steps (2.1), (2.2) and (2.3) 20 times (due to the random choices involved) and, for the algorithms {CCIA, SPSS, MS}, since none of them use of random choices, perform the three steps just once. (2:1) Execute the initialization algorithm Y, using the set X, whose execution result is a set of k initial centroids C; (2:2) Execute k-Means without its original initialization step using, as input, in addition to set X and k, the set of centroids C created in step (2.1), obtaining the clustering AG; (2:3) Calculate the values of indices Silhouette and Rand, in the induced clustering AG, obtained in (2.2), and store them; (3) For the k-M, ++ and M1 algorithms, calculate the mean and standard deviation of the validation values as well as of the number of iterations performed by the k-M to converge, in the 20 runs performed in steps (2.1), (2.2) and (2.3). For the CCIA, SPSS and MS algorithms the values found in a single execution of steps (2.1), (2.2) and (2.3) are used, considering the three algorithms always have the same result when executed more than once having the same data set as input. The analysis of the results will focus on the values of both indices as well as on the number of iterations performed by k-Means initialized by each algorithm.

Favoring the k-Means Algorithm with Initialization Methods

27

As shown in Table 2, in the MSD data the ++ algorithm had one of the best performances among the six algorithms (i.e., ++, M1, k-M (random), CCIA, SPSS and MS), taking into account the values of the S and R validation indices as well the number of iterations of the k-Means, when initialized by ++. Table 2. MSD. AG

S µ/r ++ 0.9586/0.0457 M1 0.6782/0.1095 k-M 0.6041/0.0958 CCIA 0.3403 SPSS 0.6754 MS 0.6920

R µ/r 0.9586/0.0457 0.9587/0.0715 0.9302/0.0642 0.8391 0.908 1

I µ/r 2.400/0.4245 2.850/0.4018 4.500/2.1563 7 1 2

Results related to the R index, for the LongSquare, in Table 3 have values close to 1, an evidence that the induced clusterings are a very good approximation to the clustering visually detected. Since the LongSquare has 6 groups, with approximately 150 instances/group, the average number of iterations required for the k-Means to converge, when initialized randomly or via M1, was around 12 iterations, whereas when initiated by ++, required on average approximately 9 iterations to reach convergence. This is a data domain where, in particular, the validation indices R and S do not agree much. While the index R points to the induction of good clusterings, in spite of, sometimes, with a large number of iterations by the k-Means, the S index values indicate clusterings with average quality. Table 3. LongSquare. AG

S µ/r ++ 0.5162/0.0237 M1 0.5153/0.0237 k-M 0.4974/0.0216 CCIA 0.4645 SPSS 0.5309 MS 0.5309

R µ/r 0.9367/0.0290 0.9338/0.0259 0.9240/0.0205 0.8602 0.9445 0.9445

I µ/r 8.9000/4.1821 12.4500/4.7061 12.3000/6.1081 30 2 4

Taking into account the data in the Aggregation data set, 7 groups can be visually detected. On the one hand, the mean values of the S index, in Table 4, suggest that the clusterings obtained by k-Means, initialized by ++, M1 and randomly, were just average clusterings. On the other hand, however, the mean values of the R index show

28

A. F. de Oliveira and M. do Carmo Nicoletti

the induction of clusterings similar to those visually detected. The configuration and arrangement of data in the Aggregation is not particularly suitable for a grid-based approach, such as the one adopted by M1. When the M1 algorithm is not able to find centroids through its grid-based approach, it adopts a random choice, which is the same procedure adopted by the original k-Means. Particularly, in the Aggregation data domain, the M1 performance is similar to that of the original k-Means. Results related to the CCIA, SPSS and MS follow the same tendency of those of the ++, M1 and k-M. Table 4. Aggregation AG

S µ/r ++ 0.4716/0.0255 M1 0.4644/0.0301 k-M 0.4582/0.0175 CCIA 0.4843 SPSS 0.4919 MS 0.4047

R µ/r 0.9099/0.0158 0.9030/0.0157 0.8898/0.0089 0.9202 0.9110 0.8830

I µ/r 13.8500/3.6094 15.5000/5.7576 18.0000/6.8920 20 13 28

Considering the data set 3MC the mean values of the validation index R, in Table 5 point out to the induction of good clusterings, by the 6 algorithms. The values of the S index, however, suggest average clusterings. Meanwhile, the number of iterations performed by the k-Means point out to algorithm ++ and SPSS as those that provided better initializations. Note that the CCIA provided an initialization set to the k-Means that has not required any iteration (i.e., I = 0) for the k-Means to achieve convergence.

Table 5. 3MC AG

S µ/r ++ 0.5105/0.0051 M1 0.5066/0.0074 k-M 0.5051/0.0074 CCIA 0.5126 SPSS 0.4983 MS 0.4983

R µ/r 0.8970/0.0618 0.8541/0.0843 0.8367/0.0863 0.9230 0.7498 0.7498

I µ/r 6.8500/2.6509 7.8/3.9949 6.9500/2.6921 0 8 7

In relation to the Iris data set, the results presented in the first three lines of Table 6, related to validation indices as well as to the number of iterations performed by k-Means, initialized by ++, M1 and randomly, are not statistically very different from each other. In the last three lines of Table 6, except for the number of iterations, all the other values follow the same trend as those presented in the first three lines that is, they

Favoring the k-Means Algorithm with Initialization Methods

29

do not differ significantly. It can be observed, however, that the CCIA algorithm provided k-Means with the best set of 3 initial centroids, considering that k-Means required only 3 iterations to achieve convergence, whereas with the initial centroids provided by SPSS and by MS, the k-Means performed 7 and 9 iterations, respectively, to achieve convergence. In the Iris domain, in particular, the SPSS and MS initialization methods did not contribute much, considering that the original k-Means with the random initialization needed, on average, 6.25 iterations to converge.

Table 6. Iris AG

S µ/r ++ 0.4879/0.0216 M1 0.4783/0.0255 k-M 0.4889/0.0218 CCIA 0.5048 SPSS 0.4838 MS 0.4838

R µ/r 0.8556/0.0454 0.8397/0.0600 0.8559/0.0455 0.8737 0.8679 0.8679

I µ/r 5.4000/1.4966 5.5000/2.2693 6.2500/2.4469 3 7 9

Analyzing the data shown in Table 7 obtained using the Fossil data set, it can be noted that, with respect to the values of the R index, the initialization provided by ++, on average, slightly favored the quality of the obtained clusterings. However, in contrast, the average number of iterations required for k-Means to converge was slightly higher when compared to the average of iterations required for k-Means, when initialized by M1 or, then, randomly. During the experiment the M1 was not successful when using its grid-based approach, which caused the algorithm, in all the 20 runs, to finish the search for initial centroids using random choice, meaning that in the Fossil domain, the M1 behaved exactly like the original k-Means. With respect to the average values of the R index associated to the clusterings obtained with the initializations provided by ++, M1 and random, in the Wine data set, shown in Table 8, it can be said that all the induced clusterings were very close to the one visually detected, with a difference of 0.007, at most, of the clustering visually detected. However, neither of the two methods, ++ and M1, helped to promote the kMeans performance, in respect to the number of iterations, which reached better values when the random initialization of the k-Means itself, was used. The validation values of the R and the S indices, shown in Table 9, in the Seeds data set, are quite close to each other for the five algorithms plus k-M. The recurrent problem of the M1 algorithm, that of not being able to select a set of initial centroids by means of its grid-based approach, has again occurred in the Seeds data set, and the M1 ended up adopting the random selection, turning its execution very similar to that of the original k-Means.

30

A. F. de Oliveira and M. do Carmo Nicoletti Table 7. Fossil AG

S µ/r ++ 0.4664/0.0731 M1 0.3902/0.0987 k-M 0.4325/0.0903 CCIA 0.5022 SPSS 0.3839 MS 0.5022

R µ/r 0.9619/0.0784 0.8900/0.0955 0.9318/0.0880 1 0.8851 1

I µ/r 5.3000/3.3926 4.1000/1.7578 4.7500/2.6433 3 5 2

Table 8. Wine AG

S µ/r ++ 0.3030/0.0015 M1 0.3030/0.0011 k-M 0.3024/0.0005 CCIA 0.3025 SPSS 0.3028 MS 0.3028

R µ/r 0.9342/0.0083 0.9346/0.0062 0.9313/0.0030 0.9318 0.9349 0.9349

I µ/r 4.8500/1.7684 5.6000/2.3958 4.7000/1.6462 10 6 5

Table 9. Seeds AG

S µ/r ++ 0.4247/0.0004 M1 0.4247/0.0004 k-M 0.4245/0.0004 CCIA 0.4243 SPSS 0.4243 MS 0.4252

R µ/r 0.8667/0.0025 0.8670/0.0025 0.8677/0.0023 0.8693 0.8693 0.8642

I µ/r 6.1000/1.8947 7.3000/2.3043 6.4500/2.1324 4 3 6

5 Conclusions This chapter discusses and empirically evaluates the performance of five initialization algorithms as replacements for the k-Means original random initialization process. Taking into account the eight data sets used, it can be said that the results were not too conclusive so to support one particular algorithm. However, the CCIA and the SPSS initialization algorithms could be a good choice. Based on the work described in this chapter and taking into account all the experience acquired during the development of a software system that implements the five algorithms, a few possibilities for continuing in this line of research have been considered. (1) to use a combination of the centroids

Favoring the k-Means Algorithm with Initialization Methods

31

found by each of the five algorithms and by k-M (e.g., the arithmetic mean of the values found by the algorithms), and initialize k-Means with these values; and (2) to implement a variation of the process described in (1), weighting the individual performance of the five algorithms/data domain. Acknowledgments. The authors thank UNIFACCAMP and CNPq for their support. This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior − Brasil (CAPES) − Finance Code 001.

References 1. MacQueen, J.B.: Some methods for classification and analysis of multivariate observations, In: Proceedings of 5th Berkeley Symposium on Mathematical Statistics and Probability, pp. 281–297. University of California Press (1987) 2. Al-Daoud, M., Roberts, S.A.: New methods for the initialisation of clusters. Pattern Recogn. Lett. 17, 451–455 (1996) 3. Arthur, D., Vassilvitskii, S.: K-Means++: the advantages of careful seeding. In: Proceedings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms, pp. 1027–1035. Society for Industrial and Applied Mathematics, USA (2007) 4. Maedeh, A., Suresh, K.: Design of efficient k-Means clustering algorithm with improved initial centroids. Int. J. Eng. Technol. 5(1), 33–38 (2013) 5. Khan, S.S., Ahmad, A.: Cluster center initialization algorithm for k-Means clustering. Pattern Recogn. Lett. 25, 1293–1302 (2004) 6. Pavan, K.K., Rao, A.A., Rao, A.V.D., Sridhar, G.R.: Robust seed selection algorithm for kmeans type algorithms. Int. J. Comput. Sci. Inform. Technol. (IJCSIT) 3(5), 147–163 (2011) 7. Aggarwal, C.C., Reddy, C.K.: Data Clustering Algorithms and Applications. Chapman & Hall/CRC Data Mining and Knowledge Discovery Series. CRC Press, Boca Raton (2013) 8. Han, J., Kamber, M., Pei, J.: Data Mining – Concepts and Techniques, 3rd edn. Morgan Kaufmann Publishers, Amsterdam (2012) 9. Burks, S., Harrell, G., Wang, J.: On initial effects of the k-Means clustering, In: Proceedings of the 2015 World Congress in Computer Science, Computer Engineering, & Applied Computing, USA, pp. 200–205 (2015) 10. Dua, D., Karra Taniskidou, E.: UCI Machine Learning Repository (http://archive.ics.edu/ml). University of California, School of Information and Computer Science, Irvine, CA (2017) 11. Chernoff, H.: The use of faces to represent points in n-dimensional space graphically, Technical report no. 71, Department of Statistics. Stanford University, Stanford, CA, USA (1971) 12. Hand, D.J., Daly, F., Lunn, A.D., McConway, K.J., Ostrowski, E.: Handbook of Small Data Sets, 1st edn. Chapman and Hall/CRC, London (1993) 13. Gionis, A., Mannila, H., Tsaparas, P.: Clustering aggregation. ACM Trans. Knowl. Discovery Data 1(1) (2007). https://doi.org/10.1145/1217299.1217303, http://doi.acm.org/ 10.1145/1217299.1217303, Article 4, 30 pages 14. Su, M.C., Chou, C.H., Hsieh, C.C.: Fuzzy C-Means algorithm with a point symmetry distance. Int. J. Fuzzy Syst. 7(4), 175–181 (2005) 15. Rousseeuw, P.: Silhouettes: a graphical-aid to the interpretation and validation of cluster analysis. Comput. Appl. Math. 20, 53–65 (1987) 16. Rand, W.M.: Objective criteria for the evaluation of clustering methods. J. Am. Stat. Assoc. 66(336), 846–850 (1971)

A Novel Design and Implementation of 8-Bit and 16-Bit Hybrid ALU Suhas B. Shirol(&), S. Ramakrishna, and Rajashekar B. Shettar Department of Electronics and Communication Engineering, B V Bhoomaraddi College of Engineering and Technology, Hubballi 580031, Karnataka, India [email protected], {ramakrishna,raj}@bvb.edu

Abstract. Arithmetic Logic Unit (ALU) are main constructing block of several digital computation systems like digital calculator, mobile phone, high computational computer, Digital signal processors etc. In current scenario electronic market as technology is raising every day, Fast rising technologies with handy devices demands for low power VLSI design. Ultimatum design with less delay, low power and low area is increasing. Reversible logic gates are suitable to minimize the power dissipation in the circuit, designed reversible logic is suitable. ALU design is intended with both reversible logic gates and irreversible logic gates to reduce dissipation, switching power and delay. The proposed type of design are said to be Hybrid ALU architecture. In arithmetical adder, time taken to propagate carry are minimized by using KSA and CSA, BEC is used to minimize area instead of Ripple Carry Adder. Adder design are utilized to add partial products in Vedic-multiplier, which minimizes delay in digital multiplier. Keywords: Arithmetic Logic Unit (ALU)
Carry Select Adder (CSA) Kogge-Stone Adder (KSA)
Binary to Execs one Converter (BEC)
Ripple Carry Adder (RCA)
Hybrid ALU




1 Introduction ALU plays important role in the digital computations. It is part of digital calculator, computer CPU. ALU perform arithmetic operation, Logical operation such as AND, OR, EXOR, NAND, NOR and Shift operation. ALU consists of digital adder, division, subtractor, multiplier, logical operations and shift operations [4]. In present digital electronic market as technology is updating day by day, demanding for minimizing delay, dissipation of power and area is the need of the hour and plays vital role in market. Carry Select Adder architecture are used to minimize time taken to transmit carry in digital adder, Binary to Execs one Converter (BEC) architecture is used to minimize the area instead of Ripple Carry Adder (RCA). The adding of partial products in Vedic-multiplier using adders mentioned above which in turn minimize delay in multiplier [2]. Reckless increasing technologies with digital designs demands for low power dissipation in digital VLSI design. Issues with low power digital VLSI design is power dissipation, the solution for this problem are given by reversible logic [3]. Hence, ALU are designed to minimize the delay and power dissipation using both © Springer Nature Switzerland AG 2020 A. Abraham et al. (Eds.): ISDA 2018, AISC 940, pp. 32–42, 2020. https://doi.org/10.1007/978-3-030-16657-1_4

A Novel Design and Implementation of 8-Bit and 16-Bit Hybrid ALU

33

irreversible logic gate and reversible logic gates. We can propose such design as Hybrid ALU. ALU the heart of any computational system such as digital designs and formulates significant part of computer, mobile phones, processors, MAC unit. Since in today’s digital market minimization of delay and power dissipation plays vital role.

2 Literature Survey Adder is essential sub-circuit of any ALU. Carry propagation contribute to delay in adder. Modified Carry Select Adder (CSA) are used to minimize delay, BEC are used to lessen the area and a substitute of using RCA with carry input as one [1, 8, 9]. In any digital computing system, multiplier design discovers huge importance in video processing, DFT, DSP, security and intelligent system where the computations are very high. In this, paper [2, 12] multiplier is designed using one of the popular Vedic sutra called Urdhava Triyagbhyam, Adding the partial products of multiplier is made using K-stone, Which is one the quickest adder, which minimizes delay and Vedic sutra enhance well-designed competence of multiplier. In this paper [3, 10–12] area efficient and delay efficient is discussed with modified CSA with BEC. Outcome of this modified CSA, Vedic multiplier design is delay and area competent. ALU is vital sub-system of different systems like MAC unit, computational systems, DSP etc. Reversible logic gates had grew enormous significance in low power VLSI industry as it has capability to minimize power dissipation in digital design. Power dissipation is more in irreversible logic gates when compared to reversible logic gates [4, 13]. In this paper, [5–8] the minimization of delay and power dissipation were discussed for K-stone adder design using reversible logic gate. Reversible logic gate has no data loss as one to one mapping between inputs and outputs. This is the main reason for minimization of power dissipation, hence reversible gates used in low power VLSI design. Hardware utilization are focused using code converters in improving system by minimizing complexity in the circuits. This paper discusses about hardware reduction, speed and power [6].

3 Design Methodology The ALU does the internal arithmetic operation of data in the digital processor. The opcode input is used to governor the arithmetic actions such as add, sub, mul, div, logical actions and shift actions accomplished by the ALU. A diagrammatical depiction of an ALU is shown in Fig. 1 Inputs and Outputs to ALU: 1. Operands: Two n bit integer inputs on operations are to be performed. 2. Status (Input): 1 bit input. It is carry in for addition or borrow in for subtraction. 3. Status (Output): One bit output. It is carry out from addition or borrow out from subtraction.

34

S. B. Shirol et al.

Fig. 1. Block diagram of ALU

4. Opcode: It is a code that decides operation to be performed on the operands. Number of bits in the opcode is decided by number of operation performed by ALU. If 2n operations to be performed by ALU, then opcode is n bit. 5. Result: Result of operation performed by ALU on the operands. 3.1

Design and Implementation

The below Fig. 2 shows the design and implementation of the 8 bit and 16 bit ALU.

Fig. 2. ALU implementation

3.2

Reversible Logic Gates

Reversible logic circuits are the circuits where inputs are recuperated from outputs. Features: (i) Number of Input/Output are equivalent. (ii) Input/Output vectors are mapped one to one. (iii) Fan-out is minimum.

A Novel Design and Implementation of 8-Bit and 16-Bit Hybrid ALU

35

Basic reversible logic gates are as follows: Nevertheless, reversible logic gate transfers charge from one node to another node. In this paper, the reversible logic gates used in the design are PERES, FEYNMAN, TOFFOLI, FREDKIN, MTSG and TR gate as shown in Figs. 3, 4, 5, 6, 7 and 8. • PERES gate: The below Fig. 3 shows PERES gate which has 3 input gate mapped with 3 output gate. PERES gate is used to implement Half Adder(HA) by making input as c = 0.

Fig. 3. PERES gate

• FEYNMAN gate: The below Fig. 4 shows FEYNMAN gate. The 2-inputs are A and B, the 2-outputs are P and Q. The output of Feynman Gate is XOR operation.

Fig. 4. FEYNMAN gate

• TOFFOLI gate: The below Fig. 5 shows TOFFOLI gate. The 3-inputs and 3outputs are A, B, C, P, Q and Z.

Fig. 5. TOFFOLI gate

• FREDKIN gate: The below Fig. 6 shows FREDKIN gate. The Fredkin Gate has 3outputs and 3-inputs. The input are A, B, C and the outputs are P, Q, R.

36

S. B. Shirol et al.

Fig. 6. FREDKIN gate

• MTSG gate: The below Fig. 7 shows MTSG Gate. The MTSG gate has 4-input and 4-output.

Fig. 7. MSTG gate

• TR gate: The below Fig. 8 shows TR gate. It has 3- inputs and 3-outputs.

Fig. 8. TR gate

4 Arithmetic Operations Using Hybrid Design 4.1

Adder

An adder plays vital role in digital circuit to accomplish addition of numbers. ALU and many computing computers and processors uses adders as basic building block. Adders also heart of the processor, adders are used to increment and decrement operators, and similar operations.

A Novel Design and Implementation of 8-Bit and 16-Bit Hybrid ALU

37

Therefore, adder with minimum delay and least power dissipation is designed using reversible logic gates. There are many type’s adders Parallel Prefix Adder (PPA), Carry Select Adder (CSA), CSA, RCA etc. 4.1.1 Ripple Carry Adder (RCA) Cascading of full adder give rise to Ripple Carry Adder (RCA). When full adder are cascaded, carry in of one full adder resultant is driven by the carry out of the driving full added. As the complexity of RCA is increased since the carry bit is propagated the delay time taken to propagate. 4.1.2 Carry Select Adder (CSA) CSA is combinational adder circuit, it is made of RCA and MUX. Delay minimization in RCA as carry generation and propagation is not taken into consideration, as pre calculated value of carry either one or zero. Value of carry are selected using MUX. For summing circuit any adder circuit like Carry Skip Adder, Kogge-Stone Adder, etc., can exchange RCA with other adders. CSA is shown in Fig. 9.

Fig. 9. Carry Select Adder

4.1.3 Kogge-Stone Adder (KSA) When the speed of the system required is high with performance, KSA is the adder, which discovers numerous application in VLSI engineering. KSA gives us the area efficient design with speed performance but trade-off for the power dissipation of the circuit. 4.1.4 Binary to Excess One Converter (BEC) Code converters decreases the difficulty of arithmetic circuits. Which helps in decreasing area and delay. Hence increase performance of the circuit 4-bit RCA adder uses 4 OR gate, 8 AND gate and 8 XOR gate. Where 5-bit BEC require 1 inverter, 4 XOR gate and 3 AND gate. From comparison, we can infer that BEC take less hardware area. Figure 10 below shows the circuit of BEC.

38

S. B. Shirol et al.

Fig. 10. 5-bit-Binary-to-Excess-1-Code-Converter

4.1.5 Modified Carry Select Adder To minimize the area and delay we use modified CSA. Design of Improved 8 bit and 16 bit CSA. Improved 8 bit and 16 bit CSA involves of two 4 bit KSA, 8 bit KSA and BEC. 4.1.6 Reversible Modified CSA Improved CSA are designed to attain least delay and area. In order, to minimize power dissipation the modified CSA are realized using reversible logic gate. Design of BEC includes Reversible logic gate such as PERES gate when cascaded to form Binary to excess converter (BEC). Figure 11 below shows the reversible modified CSA.

Fig. 11. BEC using reversible logic gate

5 Multiplier Multiplier is combinational logic circuit, which is used to multiply two binary integers. Multipliers are building blocks of DSP, ALU, FFT, Matrix multiplication. There are various forms of multipliers such as Vedic multiplier, array and booth multipliers. If speed performance of Multipliers is improved, it is probable to minimize the processing time needed by DSP units and ALU. In prehistoric Vedic mathematics, various sutras were considered for multiplication. Between those sutras, Urdhva Tiryakbhyam sutra is considered as more efficient multiplier. In Urdhva Tiryakbhyam, adders are used to add

A Novel Design and Implementation of 8-Bit and 16-Bit Hybrid ALU

39

the partial products, thus speed and area of multiplier is efficient. Thus, to have least delay in our design, we have used modified CSA adder to add the partial products. Figure 12 represent architecture of 8  8 and 16  16 Vedic multiplier. From architecture, it is found that in addition of partial products, addition of carry is not needed so modified CSAs area and delay are reduced by replacing KSA with RCA and half adders. Design of modified CSA with RCA and half adders.

Fig. 12. 8  8 and 16  16 Vedic multiplier

6 Results and Analysis Thus, the design of 8-bit ALU and 16-bit ALU is implemented using reversible and irreversible logic as mentioned above. The Coding of design is done by using Verilog HDL code. The code is simulated and synthesized by expending Xilinx 14.4 ISE, using Verilog language. 8 and 16 bit adders, multipliers, subtractor is designed using both reversible and irreversible logic gates using Verilog. Delay, Area and power are compared for 8 and 16 bit Hybrid ALU implemented using chip scope pro on Spartan 6 FPGA. The results are represented in below sections. Table 1 represent device utilization summary result of operations for 8-bit and 16-bit Hybrid ALU.

40

S. B. Shirol et al. Table 1. Device utilization summary (Selected Device: 6slx16csg324-3)

Slice logic utilization Number of Slice LUTs Number used as Logic Slice logic distribution Number of LUT Flip Flop pairs used Number with an unused Flip Flop Number with an unused LUT Number of fully used LUT-FF pairs Number of unique control sets IO utilization Number of IOs Number of bonded IOBs IOB Flip Flops/Latches Total delay

8-bit ALU Utilization by % utilization by design design 251 out of 9112 2

16-bit ALU Utilization by % utilization by design design 950 out of 9112 10

251 out of 9112

950 out of 9112

2

251

10

950

251 out of 251

100

950 out of 950

100

0 out of 251

-

0 out of 950

-

0 out of 251

-

0 out of 950

-

1

38 38 out of 232 1 Logic: 6.444 ns, 33.7% logic 19.133 ns

-

1

16 Route: 12.689 ns, 66.3% route

-

70 70 out of 232 1 Logic: 8.265 ns, 27.9% logic 29.640 ns

30 Route: 21.375 ns, 72.1% route

Analysis of Results 8-bit and 16-bit Hybrid ALU design comparison: Maximum obtained path de-lay, LUTs, IO Buffers, Logic delay(%) and Route delay(%) comparison in Figs. 13 and 14.

A Novel Design and Implementation of 8-Bit and 16-Bit Hybrid ALU

41

Fig. 13. 8-bit ALU comparison

Fig. 14. 16-bit ALU comparison

7 Conclusion The 8-bit and 16-bit ALU designed here makes use of reversible logic gates, which gives minimum delay, Modified CSA with reversible logic gates for addition, which contributes for minimization of delay, area. Vedic multiplier in this paper is designed by using Modified CSA using BEC, which minimize the area and time. Subtractor is

42

S. B. Shirol et al.

designed using reversible logic gates. The used adder model, multiplier model and subtractor model reduces the time involved in the operation and thus increases the speed. The design is implemented on FPGA. The whole design is coded by using Verilog HDL. The use of Xilinx 14.2 ISE to simulate and synthesize design using Verilog language.

References 1. Nautiyal, P., Madduri, P., Negi, S.: Implementation of an ALU using modified carry select adder for low power and area-efficient applications. In: 2015 International Conference on Computer and Computational Sciences (ICCCS), pp. 22–25. IEEE (2015) 2. Sudeep, M., Vucha, M., et al.: Design and FPGA implementation of high speed vedic multiplier. Int. J. Comput. Appl. 90(16) (2014) 3. Shirol, S., Ramakrishna, S., Shettar, R.: A novel design and implementation of 32-bit hybrid ALU. In: Second International Research Symposium on Computing and Network Sustainability (IRSCNS 2018) (2018) 4. Swamynathan, S., Banumathi, V.: Design and analysis of FPGA based 32 bit ALU using reversible gates. In: 2017 IEEE International Conference on Electrical, Instrumentation and Communication Engineering (ICEICE), pp. 1–4. IEEE (2017) 5. Gokhale, G., Gokhale, S.: Design of area and delay efficient vedic multiplier using carry select adder. In: 2015 International Conference on Information Processing (ICIP), pp. 295– 300. IEEE (2015) 6. Elangadi, S., Shirol, S.: Design and characterization of high speed carry select adder. Int. J. Ethics Eng. Manag. Educ. 1(6), 35–40 July 2014 7. Banerjee, A., Das, D.K.: A new ALU architecture design using reversible logic. In: 2016 Sixth International Symposium on Embedded Computing and System Design (ISED), pp. 187–191. IEEE (2016) 8. Pattnaik, S.K., Nanda, U., Nayak, D., Mohapatra, S.R., Nayak, A.B., Mallick, A.: Design and implementation of different types of full adders in ALU and leakage minimization. In: 2017 International Conference on Trends in Electronics and Informatics (ICEI), pp. 924– 927. IEEE (2017) 9. Shirol, S., Ramakrishna, S., Shettar, R.: Design and implementation of adders and multiplier in FPGA using Chipscope: a performance improvement. In: Information and Communication Technology for Competitive Strategies. Lecture Notes in Networks and Systems. https:// doi.org/10.1007/978-981-13-0586-3_2 10. Saha, P., Banerjee, A., Bhattacharyya, P., Dandapat, A.: High speed ASIC design of complex multiplier using vedic mathematics. IEEE (2011) 11. Ramkumar, B., Kittur, H.M.: Low-power and area-efficient carry select adder. IEEE Trans. Very Large Scale Integr. (VLSI) Syst. 20(2), 371–375 February 2012 12. Jais, A., Palsodkar, P.: SM-IEEE: Design and Implementation of 64 Bit Multiplier using Vedic Algorithm, 6–8 April 2016. IEEE (2016) 13. Kandasamy, N., Telagam, N., Devisupraja, C.: Design of a low-power ALU and synchronous counter using clock gating technique. In: Saeed, K., Chaki, N., Pati, B., Bakshi, S., Mohapatra, D. (eds.) Progress in Advanced Computing and Intelligent Engineering. Advances in Intelligent Systems and Computing, vol. 564. Springer, Singapore (2018)

Authentication Scheme Using Sparse Matrix in Cloud Computing Sunita Meena(&), Shivani Kapur, Vipin C. Dhobal, and Subhrat Kr. Sethi C.R.L, B.E.L, Ghaziabad, India {sunitameena,shivanikapur,vipinchandradobhal, subratkrsethi}@bel.co.in

Abstract. Cloud computing is boon in technology field which provide ondemand services in IT domain. Security and authentication threats are challenges associated with storage and access of data in cloud environment. Authentication in cloud computing is a major concern with the increase in cloud user base. The authentication schemes in cloud computing are based on hash function, biometric, logic functions to keep the data in secure and safe manner. This paper compares the different methods of authentication schemes in terms of complexity and the strength to prevent attacks and proposes a new improved method using sparse matrix approach performing better than simple matrix approach. Our method improves the usability of full matrix approach converting it to sparse matrix to authenticate client for use of cloud service. Keywords: Cloud security
Authentication
Encryption Attacks
Cloud computing
Matrix
Cryptography


Decryption


1 Introduction Cloud computing is boon for IT industries which is developed from the grid and distributed computing conceptually. It is a network of computers, connected via internet, sharing, transferring, various resources provided by cloud service providers offering the scalability, usability, resource requirements. The main aim for cloud computing is to provide new world for users handling their data with province of cloud to enable ease usage of input and output devices busk in the powerful computing capacity of cloud on-demand [1]. 1. Cloud Computing provides users to store, manage, and process data through internet instead of a personal computer or a local server. 2. Rapidly deploy and increase workload by speedy providing physical machined or virtual machines. 3. On-demand capabilities allows user to add, delete, and modify data stored and software according to the requirements in cloud data storage. 4. Work in different workloads, including the back-end operations and user’s interactive applications.

© Springer Nature Switzerland AG 2020 A. Abraham et al. (Eds.): ISDA 2018, AISC 940, pp. 43–52, 2020. https://doi.org/10.1007/978-3-030-16657-1_5

44

S. Meena et al.

5. Access to applications and services through different devices are on ease and cloud being operated whenever, wherever through internet.

2 Issues in Cloud Security Cloud Computing provides the enhancement of business. With this facility comes the security concern but as more and more information on individuals and companies is placed in the cloud, concerns are beginning to grow about just how safe an environment it is [1]. 1. Security Some argue that customer data is more secure when managed internally, while others argue that cloud providers have a strong incentive to maintain trust and as such employ a higher level of security. 2. Privacy Different from the traditional computing model, cloud computing utilizes the virtual computing technology, user’s personal data may be scattered in various virtual data center rather than stay in the same physical location, even across the national borders, at this time, data privacy protection will face the controversy of different legal systems [2]. 3. Reliability Servers in the cloud have the same problems as your own resident servers. The cloud servers also experience downtimes and slowdowns, what the difference is that users have a higher dependent on cloud service provider-CSP in the model of cloud computing. 4. Legal Issues Regardless of efforts to bring into line the lawful situation, as of 2009, supplier such as Amazon Web Services provide to major markets by developing restricted road and rail network and letting users to choose “availability zones”. On the other hand, worries stick with safety measures and confidentiality from individual all the way through legislative levels [18–22].

3 Problem Formulation and Literature Survey As cost efficiency, unlimited storage, backup and recovery, automatic software integration, easy access to information stand out as advantages, security issues stand out as the major disadvantages of this new technology [3]. With increase in large access to cloud storage its security is getting a major concern in today’s world. Data breach, loss of data with millions of information is of great concern. A study conducted by the Ponemon Institute entitled “Man In Cloud Attack” reports that over 50% of the IT and security professionals surveyed believed their organization’s security measures to protect data on cloud services are low. Attackers now have the ability to use our (or ours employees’) login information to remotely access sensitive data stored on the

Authentication Scheme Using Sparse Matrix in Cloud Computing

45

cloud; additionally, attackers can falsify and manipulate information through hijacked credentials. To protect cloud information g, authentication is the one of most important factors. In cloud computing still there is a need for well-defined authentication mechanisms. One of the first steps toward securing an IT system is to verify the identity of its users. The process of verifying a user’s identity is typically referred to as user identification and authentication [5]. Authentication is generally referred to as a mechanism that establishes the validity of the claimed identity of the individual (Fig. 1).

Fig. 1. Cloud security flow diagram

Several researchers are working to find strong authentication methods for cloud computing. A critical review of various research works is carried out. Several frameworks, models and architectures have been proposed by researchers [23–29]. Some of the prominent works in this area are covered in this section. One of the authentication architectures was proposed by Chow et al. [2]. The architecture is based on the method “what an individual does.” They proposed an implicit authentication method for mobile users. This authentication architecture is based on the history of the websites that the user visits. On one hand it is convenient and easy to use; on the other hand, it cannot be used as a replacement for regular authentication in high-risk sectors, like banking. Celesti et al. [3] proposed a reference architecture to address the identity management problem for Cloud computing. As the next development in research a new method of authentication was proposed that implemented mobile Out of Bound authentication on the Public Key Infrastructure during the login phase. To define a new message exchange flow between the entities involved within ICIMI, it has implemented a new Security Assertion Markup Language (SAML) profile defining the interaction among the home cloud authentication module(s), the foreign cloud authentication module(s) and the IDP(s). Complexity is increased and authenticating the client

46

S. Meena et al.

dependent on trust value. Dependence on making trust relationship between three clouds could be targeted with unknown behaving like a relying party and asserting party. An improved mutual authentication framework has been proposed by Kumar et al. [5]. In their research they proposed a scheme in which there are three phases. In the first phase the secret key initialization is made by the service provider to the user. In the second phase the user is registered by double authentication. In the third phase authentication is done using a nonce. The proposed scheme can also resist many attacks such as password stolen, replay attack, etc. There are some cases that need to use anonymous authentication in cloud computing. In such cases the user does not want to reveal their identity, they just want that service providers know that they are legitimate users. Zhi-hua [6] authors solve this problem by proposing non-authentication certificate anonymous scheme. The scheme proposed is based on the computational DiffieHellman problem (CDHP). The scheme presented does not have a certification center and it avoids the key Revocation and the key escrow problems in the authentication schemes based on public key certificate. Gonzalez [7] proposes a framework for studying and developing a relationship between cloud deployment models, service types, entities and lifecycle controls. The patterns of previous privacy leaks can also be used in preventing the new authentication attacks in cloud computing. Han et al. [9] proposed data storage security through a novel third party auditor scheme in cloud computing using RSA and Diffie-Hellman techniques. They design a message header and series which is attached to the data to ensure the authorization of data after transmission. Revar et al. [10] proposed a single sign-on scheme as a means of authentication in cloud environment. A single authentication allows the client to gain access to all the resources. This SSO is implemented in the cloud architecture’s top layer. In this scheme the Authentication Server is the main component that provides single sign on. This enhances the reliability, adaptability and feasibility of the Cloud. In the field of fingerprint recognition, Wang [13] proposed a system based on secret splitting of finger print biometric data. In their approach they divided and stored part of finger print data on smart card and part of it on the server. This makes attack more difficult as attacker needs to break two keys rather than one. Another biometric method in cloud computing is the face recognition. Javaid et al. [15] proposing that model credentials travel encrypted between the clients and the cloud host using public internet with an optimum performance. Different from conventional model they introduce the dedicated VPN supported firewall between cloud host and the clients to make all traffic passed by the tunnel. It configures the different services: A web server, FTP server and a storageaserver. This saves data from getting stolen through hardware or any other Trust Matrix Elements.

Authentication Scheme Using Sparse Matrix in Cloud Computing

47

4 Proposed Model Islam et al. [16] used a trust matrix using swarm intelligence in cloud computing. Trust matrix is generated using the input by user which is verified by the ant formed on three level, i.e. user, cloud data storage (CDS), cloud service provider (CSP). At each level ants keep check on the trust matrix. The authentication of user can be dropped at any level if matrix is found noncomparative. It lacks at the overhead of the matrix formation, which increase the computation of trust matrix. With increase in fault/nearer to zero value it impacts the computation time, which decrease the key value used for authentication (Figs. 2 and 3).

Fig. 2. Flow chart of security framework action (Base Paper) [16]

48

S. Meena et al.

Start

Each model contains trust value

CSP and CDS lay forms a trust matrix

Trust Matrix in CSP Layer

Conversion from Full Matrix to sparse matrix

END Yes

Continue Fig. 3. Proposed model

Authentication Scheme Using Sparse Matrix in Cloud Computing

49

U ¼ Ut  Ua R ¼ r  ðUt =Ua Þ X ic G¼ U  R=T t :

1. 2. 3. 4. 5.

Where Ut is trust degree and Ua is corresponding CU’s authentication degree. R is the degree of request potentiality. r is the priority of current request which has been decided by the CSPA. Gk;i is the trust ranking degree of the K th , UA at ti time slot. Trust Matrix is CSP layer is where T is n times values between current time and threshold time.

Our proposed Model is based on improving the computation time and cost reduction using sparse matrix approach for the Islam et al. [16] proposed trust matrix. Trust Matrix used for computing CSP Layer and CDS Layer uses the Trust Matrix including the faulty values. Our proposed Model uses the non-faulty values and computes the density decreasing the execution time and CPU time.

5 Results See Figs. 4 and 5.

Performance results for sparse vs. full matrix operation

Fig. 4. Performance comparison between sparse vs. full matrix operation

50

S. Meena et al.

Fig. 5. Performance of sparse matrix operation with increase in density with efficiency (For the comparison between the result of full and sparse matrices cpu and elapsed time computation with density range 0.02 to 0.2 are Perf. Results for full [A * inv(B)] with Total exec. Time = 0.108000 and Total cpu. Time = 0.234375. Perf. Results for sparse [A * inv(A)] with Total exec. Time = 0.031000 and Total cpu. Time = 0.062500.)

6 Conclusion and Future Scope Use of Authentication plays a vital role in Cloud Computing on both the client and server sides. These operations contain different schemes with different computational costs. The major drawbacks are computation costs and execution delay during operations of these schemes. Using the Sparse matrices in place of full matrices performs better. With the increase in densities in both the full and sparse matrices, the sparse matrices show the better result on comparison of densities, elapsed time and cpu execution time. It lowers the computation costs. The full matrix value storage in cloud is higher than the sparse matrices. The sparse matrix is used in many fields like image processing. This is an emerging technology, but due to high storage value when densities are high (i.e. number of non-zero elements are high) it takes high computation. The present research work can be extended to design and develop new methodology to meet the following additional desirable features. Robust Scenario- Learning process approached must work with robust scenarios where matrix is dense, mobility is high, and area is large. Security- Security is a vital issue in authentication. While forming a matrix it needs focus of researcher. There are some applications which result high degree of security against enemies and active/ passive eavesdropping attackers.

Authentication Scheme Using Sparse Matrix in Cloud Computing

51

References 1. Yang, J., Chen, Z.: Cloud computing research and security issues. In: 2010 International Conference on Computational Intelligence and Software Engineering (CISE), Wuhan (2010) 2. Chow, R., Jocobsson, M., Masuoka, R., Molina, J., Niu, Y., Shi, E., Song, Z.: Authentication in the clouds, a framework and its application to mobile users. In: CSSW10, Chicago, Illinois, USA, 8 October 2010 (2010) 3. Celesti, A., Tusa, F., Villari, M., Puliafito, A.: Security and cloud computing: inter cloud identity management infrastructure. In: 19th IEEE International Workshop on Enabling Technologies: Infrastructures for Collaborative Enterprises (WETICE) (2010) 4. Lee, S., Ong, I., Lim, H.T., Lee, H.J.: Two factor authentication for cloud computing. Int. J. KIMICS 8, 427 (2010) 5. Nayak, S.K., Mohapatra, S., Majhi, B.: An improved mutual authentication framework for cloud computing. Int. J. Comput. Appl. 52(5) (2012) 6. Zhi-hua, Z., Jian-jun, L., Wei, J., Bei, Z.Y.G: A new anonymous authentication scheme for cloud computing. In: The 7th International Conference on Computer Science & Education (ICCSE 2012), 14–17 July 2012 (2012) 7. Gonzalez, N.M., Rojas, M.A.T., da Silva, M.V.M., Redígolo, F.: A framework for authentication and authorization credentials in cloud computing. In: 12th IEEE International Conference on Trust, Security and Privacy in Computing and Communications (2013) 8. Banyal, R.K., Jain, P., Jain, V.K.: Multi-factor authentication framework for cloud computing. In: Fifth International Conference on Computational Intelligence, Modeling and Simulation. IEEE (2013) 9. Han, S., Xing, J.: Ensuring data storage security through a novel third party auditor scheme in cloud computing. IEEE CCIS (2011) 10. Revar, A.G., Bhavsar, M.D.: Securing user authentication using single sign-on in cloud computing. In: International Conference on Engineering (NUiCONE) (2011) 11. Le, Z., Xiong, N., Yang, B, Zhou, Y.: SC-OA: a secure and efficient scheme for origin authentication of inter domain routing in cloud computing networks. In: IEEE International Parallel & Distributed Processing Symposium (2011) 12. Zhu, H.-H., He, Q.-H., Zhu, H.-H., Tang, H., Cao, W.-H.: Voiceprint-biometric template design and authentication based on cloud computing security. In: IEEE International Conference on Cloud and Service Computing (2011) 13. Wang, P., Ku, C.C., Wang, T.C.: A New Fingerprint Authentication Scheme Based on Secret-Splitting for Enhanced Cloud Security (2011) 14. Pawle, A.A., Pawar, V.P.: Face recognition system (FRS) on cloud computing for user authentication. Int. J. Soft Comput. Eng. (IJSCE) 3(4) (2013) 15. Javaid, Z., Ijaz, I.: Secure user authentication in cloud computing. In: 5th International Conference on Information & Communication Technologies (ICICT) (2013) 16. Islam, Md.R.: Collaborative swarm intelligence based trusted computing. In: IEEE/OSA/IAPR International Conference on Informatics, Electronics and Vision (2012) 17. Singh, S.: Secured user’s authentication and private data storage access scheme in cloud computing using elliptic curve cryptography. In: 22nd International Conference on Computing for Sustainable Global Development (INDIACom) (2015) 18. Yassin, A.A., Hussain, A.A., Mutlaq, K.A.-A.: Cloud authentication based on encryption of digital image using edge detection. In: International Symposium on Artificial Intelligence and signal Processing (AISP) (2015) 19. Maxwell, J.C.: A Treatise on Electricity and Magnetism, vol. 2, 3rd edn, pp. 68–73. Clarendon, Oxford (1892)

52

S. Meena et al.

20. Manuel, P.D., Selvi, S.T., Ibrahim, M., Barr, A.-E.: Trust Management System for Grid and Cloud Resources. Research grant by Kuwait University, IEEE (2009) 21. Park, K.-W., Park, S.K., Han, J., Park, K.H.: THEMIS: towards mutually verifiable billing transactions in the cloud computing environment. In: IEEE 3rd International Conference on Cloud Computing (2010) 22. Squicciarini, A., Sundareswaran, S., Lin, D.: Preventing information leakage from indexing in the cloud. In: IEEE 3rd International Conference on Cloud Computing (2010) 23. Shamshirband, S., Patel, A., Anuar, N.B., Kiah, L.M., Abraham, A.: Cooperative game theoretic approach using fuzzy q-learning for detecting and preventing intrusions in wireless sensor networks. Eng. Appl. Artif. Intell. 32, 228–241 (2014) 24. Tiwari, A., Sanyal, S., Abraham, A., Knapskog, S.J., Sanyal, S.: A multifactor security protocol for wireless payment-secure web authentication using mobile devices. In: Guimaraes, N., Isaias, P. (eds.) International Conference on Applied Computing 2007, Salamanca, Spain, pp. 160–167 (2007). ISBN 978-972-8924-30-0 25. Eid, H., Darwish, A., Hassanien, A.E., Abraham, A.: Principle components analysis and support vector machine based intrusion detection system. In: Tenth International Conference on Intelligent Systems Design and Applications (ISDA 10), pp. 363–367. IEEE, USA (2010). ISBN 978-1-4244-8136-1 26. Haslum, K., Abraham, A., Knapskog, S.: DIPS: a framework for distributed intrusion prediction and prevention using hidden Markov Models and online fuzzy risk assessment. In: Third International Symposium on Information Assurance and Security, pp. 183–188. IEEE Computer Society press, USA (2007). ISBN 0-7695-2876-7 27. Panda, M., Abraham, A., Patra, M.: Discriminative multinomial Naive Bayes for network intrusion detection. In: Sixth International Conference on Information Assurance and Security (IAS), USA, pp. 122–127. IEEE (2010). ISBN 978-1-4244-7408-0 28. Huang, H.C., Chen, Y.H., Abraham, A.: Optimized watermarking with swarm-based bacterial foraging. J. Inf. Hiding Multimed. Sig. Process. 1(1), 51–58 (2010) 29. Herrero, A., Corchado, E., Pellicer, M., Abraham, A.: MOVIH-IDS: a mobile-visualization hybrid intrusion detection system. Neurocomputing J. 72(13–15), 2775–2784 (2009)

A Thermal Imaging Based Classification of Affective States Using Multiclass SVM C. M. Naveen Kumar(&) and G. Shivakumar(&) Department of E&I Engineering, Malnad College of Engineering, Hassan, Karnataka, India {cmn,gs}@mcehassan.ac.in

Abstract. In research performances, affective computing has become a developing area because of its large use of application in interface of human computer. Recognition of emotion is one of the art techniques state in determining present human being psychological state. Assessment of the emotional state of humans has been traditionally learned using several direct psychological self-reports and psychological measures. There are various measures to recognize emotional states of human such as facial pictures, gestures, neuro-imaging methods and physiological signals. Therefore, some of these approaches need expensive and sizeable equipment which might hinder free motion. Emotions of human are very overlapping in nature and thus it requires an efficient featureclassifier and extractor assembly. It is a novel non-invasive technique to divide emotion of human through thermal face pictures. Invariants of Hu’s moment of different patches have been fused with statistical characteristic of histogram and used as features of robust in machine of multiclass support vector based division. Here 200 highly expressive thermal images are considered for training and 120 images for testing from IVITE database. The proposed system has overall accuracy of 87.50%. Keywords: Human emotions Support vector machine


Thermal images
Statistical features


1 Introduction Emotional intelligence relates to facial expressions. On the current scenario there is little literature works on solving the FER problem in thermal images. The major drawbacks in facial analysis are texture, disturbance caused by glasses and the areas that do not represent emotions. The advancement in exclusive and minutiae technology has lead to the exploration of abstract and qualitative object like emotions. Capture by thermal cameras is temperature distribution of branches of facial vein and is non-sensitive to conditions of lightning. Therefore, we concentrate on can remedy the disadvantages and problems of visible picture identification of emotion to a great extent. However, recent researches on emotions have seen the advent of emotion analysis from thermal images of human faces. Thermal images also known as thermo grams are basically display of heat distribution of an object in form of an image. Thermal images are captured by thermal infrared camera which captures the infrared radiation emitted © Springer Nature Switzerland AG 2020 A. Abraham et al. (Eds.): ISDA 2018, AISC 940, pp. 53–63, 2020. https://doi.org/10.1007/978-3-030-16657-1_6

54

C. M. Naveen Kumar and G. Shivakumar

from an object, converts it to electrical signals and finally displays on the monitor as gray scale or pseudo-colored images. Thermal Imaging is an established and ecological tool in psycho physiological activity monitoring and l in computational physiology [13]. It is known from the recent literature that the change in the emotional and psychological state of a person gets reflected in heart rate, rate of blood flow, temperature, contraction of pupil etc. Thus it can be inferred that temperature of skin pattern may be an indicator of the state in emotion of the person. Wang et al. [11] proposes a method using the temperature difference and K-Nearest Neighbour algorithm. An emotion classification using Deep Boltzmann Machine is proposed by Him. He et al. [2]. Khan et al. [3] in his work reports a method for emotion classification based on Facial Thermal Feature Points (FTFPs) by dividing the face into square grids and analysing the thermal variations across the grids. Emotion classification on the basis of analysis of time-frequency of facial skin temporal data using genetic algorithm is proposed by Nhan et al. [4]. The methodology of segmentation of picture by Yoshiaki Sugimoto has classified face into chin, cheek, eyes, mouse, nose, etc. to generate feature of every part then fit a straight line to analysing the variation of thermal field distribution of infrared picture. However, this method of analytical depends on the pictures of thermal with maximum degree of accuracy and high requirement of equipment of photographic [5]. Masood Mehmood Khan came up with a division method in domain of three –dimension. In this paper, it mixes taxonomy of LDA with measuring temperature of particular facial points to develop the model of 3D domain [6]. In addition, taxonomy of SVM with facial point temperature combined by the Leonardo Trujillo to divide emotion as well [7]. These two methodologies act well but with boundary of professional equipment to measure facial point temperature which is a drawback of processing in real-time system. Y. Yoshitomi mentioned that process the subtraction operation between expression images and neutral images firstly, segment to extract features then combines the neural network learning to classify human emotion [8]. Compared this method with earlier ones, it stops processing many useless data to maximize the efficiency of algorithms and rate of recognition (the rate of recognition for Surprise, Happy, Sad, Neutral expressions are 100%, 95%, 85% 80% separately). Nevertheless, when subtracting with neutral images, it inevitably undergoes alignment problem in image which error may introduce by this during the whole actual operation. In Benjamin theory of Hernandez’s, generating characteristics of cheeks, eyes, forehead, mouse to analysis with taxonomy of SVM is an effective way to decrease the amount of data which the rate of recognition is around 76.6% [9]. But, it’s hard to solve how to normalize the area with segmentation in real operation. In report of Earlier, analysis the variation in thermal pictures when facial movement units move in different speed and strength. In the experiment, 9 different facial movements are played by the 4 experts to measure the feasibility of the analysis of thermal of contraction of facial muscles [10]. However, obstacle of this technology is how to tract the motion unit.

A Thermal Imaging Based Classification of Affective States Using Multiclass SVM

55

2 IVITE Database Here IVITE database is used for training and testing of system because the dataset holds around 2300 images of different classes happy, sad, fear disgust and neutral. This database gives more accurate than other database. Total images in database are around 49 folders each folder consist of 2500 images with low resolution and exhibits emotion in the great variety. Soon as our data set is trained with data set .py in the testing phase from data set having emotions can be effortlessly classified the data set holds the grey scale images of dimension 160 * 120 pixels. The faces are resized in pre-processing stage, so that they occupied same space in each image by dimension 128 * 128 [12].

3 Proposed Methodology The proposed method is implemented using Python programming; the steps involved here are as follows. First the input image is given which is the grey scale thermal image then the face is detected by using Harr Cascade. Once it is detected the region of interest (ROI) is been extracted. Later the feature extraction is done using histogram and Hu’s moment invariant. These set of features are given to train the SVM model [14]. Later the test sample is given and the emotion is classified. The Fig. 2 shows the proposed method block diagram. 3.1

Acquisition of Thermal Image

In controlled room the acquisition of thermal images was done with uniform illumination, 24 C and relative humidity of 50%. From the thermal camera the person is been seated 1 m away and instructed to manage less head moment. The image is converted from thermal to gray scale image. The obtained image of grey scale is been used as input image as the first step in classifying the emotions. 3.2

Detection of Face

The detection of face of the input image is been done by using Haar cascade technique. Detection of object using classifiers of Haar cascade is an effective detection of object method proposed by Michael Jones and Paul Viola in their paper. It is an approach based on machine learning where the trained function of cascade with lot of negative images and positive images. Then it is used to identify the object in other pictures. It will also work for face detection. The Fig. 1 shows face detection of the input picture.

Fig. 1. Face detection

56

C. M. Naveen Kumar and G. Shivakumar

Fig. 2. Proposed method block diagram

To train the classifier firstly the algorithm requires lot of negative pictures (pictures without faces) and positive pictures (pictures of faces). Then we require to generate the characteristics from it. These are just like convolution kernels. This algorithm consist of 4 steps for detecting the faces, they are: Haar Features: These are the characteristics which are same to the convolution used to identify the presence of the features in the given picture. Every characteristics result in the signal value which is measured by subtracting the pixels sum under black rectangle to the pixels sum under white rectangle. To measuring characteristics in any picture Viola Jones algorithm uses a window of 24  24 as the size of base window. The Fig. 3 shows Haar features.

Fig. 3. Haar features

A Thermal Imaging Based Classification of Affective States Using Multiclass SVM

57

Integral Image: It is the value of the pixel (x, y) obtained by the addition of all the pixel value to the left of (x, y) and above it. The obtained picture is image of integral. The Fig. 4 shows the image of integral.

Fig. 4. Integral picture

Adaboost: Adaboost is an algorithm of machine learning which eliminate all the redundant features or the features which are not important and helps in narrowing down to several thousands of features. This constructs the strengthen divider as a combination in linear of these weakened dividers. The weakened divider is nothing but the good feature or the relevant feature which is extracted by adaboost. Cascading: Here cascade classifiers are used which are composed of stages and each stage is containing strengthened classifier. Therefore all the characteristics are grouped into various stages where every stage has particular number of characteristics. Here if the given number features for stage are not detected by the given input then the input will be discarded in that stage which it fails .One advantage is that rejection of images in where less time (Fig. 5).

Fig. 5. Cascading

3.3

ROI Extraction/Rejection

Once the face detection algorithm is been applied to detect the face and then we extract the required Region of Interest (ROI) from the images. Here we have taken left eye, right eye, nose and mouth as the ROI since we could find difference in this region for different emotion. Beside these there are other ROI’s also can be taken like forehead,

58

C. M. Naveen Kumar and G. Shivakumar

left cheek, right cheek and chin. But in the obtained data base we dint consider these ROI’s because there was no much difference seen in the images. The Fig. 6 shows the ROI extraction/rejection.

Fig. 6. ROI extraction/rejection

These ROI which are marked is been taken in this work. The selections of these regions are done by giving the co-ordinate values of those ROI which include x-axis, yaxis, height and width. Once the given value maps the proper region then we are going to crop it and separate it from the face. The Fig. 7 shows the cropped ROI.

Fig. 7. Cropped ROI

3.4

Feature Extraction

The process of conversion of the captured thermal picture into a distinct and unique form is called extraction of feature so that the reference template and feature extraction can be compared. Feature extraction is done for the ROI’s which have been selected. Here we use Histogram analysis and Hu’s moment invariant for extracting the features. Analysis of Histogram: In a context of an image processing, histogram of the intensity values of the pixel normally referred by the histogram image. The pixels number in a picture is showed by the histogram at every different intensity value appeared in that picture. Histogram can also be taken for the colour pictures. Histograms are used to model the probability distributions of the levels of the intensity that give us with important information about the picture features through the distribution profile of the intensity level. The first order probability of the histogram can be mathematically represented as, PðgÞ ¼ AðgÞ=N Where, A(g) denotes the pixels number at gray level g, N denotes the pixels number in the picture, P(g) denotes the probability of histogram. There are several methods to

A Thermal Imaging Based Classification of Affective States Using Multiclass SVM

59

calculate histogram of the image. Here histogram for the cropped ROI is been calculated by using mean and variance. Mean is defined as sum of all values divided by number of values. Variance is the measure of how far each value in the data set is from the mean [1]. Hu’s Moment Invariant: After segmentation picture moments are useful to explain the objects. Simple properties of pictures which are appear via picture moments include information about its orientation, area (or total intensity), it centroid. Here we have calculated the 7 values of invariant of Hu’s moment from Eqs. 1–7. These invariants can be constructed with respect to translation, scale and rotation. I1 ¼ g02 þ g20

ð1Þ

I2 ¼ ðg20  g02 Þ2 þ 4g211

ð2Þ

I3 ¼ ðg30  3g12 Þ2 þ ð3g21  g03 Þ2

ð3Þ

I4 ¼ ðg12 þ g30 Þ2 þ ðg03 þ g21 Þ2 h i I5 ¼ ðg30  3g12 Þðg12 þ g30 Þ ðg12 þ g30 Þ2 3ðg03 þ g21 Þ2 h i þ ð3g21  g03 Þðg03 þ g21 Þ 3ðg30 þ g12 Þ2 ðg03 þ g21 Þ2

ð4Þ

h i I6 ¼ ðg20  g02 Þ ðg12 þ g30 Þ2 ðg03 þ g21 Þ2 þ 4g11 ðg12 þ g30 Þðg03 þ g21 Þ h i I7 ¼ ð3g21  g03 Þðg12 þ g30 Þ ðg12 þ g30 Þ2 3ðg03 þ g21 Þ2 h i þ ðg30  3g12 Þðg03 þ g21 Þ 3ðg12 þ g30 Þ2 ðg03 þ g21 Þ2

ð5Þ

ð6Þ

ð7Þ

These are the well known as moment of Hu invariant. The intensities of the pixels are analogous to physical density. I1, is the first one which is analogous to the inertia moment around the picture’s centroid. I7, is the last one which is invariant of skew, which enables it to separation of similar pictures of otherwise mirror pictures. All these 7 values are been calculated for the selected ROI. Here we have taken only 5th and 7th value because there was a difference in values for different emotion when compared to other values of invariants. With respect to the condition of an environment the features of moment invariant are not unique due to temperature of our skin is having more dependency on the surrounding temperature of air. To generate the features of robust we have to consider any type of normalized characteristics of the each ROI. So as the result, we are also taking histogram feature along with Hu moment invariant.

60

3.5

C. M. Naveen Kumar and G. Shivakumar

Classification

By using SVM the classification was takes place. Fused characteristics computed by the generated characteristics of face which were provided to the SVM and benefited for emotions classification. SVM is a renowned binary classifier, identifies a hyper plane decision for division of two classes and a different plane is mapped by the characteristic vector by a non-linear mapping. A single multiclass problem is decreased into problems of multiple binary division for multi-class SVM approach [1]. Here we have first trained the SVM by the features obtained from the histogram and Hu moment invariant. We have given the label of class 0 as Disgust, class label 1 as Happy and class label 2 as sad for the SVM model. Hence based on this pre-trained set in SVM it makes the test sample to classify properly and finally the emotion is been classified.

4 Experimental Results The proposed system is implemented using python programming and three basic emotions i.e. disgust, happy and sad are recognized. Simulation Outputs: Here are the screenshots of histogram (left eye and right eye, nose, mouth) and overall how all the parts are cropped and lastly the command prompt. The Figs. 8 , 9, 10 show the simulation windows.

(a)

(b)

Fig. 8. Screenshot of histogram for (a) left eye (b) right eye

A Thermal Imaging Based Classification of Affective States Using Multiclass SVM

(a)

(b)

Fig. 9. Screenshot of histogram for (a) nose (b) mouth

(b)

(a)

(c) Fig. 10. (a), (b), (c): Detected output in GUI

61

62

C. M. Naveen Kumar and G. Shivakumar

The Table 1 represents the experimental results of the proposed method. Table 1. Emotion recognition accuracy Emotions Happy Disgust Sad

Accuracy with Histogram + SVM 91.45% 94.28% 84.50%

Accuracy with Histogram Moment Invariant + SVM 94.50% 95.28% 85.78%

5 Conclusions An efficient classification methodology and recognition of emotion from thermal pictures is done. Here the recognition of emotion is dependent on moments of Hu’s invariant computed from the patches of facial taken by the thermo gram or thermal picture fused with the statistics of histogram. These fused characteristics are used to build the characteristic vectors based on which the emotions are classified using the SVM in multi-class. All the algorithms are checked on the IVITE database. Here optimized human emotion recognition system has been developed by considering four regions of interest and only two Hu’s moments which is highly correlated with the emotions. Three basic emotions i.e. disgust, happy and sad are recognized. The accuracy obtained while considering only histogram was 81.95% and the overall accuracy while considering both histogram and Hu’s moment invariant is 87.5%. The accuracy can be increased by increasing the number of training images in SVM. This proposed model can be done by using convolution neural networks. Hu moments cannot be used for the intention of face identification using complex facial pictures. In order to overcome this problem Zernike and Legendre moments can be used.

References 1. Basu, A., et al.: Human emotion recognition from facial thermal image based on fused statistical feature and multi-class SVM. In: 2015 Annual IEEE India Conference (INDICON). IEEE (2015) 2. He, S., et al.: Facial expression recognition using deep Boltzmann machine from thermal infrared images. In: Humaine Association Conference on Affective Computing and Intelligent Interaction, pp. 239–244 (2013) 3. Khan, M.M., Ingleby, M., Ward, R.D.: Automated facial expression classification and affect interpretation using infrared measurement of facial skin temperature variations. ACM Trans. Auton. Adapt. Syst. 1(1), 91–113 (2006) 4. Nhan, B.R., Chau, T.: Classifying affective states using thermal infrared imaging of the human face. IEEE Trans. Biomed. Eng. 57(4), 979–987 (2010) 5. Yoshitomi, Y., Sugimoto, Y., Tomita, S.: A method for detecting transitions of emotional stated using thermal facial image based on a synthesis of facial expressions. Robot. Auton. Syst. 31(3), 147–160 (2000)

A Thermal Imaging Based Classification of Affective States Using Multiclass SVM

63

6. Khan, M.M., et al.: Automated classification and recognition of facial expressions using infrared thermal imaging. In: IEEE Conference on Cybernetics and Intelligent Systems, vol. 1. IEEE (2004) 7. Trujillo, L., et al.: Automatic feature localization in thermal images for facial expression recognition. In: IEEE Computer Society Conference on Computer Vision and Pattern Recognition-Workshops, CVPR Workshops. IEEE (2005) 8. Miyawaki, N., Yoshitomi, Y., Tomita, S., Kimura, S.: Facial Expression Recognition Using Thermal Image Processing and Neural Network (1997) 9. Hernández, B., et al.: Visual learning of texture descriptors for facial expression recognition in thermal imagery. Comput. Vis. Image Underst. 106, 258 (2007) 10. Jarlier, S., et al.: Thermal analysis of facial muscles contractions. IEEE Trans. 2(1), 2–9 (2011) 11. Wang, S., Shen, P., Liu, Z.: Facial expression recognition from infrared thermal images using temperature difference by voting. In: IEEE 2nd International Conference on Cloud Computing and Intelligent Systems, pp. 94–98 (2012) 12. Esposito, A., et al.: A naturalistic database of thermal emotional facial expressions and effects of induced emotions on memory. In: Cognitive Behavioral Systems, pp. 158–173. Springer, Heidelberg (2012) 13. Cardone, D., et al.: New frontiers for applications of thermal infrared imaging devices: computational psychophysiology in the neurosciences. Sensors 17(5), 1042 (2017) 14. Grosan, C., Abraham, A.: Intelligent Systems: A Modern Approach. Intelligent Systems Reference Library. Springer, Heidelberg (2011)

Multidimensional Crime Dataset Analysis Prerna Kapoor and Prem Kumar Singh(&) Amity Institute of Information Technology, Amity University, 125, Noida, Uttar Pradesh, India [email protected], [email protected]

Abstract. Data analytics (DA) is defined as the process of scrutinizing different data sets to draw out the outcomes about the information they contain with the help of specialized functional systems and software. There are different areas where data analytics applications have been operated such as transportation, detection of fraud, city planning, health department, digital advertisement, etc. One of the key area of data analytics is in the crime world. Crime Analysis and prevention is a systematic approach for identifying and analyzing patterns and trends in crime. Our proposed method can predict regions which have high probability for a particular crime occurrence from previous years records and the necessary actions that can be taken place by the police authorities to provide more and more security. The necessary steps can be initiated for security reasons so that criminals think twice before performing a crime. Instead of focusing on causes of crime occurrence like criminal background of offender, etc., we are focusing on crime patterns in different regions. Crime Analysis is concerned with exploring different crime datasets, analyzing them and finding out certain patterns from them, so data analytics is a field which helps in establishing certain patterns from the data. In this paper, we are going to represent the crime data in the form of multipolarity to find relationships between the objects and the attributes. Since, the crime data is very large in size and in unstructured manner, so there is a need to first normalize the data and then find relationships among them by representing them in the form of m-Polar Fuzzy Contexts and m-Polar Fuzzy Concepts. Keywords: Knowledge Discovery
Formal Concept Analysis (FCA) m-Polar fuzzy set
m-Polar fuzzy graph




1 Introduction Knowledge Discovery in Databases (KDD) and Data Mining (DM) is a research area where different approaches are refined for fetching knowledge from data [1, 18–20]. When awareness is less about the data or in other words, we can say that lack of knowledge is there about the data and the scrutiny goals are vague, then visual data exploration and visual analytics are needful. So, for visual data exploration, Formal Concept Analysis (FCA) was introduced by Rudolf Wille in the early 1980s. It is defined as a method for formulation conceptual anatomy out of data. These anatomies can be represented graphically as conceptual hierarchies which help in analyzing complex structures and the steadiness within the data. Pattern structures is another © Springer Nature Switzerland AG 2020 A. Abraham et al. (Eds.): ISDA 2018, AISC 940, pp. 64–72, 2020. https://doi.org/10.1007/978-3-030-16657-1_7

Multidimensional Crime Dataset Analysis

65

elongation of FCA that grants processing complex data such as datasets or graphs [2]. Multi-polarity plays an important role in different fields of technology. Chen et al. [3] introduced the theory of m-polar fuzzy sets. In m-Polar fuzzy set, membership value of an element belongs to [0, 1]m which epitomizes m different properties of an element [4]. The crime data that we gather is from different sources such as Bureaus, Police Departments, etc., such datasets contain different attributes (k) where k can be multidecision attributes or multi-polar information. So in such cases, where k  2, multipolar information emerges and they cannot be simply represented by unipolar or bipolar graphs. So, here the need of m-Polar fuzzy sets arises to describe the relationship of data. In this paper, we are going to apply m-Polar fuzzy concept in crime datasets to find out different crime patterns. Crime analysis is concerned with exploring different crime datasets, then analyzing them and finding out patterns and relationships from such data sets. Since, there is a high volume of crime data, so data analytics is a field which helps in establishing certain patterns from the data. By constructing a m-polar concept lattice, we can visualize the collective information and knowledge it provides so that detailed investigation can be carried on for particular criminals or particular spot areas as per the requirement of the investigator. Once we represent the data in the form of lattice, a pattern can be analyzed and it becomes easy for the investigator to pick up key patterns as depicted by the lattice and relate it with the criminal case rather than going through all the reports one by one. Kester [5] introduced a method for categorizing attributes of crime using FCA. Qazi et al. [6] introduced an associative search model and used FCA to find crime hotspots and profile of offenders. Andrews et al. [7] proposed an FCAbased approach for finding new patterns from existing data. Sarwar and Akram [4] introduced the concepts m-Polar fuzzy context, m-Polar fuzzy concept lattice and discussed certain applications of m-Polar fuzzy concept lattice or detecting women and child trafficking suspects. Poelmans et al. [8] used FCA to classify certain cases as domestic violence or non-domestic violence. Singh [9] recently introduced a method to identify patterns when acceptance is at par for the m-Polar fuzzy attributes. The organization of the paper is as follows: The Sect. 1 replenishes description about KDD and Data Mining, Formal Concept Analysis (FCA), Multi-polarity, need of m-Polar and how to apply m-Polar in crime pattern analysis. Section 2 provides a background of different concepts such as fuzzy, fuzzy logic, prolongation of unipolar to bipolar. Section 3 enlists certain definitions related to FCA. Section 4 provides the illustration of our proposed work followed by conclusion and future work in Sect. 5.

2 Background The word “fuzzy” means “vagueness”. Fuzziness transpires when the boundary of an information is not clearly defined. Fuzzy logic was introduced by the American Computer Scientist Zadeh [10].

66

P. Kapoor and P. K. Singh

Fuzzy Logic is defined as a method of decision making which involves all possibilities between YES and No. It is a form of logic which is based on the fuzzy set concept. The association in fuzzy sets is described by the values which are in the form of [0, 1]. The fuzzy sets impersonate vagueness which is due to human perception. Fuzzy set theory is an elongation of the Classical set theory where the elements have degree of membership. Fuzzy logic has been solicited as one of the best techniques in relation with human reasoning and decision making [11]. FCA is progressively applied in knowledge discovery, data analysis, retrieval of information, etc. Formal Concept Analysis (FCA) initiates from a binary table, also known as a cross-table which epitomizes relationship between objects and their attributes [12]. The objects are delineated by table rows and attributes by table columns. The Table 1 shows the objects and the attributes where x manifests that the analogous object has the analogous attribute.

Table 1. Table with logical attributes I a1 b1 b2 b3 x b4 x

a2 a3 a4 x x x x x

Our primary concern is use FCA for knowledge discovery in databases. Knowledge Discovery in Databases (KDD) is defined as the process of perceiving useful knowledge from a set of data. It is the most widely used technique of data mining. The process of KDD is defined in Fig. 1 as follows.

Fig. 1. KDD process

Multidimensional Crime Dataset Analysis

67

As we know that data analytics is an umbrella term covering different aspects such as data mining, FCA, etc. So we are focusing on pattern analysis through FCA but before that we need to get overview of about fundamentals of FCA and different definitions of the concepts. FCA is prolonged from unipolar to bipolar and three-polar fuzzy space. It has been perceived that many data sets embrace multi-polar or multidecision attributes.

3 Preliminaries Definition 3.1 (Ref. [4]) A fuzzy subset of a universe U (a fuzzy set) is a mathematical object M described by its characteristic function (membership function) µA: U ! [0, 1]. The classical membership degree are represented by 1 (is a member) and 0 (not a member). Definition 3.2 (Ref. [8]) An m-polar fuzzy set (or a [0, 1]m-set) on U is exactly a mapping A: U ! [0, 1]m. The set of all m-polar fuzzy sets on U is denoted by m(U). Definition 3.3 (Ref. [9]) e m-polar (or a [0, 1]m) fuzzy relation is mapping: l R(xi, yj) ! [0, 1]m where xi 2 U, yj 2 V. It means for each pair of elements from U and V is assigned an m-tuple of lattice e on the set U and V can values, i.e. an element of [0, 1]. Thus, m-polar fuzzy relation ( R) e e e e be represented as follows: R(xi, yj) = {(xi, yj), l R(xi, yj)} where l R(xi, yj) = (l 1 R e (xi, yi), …, lm R(xi, yi)) and the relation is mapped in the interval [0, 1]m. Definition 3.4 (Ref. [10]) An m-polar fuzzy formal context is defined as a 3-tuple Ft = (Y, Z, E) where Y = {y1, y2, … yn} is the collection of objects, Z = {z1, z2, … zr} is the collection of attributes and E is an mF set on Y  Z i.e. E: Y  Z -> [0, 1]m. Definition 3.5 (Ref. [7]) Denoted by B(X, Y, I) the collection of all formal concept of (X, Y, I) i.e. B(X, Y, I) = {(A, B) 2 2X  2Y |A" = B, B# = A}. B(X, Y, I) equipped with the sub-concept and super-concept ordering 14000∨attribute_Value