Next Generation Computing Technologies on Computational Intelligence: 4th International Conference, NGCT 2018, Dehradun, India, November 21–22, 2018, Revised Selected Papers [1st ed. 2019] 978-981-15-1717-4, 978-981-15-1718-1

Table of contents :
Front Matter ....Pages i-xix
Front Matter ....Pages 1-1
Texture Redefined: A Second Order Statistical Based Approach for Brodatz Dataset Samples 1–35 (A) (Amit Kumar Shakya, Shalini Tiwari, Anurag Vidyarthi, Rishi Prakash)....Pages 3-12
A Taxonomy on Machine Learning Based Techniques to Identify the Heart Disease (Anand Kumar Srivastava, Pradeep Kumar Singh, Yugal Kumar)....Pages 13-25
Prediction and Analysis of Ecological Climatic Conditions (Humidity/Temperature) of Different Countries Fly-Ash Using Data Mining Techniques (Divya Gupta, Kanak Saxena)....Pages 26-37
NStackSenti: Evaluation of a Multi-level Approach for Detecting the Sentiment of Users (Md Fahimuzzman Sohan, Sheikh Shah Mohammad Motiur Rahman, Md Tahsir Ahmed Munna, Shaikh Muhammad Allayear, Md. Habibur Rahman, Md. Mushfiqur Rahman)....Pages 38-48
A Large Scale Study for Identification of Sarcasm in Textual Data (Pulkit Mehndiratta, Devpriya Soni)....Pages 49-63
Security Issues and Solutions in Cloud Robotics: A Survey (Saurabh Jain, Rajesh Doriya)....Pages 64-76
Emotion Recognition in Poetry Using Ensemble of Classifiers (P. S. Sreeja, G. S. Mahalakshmi)....Pages 77-91
An Approach to Threshold Based Human Skin Color Recognition and Segmentation in Different Color Models (B. S. Sathish, P. Ganesan, S. Dola Sanjay, A. Ranganayakulu, S. Jagan Mohan Rao)....Pages 92-101
Neuro-Fuzzy Approach for Reconstruction of 3-D Spine Model Using 2-D Spine Images and Human Anatomy (Saurabh Agrawal, Dilip Singh Sisodia, Naresh Kumar Nagwani)....Pages 102-115
Multi-objective Planning of Rural Distribution System Using PSO (Meena Kumari, V. R. Singh, Rakesh Ranjan)....Pages 116-127
A Review of Reverse Dictionary: Finding Words from Concept Description (Bushra Siddique, Mirza Mohd Sufyan Beg)....Pages 128-139
A Graph Based Analysis of User Mobility for a Smart City Project (Jai Prakash Verma, Sapan H. Mankad, Sanjay Garg)....Pages 140-151
Parkinson’s Disease Diagnosis by fMRI Images Using MFCC Feature Extraction (Ritika Dubey, Aditi Narang, Vijay Khare)....Pages 152-160
Front Matter ....Pages 161-161
A New Approach to Train LSTMs, GRUs, and Other Similar Networks for Data Generation (Ravin Kumar)....Pages 163-171
Automatic Road Extraction from Semi Urban Remote Sensing Images (Pramod Kumar Soni, Navin Rajpal, Rajesh Mehta)....Pages 172-182
A New Hybrid Backstepping Approach for the Position/Force Control of Mobile Manipulators (Manju Rani, Dinanath, Naveen Kumar)....Pages 183-198
Distributed Multi-criteria Based Clusterhead Selection Approach for MANET (Preeti Yadav, Ritu Prasad, Praneet Saurabh, Bhupendra Verma)....Pages 199-209
Enhanced Slicing+: A New Privacy Preserving Micro-data Publishing Technique (Rajshree Srivastava, Kritika Rani)....Pages 210-220
Discovering Cyclic and Partial Cyclic Patterns Using the FP Growth Method Incorporated with Special Constraints (Pragati Upadhyay, Narendra Kohli, M. K. Pandey)....Pages 221-235
Front Matter ....Pages 237-237
iDJ: Intrusion Detection System in Java (Radhesh Khanna, Shashank Sharma)....Pages 239-247
An Effective Optimisation Algorithm for Sensor Deployment Problem in Wireless Sensor Network (Vishal Puri, A. Ramesh Babu, T. Sudalai Muthu, Sonali Potdar)....Pages 248-258
A Review of Sensor Deployment Problem in Wireless Sensor Network (Vishal Puri, A. Ramesh Babu, T. Sudalai Muthu, Kirti Korabu)....Pages 259-268
Analysis of Energy Efficient Framework for Static and Mobile Nodes in WSN-Assisted IoT (Anurag Shukla, Sarsij Tripathi)....Pages 269-280
Energy Aware SEP Based on Dual Cluster Heads and Dual Sink Approach in WSN (Sanjay Kumar, Narendra Harariya, Arka Bhowmik)....Pages 281-295
Impact of Cranial Electrical Stimulation Based Analysis of Heart Rate Variability in Insomnia ( Khyatee, Aparna Sarkar, Rajeev Aggarwal)....Pages 296-307
Front Matter ....Pages 309-309
EAAP: Efficient Authentication Agreement Protocol Policy for Cloud Environment (Narander Kumar, Jitendra Kumar Samriya)....Pages 311-320
Multi-path Priority Based Route Discovery Mechanism (Swati Atri, Sanjay Tyagi)....Pages 321-329
Secure and Trustworthy Cloud: Need of Digital India, an e-Governance Project (Archana B. Saxena, Meenu Dave)....Pages 330-339
Hyperheuristic Framework with Evolutionary and Deterministic Algorithms for Virtual Machine Placement Problem (Amol C. Adamuthe, Akshayya Jadhav)....Pages 340-350
Efficient Directional Information Dissemination Approach in Vehicular Ad Hoc Networks (Sandeep Kad, Vijay Kumar Banga)....Pages 351-363
Software Agents in Support of Scheduling Group Training (Giorgi Mamatsashvili, Konrad Gancarz, Weronika Łajewska, Maria Ganzha, Marcin Paprzycki)....Pages 364-375
Back Matter ....Pages 377-378

Citation preview

Manish Prateek · Durgansh Sharma · Rajeev Tiwari · Rashmi Sharma · Kamal Kumar · Neeraj Kumar (Eds.)

Communications in Computer and Information Science

922

Next Generation Computing Technologies on Computational Intelligence 4th International Conference, NGCT 2018 Dehradun, India, November 21–22, 2018 Revised Selected Papers

Communications in Computer and Information Science Commenced Publication in 2007 Founding and Former Series Editors: Phoebe Chen, Alfredo Cuzzocrea, Xiaoyong Du, Orhun Kara, Ting Liu, Krishna M. Sivalingam, Dominik Ślęzak, Takashi Washio, Xiaokang Yang, and Junsong Yuan

Editorial Board Members Simone Diniz Junqueira Barbosa Pontifical Catholic University of Rio de Janeiro (PUC-Rio), Rio de Janeiro, Brazil Joaquim Filipe Polytechnic Institute of Setúbal, Setúbal, Portugal Ashish Ghosh Indian Statistical Institute, Kolkata, India Igor Kotenko St. Petersburg Institute for Informatics and Automation of the Russian Academy of Sciences, St. Petersburg, Russia Lizhu Zhou Tsinghua University, Beijing, China

922

More information about this series at http://www.springer.com/series/7899

Manish Prateek Durgansh Sharma Rajeev Tiwari Rashmi Sharma Kamal Kumar Neeraj Kumar (Eds.) •

•

•

•

•

Next Generation Computing Technologies on Computational Intelligence 4th International Conference, NGCT 2018 Dehradun, India, November 21–22, 2018 Revised Selected Papers

123

Editors Manish Prateek University of Petroleum and Energy Studies Dehradun, India

Durgansh Sharma University of Petroleum and Energy Studies Dehradun, India

Rajeev Tiwari University of Petroleum and Energy Studies Dehradun, India

Rashmi Sharma University of Petroleum and Energy Studies Dehradun, India

Kamal Kumar National Institute of Technology Srinagar, Uttrakhand, India

Neeraj Kumar Thapar University Patiala, India

ISSN 1865-0929 ISSN 1865-0937 (electronic) Communications in Computer and Information Science ISBN 978-981-15-1717-4 ISBN 978-981-15-1718-1 (eBook) https://doi.org/10.1007/978-981-15-1718-1 © Springer Nature Singapore Pte Ltd. 2019 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 Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Preface

These proceedings comprise the best research papers presented at the 4th International Conference on Next-Generation Computing Technologies (NGCT 2018) organized by the School of Computer Science at the University of Petroleum and Energy Studies (UPES), Dehradun, India, during November 21–22, 2018. The conference theme was “Computational Intelligence for Next-Generation Computing” and all the tracks and sub-tracks focused on contemporary research in computing and information technology. NGCT 2018 offered a platform to researchers, experts, academics, and industry fellows to share and discuss their research findings with the aim of offering human beings a better daily living with the help of next-generation computing technologies. NGCT 2018 followed a strict peer-review process, with 31 top-quality papers being selected and presented out of 319 submissions from various parts of the globe. The present proceedings contain four parts, namely, “Image Processing, Pattern Analysis, and Machine Learning,” “Information and Data Convergence,” “Disruptive Technologies for Future,” and “E-Governance and Smart World.” We express our thanks to the university leaders, advisory and technical board members, keynote speakers, and Organizing Committee members. We extend our thanks to the conference co-sponsors UCOST, CSIR, DRDO, Elsevier, CETPA, Sumeru Infrastructures, and The Solitaire as well as the publication partners Springer and Inderscience publishers. October 2019

Manish Prateek Durgansh Sharma Rajeev Tiwari Rashmi Sharma Kamal Kumar Neeraj Kumar

Organization

Steering Committee Chief Patron S. J. Chopra

UPES, Dehradun, India

Patron Deependra Kumar Jha

UPES, Dehradun, India

Co-patron Kamal Bansal

UPES, Dehradun, India

General Chair Manish Prateek

UPES, Dehradun, India

Conference Chair Durgansh Sharma

UPES, Dehradun, India

Conference Secretary Rajeev Tiwari Ajay Prasad

UPES, Dehradun, India UPES, Dehradun, India

Advisory Committee Sartaj Sahni William Stallings Nitin Henry Hexmoor Vijay K. Vaishnavi Dennis K. Peters Laurent George Rafik Zitouni Jaouhar Fattahi Karan Verma Zarinah Mohd Kasirun Su Cheng Haw Shamimul Qamar Benoit Favre Sugam Sharma Ding-zhu Du

University of Florida, USA Independent Consultant, USA University of Cincinnati, USA Southern Illinois University, USA Georgia State University, USA Memorial University, Canada University of Paris-East, France ECE Paris, France Valcartier Research Centre, Canada Universiti Teknologi PETRONAS, Malaysia University of Malaya, Malaysia Multimedia University, Malaysia King Khalid University, Saudi Arabia Université d’Aix-Marseille, France Iowa State University, USA University of Texas, USA

viii

Organization

Bharat K. Bhargava Valentina Emilia Balas Deepak Kumar Gupta Nabil Belgasmi B. V. Babu Mohamed Belaoued Jindong Liu Brian Lee Piyush Maheshwari K. Mustafa M. N. Hoda Niladri Chatterjee K. Chandrasekaran M. L. Saikumar K. K. Shukla Umapada Pal Pabitra Pal Choudhury Anirban Basu Preeti Bajaj Prosenjit Pal R. Rajendra Prasath Kunwar Singh Vaisla Sanjay Sood Bhabatosh Chanda Gopi Krishna Durbhaka Lokanatha C. Reddy H. L. Mandoria Subodh Srivastava Muthukumar Subramanyam Parma Nand Arindam Wiswas Vinay Rishiwal Sudhir Kumar Sharma Chandrasekaran Subramaniam Anil Kumar Verma R. B. Patel Shrihari Honwad Ashok Kumar Gwal Annappa B. N. Muthu Kumaran G. R. Sinha

Purdue University, USA Aurel Vlaicu, University of Arad, Romania Fayetteville State University, USA Banque de Tunisie, Tunisia Graphic Era University, India University of Constantine, Algeria Imperial College London, UK Athlone Institute of Technology, Ireland Amity University, Dubai Jamia Millia Islamia University, India BVICAM, India Indian Institute of Technology, Delhi, India NIT, Suratkal, India Institute of Public Enterprise, India IIT, Varanasi, India Indian Statistical Institute, India Indian Statistical Institute, India CSI, India Council, India IEEE, India Council, India Prophecy Sensorlytics, LLC, India Indian Institute of Information Technology, India Uttarakhand Technical University, India C-DAC, Mohali, India Indian Statistical Institute, Kolkata, India Sathyabhama University, India Dravidian University, India G. B. Pant University of Agriculture and Technology, India JNTU, India IIIT, Trichy, India Sharda University, India Asansol Engineering College, India Rohilkhand University, India IITM-Institute of Information Technology & Management, India SRITE, India Thapar University, India Chandigarh Engineering College, India GD Goenka University, India Governing Council at Association of Indian Universities, India NIT, Karnataka, India Anna University, India International Institute of Information Technology, India

Organization

Murtaza Rizvi Ajith Abraham Deepak Garg K. V. Arya Md. Abulaish M. P. Singh D. P. Vidyarthi R. K. Agrawal Chiranjeev Kumar Manu Sood

National Institute of Technical Teachers’ Training and Research, India Machine Intelligence Research Labs, India Bennet University, India IET, India SAARC University, India B. R. Ambedkar University, India JNU, India JNU, India ISM, Dhanbad, India H.P. University, India

Keynote Speakers Marcin Paprzycki E. G. Rajan Deepak Bhaskar Phatak Anurag Mishra

Systems Research Institute, Polish Academy of Sciences, Poland Founder President of the Pentagram Research Centre (P) Ltd., Hyderabad, India Former Professor at IIT, Bombay, India Deendayal Upadhyay College, Delhi University

Technical Program Committee Technical Program Chairs Rajeev Tiwari Neeraj Kumar Rashmi Sharma

UPES, Dehradun, India Thapar University, Patiala, India UPES, Dehradun, India

Editors Manish Prateek Durgansh Sharma Rajeev Tiwari Rashmi Sharma Kamal Kumar Neeraj Kumar

UPES, Dehradun, India UPES, Dehradun, India UPES, Dehradun, India UPES, Dehradun, India NIT, Uttrakhand, India Thapar University, Patiala, India

Members Natarajan Meghanathan Noor Zaman Noor Elaiza Binti Abdul Khalid Visvasuresh Victor Govindaswamy Nagarajan Kathiresan Sanjay Misra

Jackson State University, USA King Faisal University, Saudi Arabia University Technology Mara Shah Alam, Malaysia Concordia University, Canada SIDRA Medical and Research Center, Qatar Federal University of Technology, Nigeria

ix

x

Organization

Alireza Haghpeima Mithun Mukherjee Amit Banerjee Md. Ezaz Ahmed K. L. Rahmani Nguyen Cuong Ahmed Ali Saihood Arun Dev Dhar Dwivedi Kusum Yadav Xiao-Zhi Gao Mudassir Khan Upasana Geetanjali Singh Gorachand Dutta Shailendra Misra Mayank Singh Abhijit Das Mohd. Shoab Angelina Geetha Yogita Gigras Raman Chadha S. Arvind Hemraj Saini V. Singh Vr Rajib Kar Jasbir Saini Abdus Samad Pallav Kumar Baruah Rajiv Pandey Diptendu Sinha Roy Chittaranjan Pradhan Bimal Roy Sangram Ray Gopalakrishnan T. Harish Mittal J. K. Deegwal Saurabh Mukherjee Arvind Rehalia Arun Sharma Satish Kumar Singh Shrivishal Tripathi Tanvir Ahmad

Islamic Azad University of Mashhad, Iran Guangdong University of Petrochemical Technology, China Innovative Photonics Evolution Research Center, Japan Saudi Electronic University, Saudi Arabia Saudi Electronic University, Saudi Arabia Quang Nam University, Vietnam Thiqar University, Iraq University of Bordeaux, France University of Hail, Saudi Arabia Aalto University, Finland King Khalid University, Saudi Arabia University of Kwa-Zulu Natal (Westville Campus), South Africa University of Bath, UK University of Kwa-Zulu Natal (Westville Campus), South Africa University of Kwa-Zulu Natal (Westville Campus), South Africa Inria Sophia Antipolis - Méditerranée, France Shaqra University, Saudi Arabia B.S. Abdur Rahman University, India The Northcap University, India Chandigarh Group of Colleges, India CMR Institute of Technology, India Jaypee University of Information Technology, India National Physical Laboratory, India National Institute of Technology, India DCR University of Science and Technology, India Aligarh Muslim University, India Sri Sathya Sai Institute of Higher Learning, India Amity University, India National Institute Technology, India KIIT University, India Indian Statistical Institute, India National Institute of Technology, India Bannari Amman Institute of Technology, India B.M. Institute of Engineering and Technology, India Government Engineering College, India Banasthali Vidyapith, India Bharti Vidyapeeth, India Indira Gandhi Delhi Technical University for Women, India Indian Institute of Information Technology, India Indian Institute of Technology Jodhpur, India Jamia Millia Islamia University, India

Organization

Suraiya Jabin V. K. Jain Sudeep Tanwar Dhirendra Mishra Anil Kumar Yadav Niyati Baliyan Shelly Sachdeva Ranjeet Kumar Sanjeev Sharma Sherin Zafar K. Saravanan M. Sandhya S. Geetha Ranjit Rajak Ashutosh Tripathi Rajat Saxena Lalit B. Damahe Umesh L. Kulkarni Nihar Ranjan Roy Rashmi Thakur Xavier Arputha Rathina Emmanuel Shubhakar Pilli Manuj Aggarwal Tarun Goyal Dheeraj Malhotra Gaurav Agrawal Vaibhav Muddebihalkar T. R. V. Anandharajan K. M. Mehata Neha Gulati Akhilesh Kumar Sharma Rajeev Kumar Baldev Singh Vishnu Srivastava Rakesh Kumar Bansal H. N. Suresh Y. J. Nagendra Kumar A. K. Verma Daya Gupta Jyotirmay Patel S. Ghosh Sajai Vir Singh Sandeep Paul

xi

Jamia Millia Islamia University, India Mody University of Science and Technology, India Nirma University, India Narsee Monjee Institute of Management Studies, India UIET-CSJM University, India Indira Gandhi Delhi Technical University for Women, India NIT, Delhi, India Indian Institute of Information Technology, India Rajiv Gandhi Technological University, India Jamia Hamdard University, India Anna University Regional Campus, India B.S. Abdur Rahman University, India VIT University, India Dr. Harisingh Gour University, India Amity University, India Indian Institute of Technology, India Yeshwantrao Chavan College of Engineering, India VIT Wadala, India GD Goenka University, India Thakur college of Engineering and Technology, India B.S. Abdur Rahman University, India MNIT, India ARSD College, India AIETM, India Guru Gobind Singh Indraprastha University, India Inderprastha Engineering College, India Savitribai Phule University of Pune, India Einstein College of Engineering, India Hindustan University, India University Business School, India Manipal University, India Teerthanker Mahaveer University, India VIT, India CSIR CEERI, India Maharaja Ranjit Singh Punjab Technical University, India Bangalore Institute of Technology, India Gokaraju Rangaraju Institute of Engineering and Technology, India Thapar University, India Delhi Technological University, India MIET, India Galgotias University, India Jaypee Institute of Information Technology, India Dayalbagh Educational Institute, India

xii

Organization

Sunil Sikka Kuldeep Kumar Sunita Varma Lavika Goel Sunanda Dixit Vijander Singh Pratiyush Guleria Saiyedul Islam Purnima Sharma Manoj Kumar Mohd Imran Syed Aamiruddin Sudhanshu Kumar Jha S. Balan Mohd Vasim Ahamad Sonal Purohit Om Prakash Sharma Devendra Kumar Sharma Vandana Niranjan Surya Prakash B. Narendra Kumar Rao Rohit Beniwal Amit Patel Puneet Mishra Atul Garg Savina Bansal Vinay Nassa Manu Sood Neeta Singh Damodar Reddy Sujata Pandey Seema Maitrey Mahinder Singh Aswal A. Srinivasan Gollapudi Ramesh Chandra Himanshu Monga Sudipta Roy Dilip Singh Sisodia Ravi Mishra Rakesh Chandra Arif Mohammad Abhineet Anand Rakesh Kumar Devendra Prasad

Amity University, India Birla Institute of Technology and Science, India SGSITS, India Birla Institute of Technology & Science, India DSCE, India Amity University, India NIELIT, DOEACC Society, India BITS, Pillani, India Mody University of Science and Technology, India Shri Mata Vaishno Devi University, India Aligarh Muslim University, India Guru Govind Singh Indraprastha University, India National Institute of Technology Jamshedpur, India Government Arts College, India University Women’s Polytechnic, India FMS-WISDOM Banasthali University, India Poornima College of Engineering, India Meerut Institute of Engineering and Technology, India IGDTUW, India Indian Institute of Technology, India Sree Vidyanikethan Engineering College, India Delhi Technological University, India RGUKT IIIT, India University of Lucknow, India Maharishi Markandeshwar University, India Punjab Technical University, India South Point Technical Campus, India Himachal Pradesh University, India Gautam Buddha University, India National Institute of Technology, India Amity University, India Krishna Institute of Engineering and Technology, India Gurukul Kangri University, India MNM Jain Engineering College, India VNR Vignana Jyothi Institute of Engineering and Technology, India Shiva Group of Institutions, India Assam University, India National Institute of Technology, India Indian Institute of Technology, BHU, India IIIT, Bhubaneshwar, India Integral University, India Galgotia University, India National Institute of Technical Teachers Training and Research, India Chandigarh Group of Colleges, India

Organization

Harmunish Taneja Amit Doegar Mala Kalra Mukesh Kumar Jyoti Thaman Sanjeev Rana Amit Sabharwal Leena Arya Himanshu Aggarwal J. Satheesh Kumar Kamal Kant Rama Krishna Challa Charu Paawan Sharma Rashid Ali S. R. Balasundaram Saru Dhir B. Surendiran Sapna Tyagi Anurag Singh Avdhesh Kumar Gupta Paresh Virparia Sandeep Vijay Udai Pratap Rao B. Balamurugan Rabindra Jena S. N. Pandey Ved Prakash Susheela Dahiya Partha Pakray Pradeep Tomar Mansaf Alam Deepika Koundal Mukesh Kumar Atul M. Gonsai A. Malathi Manik Sharma N. N. Das S. K. Singh Prashant Ahlawat Dheeraj Jain Nishant Malik Kumud Ranjan Jha

xiii

DAV College, India National Institute of Technical Teachers Training and Research, India National Institute of Technical Teachers Training and Research, India YMCA, India Independent Consultant, India MM University, India IIT, Kharagpur, India ITS Engineering College, India Punjabi University, India Bharathiar University, India NIT, Hamirpur, India NITTR, India Jaypee Institute of Information Technology, India University of Petroleum and Energy Studies, India AMU, India NIT, Trichy, India Amity University, India NIT, Puducherry, India Institute of Management Studies, India NIT, Delhi, India Institute of Management Studies, India Sardar Patel University, India DIT University, India Sardar Vallabh Bhai National Institute of Technology, India VIT University, India Institute of Management Technology, India MNNIT, India University of Petroleum and Energy Studies, India University of Petroleum and Energy Studies, India NIT, Mizoram, India GBU, India Jamia Millia Islamia Central University, India Chitkara University, India Swami Keshvanand Institute of Technology, India Saurashtra University, India Government Arts College, India DAV University, India Manav Rachna University, India Sanskriti University, India GITM, India Sangam University, India CERT, India SMVDU, India

xiv

Organization

Rakesh Kumar Jha Ajay Kaul Soumen Bag Gesu Thakur Dinesh Sharma Saptarshi Roy Chowdhury Anwesa Das Ritika Sudhansh Sharma Ashish Bagwari Sanjay Kumar Alaknanda Ashok Vishal Kumar Vinay Rishiwal Pramod Kumar A. Senthil Ajay Kr. Singh S. R. Biradar Anand Sharma S. K. Vasistha Rabins Porwal Prateek Srivastava Manoj Diwakar Manu Pratap Singh P. K. Mishra Sunita Agarwal B. K. Verma Rajneesh Kumar Shashi Bhushan Nisheeth Joshi Susheel Chandra Dimri Iti Mathur Nipur Bali Ritika Arora Mehra Rishi Asthana Mahendra Singh Aswal Madhu Gaur Anil Prabhakar Manuj Darbari Rajdev Tiwari Somesh Kumar Ashish Sharma Diwakar Bhardwaj

SMVDU, India SMVDU, India IIT (ISM) Dhanbad, India IMS, India Amity, India NSHM Knowledge Campus, India NSHM Knowledge Campus, India DIT University, India IGNOU, India Uttarakhand Technical University, India Uttarakhand Technical University, India G. B. Pant University of Agriculture and Technology, India Uttarakhand Technical University, India M. J. P. Rohilkhand University, India Krishna Institute of Engineering and Technology, India Mody University, India Mody University, India SDM College of Engineering and Technology, India Mody University, India Mody University, India Lal Bahadur Shastri Institute of Management, India Sir Padampat Singhania University, India DIT University, India Dr. B.R. Ambedkar Open University, India Banaras Hindu University, India MNNIT, India Chandigarh Engineering College, India Maharishi Markandeshwar University, India Chandigarh Engineering College, India Banasthali University, India Graphic Era University, India Banasthali University, India Gurukul Kangri University, India Dehradun Institute of Technology, India IMS Engineering College, India Gurukul Kangri University, India G.L. Bajaj Institute of Technology and Management, India IIT Chennai, India Babu Banarasi Das National Institute of Technology, India GNIT, India NIET, India GLA University, India GLA University, India

Organization

Anand Singh Jalal Anil Kumar Sagar Rashi Agrawal Rajiv Kumar Hitendra Garg Himani Maheshwari Musrrat Ali Umesh Chandra Sunil Shukla Mohammad Mazhar Afzal Sangeet Srivastava Sanjay Tanwani Urjita Thakar Nirmal Dagdee Nikhil Marriwala

xv

GLA University, India Galgotia University, India Sharda University, India Sharda University, India Hindustan College, India Uttaranchal University, India Glocal University, India Banda University of Agriculture and Technology, India Desh Bhagat University, India Glocal University, India North Cap University, India Devi Ahilya Vishwavidyalaya, India S.G.S. Institute of Technology and Science, India Sushila Devi Bansal College of Technology, India Kurukshetra University, India

Contents

Image Processing, Pattern Analysis and Machine Vision Texture Redefined: A Second Order Statistical Based Approach for Brodatz Dataset Samples 1–35 (A) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Amit Kumar Shakya, Shalini Tiwari, Anurag Vidyarthi, and Rishi Prakash A Taxonomy on Machine Learning Based Techniques to Identify the Heart Disease . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Anand Kumar Srivastava, Pradeep Kumar Singh, and Yugal Kumar Prediction and Analysis of Ecological Climatic Conditions (Humidity/Temperature) of Different Countries Fly-Ash Using Data Mining Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Divya Gupta and Kanak Saxena NStackSenti: Evaluation of a Multi-level Approach for Detecting the Sentiment of Users. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Md Fahimuzzman Sohan, Sheikh Shah Mohammad Motiur Rahman, Md Tahsir Ahmed Munna, Shaikh Muhammad Allayear, Md. Habibur Rahman, and Md. Mushfiqur Rahman

3

13

26

38

A Large Scale Study for Identification of Sarcasm in Textual Data . . . . . . . . Pulkit Mehndiratta and Devpriya Soni

49

Security Issues and Solutions in Cloud Robotics: A Survey . . . . . . . . . . . . . Saurabh Jain and Rajesh Doriya

64

Emotion Recognition in Poetry Using Ensemble of Classifiers . . . . . . . . . . . P. S. Sreeja and G. S. Mahalakshmi

77

An Approach to Threshold Based Human Skin Color Recognition and Segmentation in Different Color Models . . . . . . . . . . . . . . . . . . . . . . . B. S. Sathish, P. Ganesan, S. Dola Sanjay, A. Ranganayakulu, and S. Jagan Mohan Rao Neuro-Fuzzy Approach for Reconstruction of 3-D Spine Model Using 2-D Spine Images and Human Anatomy . . . . . . . . . . . . . . . . . . . . . . Saurabh Agrawal, Dilip Singh Sisodia, and Naresh Kumar Nagwani Multi-objective Planning of Rural Distribution System Using PSO . . . . . . . . Meena Kumari, V. R. Singh, and Rakesh Ranjan

92

102 116

xviii

Contents

A Review of Reverse Dictionary: Finding Words from Concept Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bushra Siddique and Mirza Mohd Sufyan Beg A Graph Based Analysis of User Mobility for a Smart City Project. . . . . . . . Jai Prakash Verma, Sapan H. Mankad, and Sanjay Garg Parkinson’s Disease Diagnosis by fMRI Images Using MFCC Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ritika Dubey, Aditi Narang, and Vijay Khare

128 140

152

Information and Data Convergence A New Approach to Train LSTMs, GRUs, and Other Similar Networks for Data Generation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ravin Kumar Automatic Road Extraction from Semi Urban Remote Sensing Images . . . . . Pramod Kumar Soni, Navin Rajpal, and Rajesh Mehta

163 172

A New Hybrid Backstepping Approach for the Position/Force Control of Mobile Manipulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Manju Rani, Dinanath, and Naveen Kumar

183

Distributed Multi-criteria Based Clusterhead Selection Approach for MANET. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Preeti Yadav, Ritu Prasad, Praneet Saurabh, and Bhupendra Verma

199

Enhanced Slicing+: A New Privacy Preserving Micro-data Publishing Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rajshree Srivastava and Kritika Rani

210

Discovering Cyclic and Partial Cyclic Patterns Using the FP Growth Method Incorporated with Special Constraints . . . . . . . . . . . . . . . . . . . . . . Pragati Upadhyay, Narendra Kohli, and M. K. Pandey

221

Disruptive Technologies for Future iDJ: Intrusion Detection System in Java . . . . . . . . . . . . . . . . . . . . . . . . . . . Radhesh Khanna and Shashank Sharma An Effective Optimisation Algorithm for Sensor Deployment Problem in Wireless Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Vishal Puri, A. Ramesh Babu, T Sudalai Muthu, and Sonali Potdar A Review of Sensor Deployment Problem in Wireless Sensor Network . . . . . Vishal Puri, A. Ramesh Babu, T Sudalai Muthu, and Kirti Korabu

239

248 259

Contents

xix

Analysis of Energy Efficient Framework for Static and Mobile Nodes in WSN-Assisted IoT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Anurag Shukla and Sarsij Tripathi

269

Energy Aware SEP Based on Dual Cluster Heads and Dual Sink Approach in WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sanjay Kumar, Narendra Harariya, and Arka Bhowmik

281

Impact of Cranial Electrical Stimulation Based Analysis of Heart Rate Variability in Insomnia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Khyatee, Aparna Sarkar, and Rajeev Aggarwal

296

E-Governance and Smart World EAAP: Efficient Authentication Agreement Protocol Policy for Cloud Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Narander Kumar and Jitendra Kumar Samriya Multi-path Priority Based Route Discovery Mechanism . . . . . . . . . . . . . . . . Swati Atri and Sanjay Tyagi

311 321

Secure and Trustworthy Cloud: Need of Digital India, an e-Governance Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Archana B. Saxena and Meenu Dave

330

Hyperheuristic Framework with Evolutionary and Deterministic Algorithms for Virtual Machine Placement Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . Amol C. Adamuthe and Akshayya Jadhav

340

Efficient Directional Information Dissemination Approach in Vehicular Ad Hoc Networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sandeep Kad and Vijay Kumar Banga

351

Software Agents in Support of Scheduling Group Training. . . . . . . . . . . . . . Giorgi Mamatsashvili, Konrad Gancarz, Weronika Łajewska, Maria Ganzha, and Marcin Paprzycki

364

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

377

Image Processing, Pattern Analysis and Machine Vision

Texture Redefined: A Second Order Statistical Based Approach for Brodatz Dataset Samples 1–35 (A) Amit Kumar Shakya(&), Shalini Tiwari, Anurag Vidyarthi, and Rishi Prakash Department of Electronics and Communication, Graphic Era (Deemed to be University), 566/6 Bell Road Clement town, Dehradun, Uttarakhand, India [email protected] Abstract. In this research we have used the Brodatz dataset for redefining the texture features. We have computed the second order image statistical parameters like contrast, correlation, energy and homogeneity for defining the features. These features are texture visual features which are affected by the human visual perception. We have computed these features for the first 35 textured surfaces obtained from the Brodatz dataset and on the basis of this we have concluded that which surface have obtained maximum and minimum statistical value and its effects on the human visual perception. Keywords: Brodatz dataset
Texture
Contrast Homogeneity
Human visual perception


Correlation
Energy


1 Introduction Texture is playing an important role in image analysis and interpretation. The features derived from the texture analysis are widely used in the field of remote sensing [1], bio medical imaging [2], SAR data analysis [3], fabrics defect detection, computer vision and pattern recognition and many more. Texture is categorised in many forms like smooth [4], coarse [5], regular [6], irregular [7], fine, etc., it is considered as the statistical property of the surface. Today the texture is used in the field of remote sensing where interpretation of the large dataset has become quite easy. Texture features based upon the statistical features was first defined by Haralick, he proposed a set of 14 features for defining the texture i.e. contrast, correlation, energy, homogeneity, entropy, sum average, sum variance, inverse difference moment, measure of correlation, maximum correlation coefficients etc., [8]. Later another scientist Gotlieb categorised these 14 features in four different categories i.e. correlation measures, entropy measures, statistical measures and finally texture visual measures. Contrast, correlation, energy and homogeneity forms the texture visual measures group, these features all together combined to form the human visual perception i.e. surfaces which look similar should have same sought of relation in between them. In this research work we have quantified the texture features and on the basis of this we have concluded that which sample feature have obtained the maximum or minimum feature value. © Springer Nature Singapore Pte Ltd. 2019 M. Prateek et al. (Eds.): NGCT 2018, CCIS 922, pp. 3–12, 2019. https://doi.org/10.1007/978-981-15-1718-1_1

4

A. K. Shakya et al.

2 Methodology Based on the Second Order Statistical Parameters Statistical measures of the image are classified in three categories first order, second order and third or higher order, among these higher order is useless from the point of view of human visual perception, and the first order do not provide sufficient information about the statistical features. First order computes mean, variance, standard deviation. Second order features are most suitable when it comes to human visual perception. Here we are providing information about texture visual parameters only. 1. Contrast: fr ¼

Xgl1 0

n2 

Xgl Xgl r¼1

c¼1

pðr; cÞ

ð1Þ

Here gl ¼ number of distinct grey levels, jr  cj ¼ n ðdifference in the grey level of the image pixelÞ. pðr; cÞ ¼ image pixel of the texture evaluation The contrast of the image is always greater than zero i.e. ‘0’. 2. Correlation: fr ¼

X X ðr; cÞpðr; cÞ  lx ly r c rx ry

ð2Þ

Here lx ; ly and rx ; ry . are the mean and standard deviation of the row and column respectively. The range of the correlation is (−1, 1). 3. Energy: X X fr ¼ ð3Þ jpðr; cÞj2 r c The range of the energy is (0, 1). If the energy has obtained value 1 then it means that it is a constant image. It is also known as angular second moment. 4. Homogeneity: X X 1 fr ¼  pðr; cÞ ð4Þ r c 1 þ ð r  cÞ 2 It is the measure of the spatial closeness of the image pixel. The range of the homogeneity is (0, 1). It is also known as inverse difference moment (IDM). In this experiment we have taken first 35 sample texture images of Brodatz texture dataset and evaluated the second order statistical parameters for the dataset. Finally we have quantified each sample on the basis of second order statistical parameters and obtained the result as maximum and minimum contrast, correlation, energy and homogeneity.

3 Classification of the Brodatz Dataset Sample 1–35 In this classification scheme we have classified the texture of the surface on the basis of statistical measures and computed there grey versions and finally make a quantification of the surface texture features. Here we have presented first 35 samples of Brodatz

Texture Redefined: A Second Order Statistical Based Approach

5

texture which are classified in categories of grass, bark, straw, weave, cloth etc. The entire surfaces are different from each other in term of spectral as well as spatial relationship between the image pixels. Here we are presenting the original texture surfaces as prepared by Brodatz (Fig. 1).

(1)Grass

(2) Bark

(3) Straw

(4) Herringbone weave (5) Woollen cloth

(6) Pressed calf leather (7) Beach sand (8) Water (D 68)

(11)Pigskin

(12) Brick wall

(9) Wood grain

(13) Plastic bubbles (14) Grass

(10) Raffia

(15) Bark

(16)Straw (17) Herringbone weave (18) Woollen cloth 2 (19) Pressed calf leather (20) Beach sand

(21)Water

(22) Wood grain

(26) Plastic bubbles (27) Grass

(23) Raffia

(24)Pigskin

(28) Bark

(29) Straw

(25) Brick wall

(31) Herringbone weave (32) Pressed calf leather (33) Beach sand (34) Water

Fig. 1. Brodatz dataset of the original texture images

(30) Woollen cloth

(35) Wood grain

6

3.1

A. K. Shakya et al.

Grey Level Classification of the Brodatz Dataset 1–35

Here we are showing the grey levels of the dataset which are obtained along with the quantification of the texture features. The grey level representation is the first step for visual interpretation, which is followed by the texture feature quantification (Fig. 2).

(1) Grass

(5) Woollen Cloth

(9) Wood Grain

(13) Plastic Bubble

(17) Herring weave

(2) Bark

(3) Straw

(6) Pressed calf leather (7) Beach Sand

(10) Raffia

(11) Pigskin

(14) Grass

(15) Bark

(18) Woollen cloth

(4) Herringbone weave

(8) Water

(12) Brick Wall

(16) Straw

(19) Pressed calf leather (20) Beach Sand

Fig. 2. Visual variations in the Brodatz dataset through GLCM

Texture Redefined: A Second Order Statistical Based Approach

(21) Water

(22) Wood grain

(23) Raffia

(24)Pigskin

(25) Brick wall

(26) Plastic bubble

(27) Grass

(28) Bark

(29) Straw

(30) Woollen cloth

(33) Beach sand

7

(31) Herring bone weave (32) Pressed calf leather

(34) Water

(35) Wood grain

Fig. 2. (continued)

Now we have categorised the dataset in terms of texture features i.e. on the basis of contrast, correlation, energy and homogeneity. The quantification of the texture feature is done to obtain specific behaviour of the GLCM features (Table 1). Con* = Contrast, Corr* = Correlation, E* = Energy, H* = Homogeneity, D = Degree

Table 1. Quantification of the statistical features for the Brodatz dataset

S.No

Dataset

1 Grass D9

Con* Corr* E* H*

0D 3.3915 0.3981 0.0302 0.5274

Texture features 45 D 90 D 135 D 3.9206 2.8980 4.2879 0.2995 0.4822 0.2339 0.0293 0.0333 0.0286 0.5124 0.5685 0.4986

Average 3.6245 0.3534 0.0303 0.5267

8

A. K. Shakya et al.

2

Bark D12

3

Straw D15

4

Herringbone Weave D15

5

Woollen cloth D19

6

Pressed calf leather D24

7

Beach sand D29

8

Water D38

9

Wood grain D68

10

Raffia D84

11

Pigskin D92

12

Brick wall D94

Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr*

1.6728 0.6285 0.0637 0.6467 1.8232 0.4105 0.0631 0.6150 1.2158 0.3836 0.1032 0.6757 0.7179 0.5334 0.1278 0.7302 2.6037 0.1848 0.0538 0.5717 0.9202 0.4548 0.1530 0.7279 0.4377 0.3508 0.2621 0.7873 1.2596 0.2732 0.1356 0.6975 0.7290 0.6260 0.1101 0.7486 0.5475 0.8274 0.1584 0.7739 1.0180 0.5311

2.1225 0.5227 0.0600 0.6221 1.4548 0.5190 0.0735 0.6702 1.6987 0.0967 0.0948 0.6103 0.9000 0.3906 0.1207 0.7016 2.8313 0.1015 0.0522 0.5541 1.1919 0.2627 0.1439 0.6979 0.5198 0.1264 0.2633 0.7876 1.3510 0.1735 0.1360 0.6957 1.2889 0.3132 0.0861 0.6539 0.8266 0.7340 0.1457 0.7325 1.1260 0.4635

1.3834 0.6889 0.0709 0.6827 1.9137 0.3672 0.0631 0.6128 1.2861 0.3167 0.1036 0.6729 0.7191 0.5132 0.1315 0.7380 1.5442 0.5101 0.0678 0.6521 1.0244 0.3664 0.1524 0.7229 0.2414 0.5996 0.3775 0.9200 0.3001 0.8182 0.2484 0.9010 1.0742 0.4277 0.0959 0.6986 0.7221 0.7677 0.1556 0.7619 0.5197 0.7525

2.2836 0.4864 0.0575 0.6049 2.8015 0.0734 0.0571 0.5520 1.7924 0.0468 0.0933 0.6050 0.8640 0.4150 0.1222 0.7074 2.6329 0.1645 0.0535 0.5674 1.3344 0.1743 0.1378 0.6763 0.5425 0.0878 0.2596 0.7756 1.3924 0.1481 0.1343 0.6902 1.4992 0.2009 0.0846 0.6327 0.8281 0.7335 0.1474 0.7383 1.1351 0.4591

1.865575 0.581625 0.0630 0.6391 1.9983 0.3425 0.0642 0.6125 1.4982 0.21095 0.09872 0.64097 0.80025 0.46305 0.12555 0.7193 2.403025 0.240225 0.056825 0.586325 1.117725 0.31455 0.146775 0.70625 0.43535 0.29115 0.290625 0.817625 1.075775 0.35325 0.163575 0.7461 1.147825 0.39195 0.094175 0.68345 0.731075 0.76565 0.151775 0.75165 0.9497 0.55155

Texture Redefined: A Second Order Statistical Based Approach

12 13

Brick wall D94 Plastic Bubble D112

14

Grass D9 HE

15

Bark D12 HE

16

Straw D15 HE

17

Herringbone Weave D16 HE

18

Woollen cloth D19 HE

19

Pressed calf leather D24 HE

20

Beach Sand D29 HE

21

Water D38 HE

22

Wood Grain D68 HE

E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H*

0.1548 0.7468 0.9901 0.6860 0.0738 0.7200 6.5910 0.3731 0.0184 0.4673 4.1313 0.6095 0.0238 0.5424 6.5233 0.3839 0.0185 0.4716 6.9949 0.3385 0.0180 0.4652 4.8269 0.5395 0.0214 0.5149 8.3313 0.2029 0.0173 0.4334 5.9448 0.4387 0.0199 0.5020 8.3755 0.2085 0.0167 0.4351 7.4935 0.3023 0.0180 0.4626

0.1503 0.7337 1.3115 0.5804 0.0644 0.6763 7.4631 0.2883 0.0172 0.4503 4.9918 0.5272 0.0213 0.5166 4.6801 0.5569 0.0227 0.5359 9.6059 0.0889 0.0161 0.3969 5.9601 0.4296 0.0190 0.4816 9.0356 0.1334 0.0164 0.4163 7.3861 0.3006 0.0175 0.4630 8.3301 0.2104 0.0168 0.4373 7.4900 0.3006 0.0180 0.4609

0.1794 0.8110 1.0925 0.6506 0.0714 0.7080 5.5255 0.4731 0.0204 0.5094 3.3395 0.6836 0.0272 0.5798 6.6917 0.3665 0.0182 0.4691 6.5565 0.3783 0.0182 0.4643 4.4943 0.5699 0.0222 0.5263 4.6931 0.5499 0.0218 0.5211 6.1893 0.4142 0.0194 0.4967 1.5333 0.8546 0.0383 0.6720 0.9298 0.9134 0.0493 0.7419

0.1500 0.7328 1.4700 0.5297 0.0625 0.6643 8.1747 0.2205 0.0166 0.4349 0.4751 5.5423 0.0201 0.4975 9.9553 0.0574 0.0157 0.4010 10.1318 0.0390 0.0158 0.3916 5.6600 0.4584 0.0194 0.4887 8.4051 0.1939 0.0168 0.4302 8.5367 0.1917 0.0164 0.4352 8.8903 0.1574 0.0164 0.4219 7.4900 0.3006 0.0180 0.4609

0.158625 0.756075 1.216025 0.611675 0.068025 0.69215 6.938575 0.33875 0.01815 0.465475 3.234425 0.84065 0.0231 0.534075 6.9626 0.341175 0.018775 0.4694 8.322275 0.211175 0.017025 0.4295 5.235325 0.49935 0.0205 0.502875 7.616275 0.270025 0.018075 0.45025 7.014225 0.3363 0.0183 0.474225 6.7823 0.357725 0.02205 0.491575 5.850825 0.454225 0.025825 0.531575

9

10

23

A. K. Shakya et al.

Raffia D84 HE

23 24

Pig skin D92 HE

25

Brick wall D94 HE

26

Plastic bubble D112 HE

27

Grass D9_1024

28

Bark D12_1024

29

Straw D15_1024

30

Woollen cloth D19_1024

31

Herringbone Weave D16_1024

32

Presses calf leather D24_1024

33

Beach sand D29_1024

Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr* E* H* Con* Corr*

3.8777 0.6305 0.0270 0.5737 4.5635 0.5879 0.0261 0.5375 4.7922 0.5461 0.0241 0.5560 3.0991 0.7041 0.0294 0.6076 1.1983 0.7142 0.0563 0.6677 0.7409 0.8255 0.0823 0.7401 0.9723 0.7267 0.0665 0.6882 1.4079 0.6081 0.0772 0.6689 1.4192 0.7325 0.0449 0.6529 1.0350 0.6331 0.0906 0.7029 0.4615 0.7209

6.9109 0.3399 0.0179 0.4622 6.6832 0.3953 0.0210 0.4839 5.1596 0.5098 0.0228 0.5404 4.1866 0.5993 0.0243 0.5576 1.6318 0.6108 0.0504 0.6341 0.9952 0.7656 0.0733 0.7034 0.7494 0.7894 0.0756 0.7313 1.6567 0.5386 0.0719 0.6454 1.9255 0.6371 0.0387 0.6061 1.2549 0.5551 0.0840 0.6783 0.5964 0.6394

5.5717 0.4677 0.0214 0.5202 5.3173 0.5190 0.0238 0.5231 2.7641 0.7374 0.0329 0.6171 3.3982 0.6748 0.0277 0.5941 1.0036 0.7606 0.0626 0.7044 0.6546 0.8460 0.0864 0.7566 0.3760 0.8944 0.1071 0.8292 1.1471 0.6805 0.0831 0.6934 1.1859 0.7766 0.0471 0.6669 0.6643 0.7645 0.1063 0.7589 0.4543 0.7253

6.9109 0.3399 0.0179 0.4622 6.4730 0.4143 0.0214 0.4938 5.2149 0.5046 0.0226 0.5387 4.6631 0.5536 0.0231 0.5447 1.7063 0.5930 0.0489 0.6242 0.9949 0.7657 0.0728 0.7015 1.3218 0.6285 0.0589 0.6452 1.5108 0.5792 0.0744 0.6570 1.9928 0.6244 0.0382 0.6011 1.1955 0.5762 0.0855 0.6858 0.6576 0.6024

5.8178 0.4445 0.02105 0.504575 5.75925 0.479125 0.023075 0.509575 4.4827 0.574475 0.0256 0.56305 3.83675 0.63295 0.026125 0.576 1.385 0.66965 0.05455 0.6576 0.8464 0.8007 0.0787 0.7254 0.854875 0.75975 0.077025 0.723475 1.430625 0.6016 0.07665 0.666175 1.63085 0.69265 0.042225 0.63175 1.037425 0.632225 0.0916 0.706475 0.54245 0.672

Texture Redefined: A Second Order Statistical Based Approach

33

Beach sand D29_1024 Water D38_1024

34

35

E* H* Con* Corr* E* H* Con* Corr* E* H*

Wood grain D68_1024

0.1841 0.8107 0.2679 0.3799 0.4222 0.8669 0.4810 0.5948 0.3320 0.8319

0.1690 0.7814 0.2710 0.3726 0.4208 0.8654 0.4730 0.6016 0.3319 0.8321

0.1842 0.8109 0.2267 0.4753 0.4438 0.8868 0.1507 0.8732 0.4024 0.9249

0.1620 0.7674 0.2811 0.3494 0.4161 0.8606 0.5063 0.5735 0.3276 0.8267

11

0.174825 0.7926 0.261675 0.3943 0.425725 0.869925 0.40275 0.660775 0.348475 0.8539

9 8 7 6 5 4 3 2 1 0

Comparative Contast of Sample 1 to 35

Raamge of Texture Feature Correlation

Range of Texture Feature Contrast

Now the comparative plot of the obtained texture features (Fig. 3).

1 4 7 101316192225283134 Brodatz Samples (1 to 35)

Comparative Correlation of Sample 1 to 35 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 20

(b)

(a) Comparative Energy of Sample 1 to 35 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0 20 40 Brodatz Sample (1 to 35)

(c)

Comparative Homogeniety of Sample 1 to 35 1 Range of Texture Feature Homogeniety

Raange of Texture Feature Energy

40

Brodatz Sample (1 to 35)

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

20

40

Brodatz Sample ( 1 to 35 )

(d)

Fig. 3. (a) Comparative plot of the feature contrast (b) Comparative plot of the feature correlation (c) Comparative plot of the feature energy (d) Comparative plot of the feature homogeneity

12

A. K. Shakya et al.

4 Conclusion The result of this research work is Herringbone weave D16_H6 has obtained highest contrast whereas Wood Grain D6Q_1024 obtained minimum contrast. Bark D12_HE has obtained maximum correlation, whereas Herringbone weave D15 have obtained minimum correlation. Energy is highest for Water D38_1024 and minimum for Grass D9_HE. The feature homogeneity has obtained maximum value for Water and minimum for Herringbone weave D15. Acknowledgement. Authors like to express their deep gratitude and thanks to Department of Electrical Engineering, University of South California (USC), for providing the Brodatz dataset used in this research work.

References 1. Yuan, J., Wang, D., Li, R.: Remote sensing image segmentation by combining spectral and texture features. Trans. Geosci. Remote Sens. 52(1), 16–24 (2014) 2. Depeursinge, A., Rodrigueza, A.F., Villeb, D.V., Muller, H.: Three–dimensional solid texture analysis in biomedical imaging: review and opportunities. Med. Image Anal. 18(1), 176–196 (2014) 3. Zhai, W., Shen, H., Huang, C., Pei, W.: Fusion of polarimetric and texture information for urban building extraction from fully polarimetric SAR imagery. Remote Sens. Lett. 7(1), 31– 40 (2016) 4. Jamkara, S.S., Raob, C.B.K.: Index of aggregate particle shape and texture of coarse aggregate as a parameter for concrete mix proportioning. Elsevier Cement Concrete Res. 34(11), 2021–2027 (2004) 5. Lu, Y., Tsin, Y., Lin, W.C.: The promise and perils of near-regular texture. Int. J. Comput. Vis. 62(1/2), 142–159 (2005) 6. Nemoto, K., Yanagi, K., Aketagawa, M.: Development of a roughness measurement standard with irregular surface topography for improving 3D surface texture measurement. Meas. Sci. Technol. 20, 1–7 (2009) 7. Haralick, R.M.: Statistical and structural approaches to texture. Proc. IEEE 67(5), 786–804 (1979) 8. Gotlieb, C.C., Kreyszig, H.E.: Texture descriptors based on co-occurrence matrices. Comput. Vis. Graph. Image Process. 51(1), 70–86 (1990)

A Taxonomy on Machine Learning Based Techniques to Identify the Heart Disease Anand Kumar Srivastava1,2(&), Pradeep Kumar Singh2(&), and Yugal Kumar2(&) 1

Department of Computer Science and Engineering, ABES Engineering College, Ghaziabad, UP, India [email protected] 2 Department of Computer Science and Engineering, Jaypee University of Information Technology, Waknaghat, Solan, HP, India [email protected], [email protected]

Abstract. Every year average death of human being is 17.7 million caused by Heart Disease or Cardiovascular diseases (CVDs), which is 31% of all global deaths reflected in Survey of World Heart Day 2017 [33]. In September 2016, many countries have taken the various Global Hearts Initiative in prediction and diagnosis of heart Diseases at earlier stages so that it can be cure perfectly [5]. Many authors have studied in this filed to optimize the performance of various ML techniques using various approaches. In latest studies, many groups have uncovered that many optimization algorithm like Differential Evolution, Genetic Variants, and Particle Swarm optimization are associated with prediction algorithm like K-Nearest neighbor, Decision Tree, Neural Network, Support Vector machine, Logistic Regression etc. to make efficient medical system for CVDs. The Objective of our current study is to analyze the comparatively study of the ML techniques in terms of performance measure of different ML techniques that have been used by various authors in their research work in earlier studies of heart disease prediction and diagnosis process. Keywords: Heart disease
Data mining
Machine learning
Heart failure
Logistic regression (LR)
K- means (KM)
Decision Tree (DT)
Support vector machine (SVM)
Multivariate adaptive regression splines (MARS)
Rough sets (RS)
K-Nearest neighbor (KNN)
Genetic algorithm (GA)
Neural network (NN)
Differential Evolution (DE)
Dictionary-based keyword spotting (DBS)
Natural language Process (NLP)
Random forest (RF)
Classification and regression tree (CART)
Naïve Bayes (NB)
Classification by clustering (CC)
Multilayer perceptron (MLP)
Radial basis function (RBF)
Single conjunctive rule learner (SCRL)
Bagged Tree (BT)
AdaBoost (AB)
Particle swarm optimization (PSO)

1 Introduction In human body heart is very crucial part, through blood vessels it pumps the blood to complete body parts, so other body parts get the oxygen and essential nutrients from bloods as well as heart assists in the removal of metabolic wastes. © Springer Nature Singapore Pte Ltd. 2019 M. Prateek et al. (Eds.): NGCT 2018, CCIS 922, pp. 13–25, 2019. https://doi.org/10.1007/978-981-15-1718-1_2

14

A. K. Srivastava et al.

So it’s very important that heart works properly. According to WHO on the basis of one survey as data available in 2015 [32], maximum human deaths (appox 17.7 million) are caused by the heart diseases and there are some factors that put people at increase risk for heart disease like Smoking, high cholesterol, high blood pressure, genetic effect, diabetes, obesity, etc. and few symptom that indicate the heart disease Chest pain (angina), Shortness of breath, Sweating, Irregular heartbeat, Nausea, heart burn, heart attacks. Machine learning techniques can be used in heart disease prediction at earlier stages so that it can be cure within a desirable time period. In last 15 years many studied carried out in this field. Objective of each studied is to optimize the implemented algorithm with some suitable proposal. 1.1

Data Mining Techniques

Data Mining techniques are the process to convert raw data into useful information or we can say that it is the process to find some useful hidden patterns of data to predict some outcomes. These are the important DM techniques: – – – – –

Decision Tree Clustering Association Classification Prediction

In data mining techniques Association method works on the rule if-else procedure. It helps to discover the patterns between two items or data. Association rules are useful for analyzing and predicting the different diseases. Classification data mining technique is the example of supervised learning technique. Classification algorithm predicts categorical class level. Prediction algorithm is also example of supervised learning technique but it predicts continuous valued function. For example we can suggest a classification model that predicts that heart is working correctly or not, that means answer will be yes or not on the basis of symptoms of patients. And in same case prediction algorithm predict the probability of chances of heart failure. Clustering is the example of unsupervised learning techniques. It is the process of creating the group or classes of similar behavioral objects or data. In clustering process on the basis of similar characteristics of objects or data and using any automatic techniques we makes meaningful cluster or classes. J. Ross Quinlan in 1980 developed decision tree technique also known as ID3 later he presented C4.5. Decision tree algorithm concept is based on greedy approach and follows the tree structure. 1.2

Machine Leaning Algorithm

ML uses the DM techniques and other learning techniques to build model, so that we can predict the future outcomes. If we compare the DM and ML, Dm is the process to extract the hidden information or patterns from a particular dataset whereas ML deals

A Taxonomy on Machine Learning Based Techniques

15

the algorithm improves automatically through experience based on data. Basically ML uses the learning techniques, on the basis of leaning algorithm ML can be categorize into 3 category Supervised Learning, unsupervised learning and reinforcement learning. 1.2.1 Supervised Learning It can be explained that it uses training or labeled data to learn the mapping function F() from the (Z) input variable with (Y) output variable. Y ¼ FðZÞ Again supervised leaning problem can be divided into 2 category classification and regression. Examples of Supervised learning are Logistic Regression, KNN, CART, blog– Linear Regression, Naïve Bayes. Ensembling is one type of supervised learning, In which we use the multiple combinations of weal ML models to predict future outcomes on new sample. Some examples are Bagging with random forest, Boosting with XGBoost. In [4], Author used the different ensembling methods in prediction of of heart disease and computed the accuracy of each techniques. 1.2.2 Unsupervised Learning It only processes the input variable (Z) with no corresponding output variables and uses the unlabeled data to build mode for future prediction. Unsupervised learning problems can be divided into 3 types Association, clustering and Dimensionality Reduction. Dimensionality Reduction can be done using Feature selection methods and Extraction methods. In the process of Feature Extraction actually data transform from multi dimensional space to small or low dimensional space. K- means, Apriori, and PCA are examples of unsupervised learning.

2 Related Work Since the last decade, many studies have been implemented and evaluated the performance measures in process of heart disease prediction and diagnosis. Different Authors used different algorithm with different feature set in prediction and diagnosis of heart diseases especially in heart failure. Vivekandan et al. implemented a scheme to used the modified Differential Evolution algorithm for optimize feature selection [1]. Author has identified 13 features for the heart disease prediction, diagnosis and dataset taken from UCI repositories [2], further reduced these attributes to nine attributes for improving the accuracy. In this study author adapt the Fuzzy AHP algorithm to obtain the relative weight of each feature. Further author implemented feed forward NN algorithm for prediction and diagnosis of heart disease. In [3], the author adapt the SVM classifier along with forward feature selection, back elimination feature and forward feature inclusion and implemented SPECTF dataset from UCI. Author concluded that, current approach obtained smaller subsets

16

A. K. Srivastava et al.

and increases the performance accuracy of diagnosis compared to back elimination techniques and forward inclusion. Shao et al. [4] proposed as feature selection algorithm rough set, multivariate adaptive regression splines and logistic regression to reduce the set of critical features for heart disease prediction and diagnosis with higher accuracy. In this study author analyzed hybrid multivariate adaptive regression splines artificial neural network algorithm is the best alternative because this method contained the less number of explanatory variables (features) and provided the efficient classification accuracy on prediction. On the other hand Oluwarotimi et al. [6] suggested hybrid algorithm based on Fuzzy-AHP and ANN mythology for Heart Failure and risk prediction using medical dataset [23]. This prediction process consists of two sequential stages. At first stage author implemented Fuzzy-AHP to choose rank and obtained the local weights of Heart Failure features and for set of given attributes author computed global weight, Author’s implanted algorithm and declared that an average accuracy is 91.10%, which is more better than conventional Artificial neural network method. (4.40% higher in comparison) Jabbar et al. implemented a classification algorithm which is the combination of genetic and KNN algorithm to predict and diagnose the heart disease of a patient for Andhra Pradesh population [7]. Author used 7 data sets computed the accuracy with and without GA along with KNN and concluded. In [19], Performance Analysis of Ensemble classifier Random Forest, AdaBoost, Bagged Tree with PSO on parameter like accuracy, sensitivity, specificity, PPV, NPV are performed. Author used the Heart Disease Dataset which contains 270 instances and 14 attributes and concluded that Particle Swarm Optimization (PSO) is computationally efficient (inexpensive) in context to speed and memory. Pouriyeh et al. [16] compared MLP, Decision Tree, NB, K-NN, SVM, SCRL and RBF classifiers in two experiments. In first experiment author evaluated the accuracy with the help of 10 fold cross validation method and further to estimate the efficiency performance of each algorithm. Author used K = 1, 3, 9 and 15 In case of KNN classifier and obtained the performance of KNN is best at K = 9 using whole data set [17]. In second experiments author use the bagging, boosting and stacking concept for evaluation of algorithm efficiency. In this study Author found that, no improvement in support vector machine performance but, maintained its performance (accuracy) level in case of Bagging. SCRL improved 10.56%, it increased from 69.96% to 80.52% and Decision tree improved 0.99% (from 77.55%–78.54%). In case of Boosting SCRL improved 11.22% and Decision Tree improved 4.62% (from 77.55% to 82.17%). The others algorithm performance remained same as earlier. Where the combination of SVM and MLP has the best accuracy 84.15% in case of stacking. In [15], Soni et al. analyzed the performance efficiency of Naïve Bayes, Decision Tree, ANN, classification via clustering Algorithm in heart disease prediction and diagnosis. In this study, research is divided into two phases, the first part author used the 25 set of Attributes and implemented the algorithm, in second phase author used the Genetic Algorithm for optimizing the feature set 25 to 15 while applying the classifying algorithm and found that efficiency is optimized as compare with first case.

A Taxonomy on Machine Learning Based Techniques

17

In optimization of feature set from data set, Yan et al. used the genetic algorithm in selection of 24 critical features among 40 features available from data set [14]. With these features implemented system provides better diagnostic performance. Anbarasi et al. compared the performance measure in the context of accuracy in percentage of Decision Tree, classification via clustering, Naïve bayes, in prediction of heart disease [12]. Author used the Genetic Algorithm to select the critical attributes which participate more closely in the heart risk, HF prediction and diagnosis. This result reduces the required no of tests taken by patient indirectly. 13 attributes (features) are compressed to six attributes using genetic search with dataset Cleveland Heart Disease database [31]. Further apply the classifiers and investigated the performance. Guidi et al. implemented Heart Failure clinical decision support system along with a portable kit to acquire the set of clinical attributes [11]. This system provided various functioning with the using of different ML techniques. Author selected 12 attributes from database, data taken in period of 2001 to 2008 from Hospital of Maria Nuova, Florence, Italy. Author compared different types of ML techniques and declared that the CART method provide the best result among all ML. Also give the humanly understandable decision making process, Classification and regression tree provides the accuracy of 87.6% in type prediction and 81.8% in severity assessment Olaniyi et al. compared the efficiency performance efficiency of SVM and Back propagation Neural Network in heart disease prediction and diagnosis [9]. In this research implementation works are divided into two phases, the first part author used back propagation training and testing of the network while in second part prediction applied using support vector machine. In [26], Author implemented a automatic system that automate the feature design for numerical sequence classification using Genetic programming, system is called Autofeed which is fully data driven. Fen et al. designed comprehensive risk model using a modified method of random survival forest technique for predicting and diagnose the heart failure mortality with a high level of performance accuracy [20]. Author worked on MIMIC II database With 32 variables and achieved 82.1% performance. In 2017 [21], Jin et al. designed robus and effective paradigm for heart failure prediction at earlier stage using short-term memory network (LSTM) model and author compared with methods such as Adaboost, Rnadom Forest and LR. Author showed that these methods performs best in the prediction and diagnose of heart failure. Pu et al. proposed the method, with this we predict and prevent all kinds of diseases including cardiovascular diseases on the basis of Genes detection [22]. Because in human body Genes generally stable for long time after birth. Author finds that gene detection plays an important role in prediction of heart diseases. Hui et al. [8] worked with data set [25] and proposed a system which extract the information and identify the risk factor for heart disease from clinical text using NLP techniques. Author analyzed the characteristics of clinical evidence and categorized into 3 main types, sentence level clinical measurements, token level clinical entities and sentence level clinical fact. Nahar et al. Analyzed the promising techniques for feature selection for prediction for heart disease author proved that motivated feature selection process (MFS) and

18

A. K. Srivastava et al.

MFS along with computerized feature selection process (CFS) are the best suited feature selection techniques [28]. Author used the SVM and Naïve Base classifier prediction and found that SVM predict more efficient then Naïve Base.

3 Research Methodology This section discusses the different approaches or techniques used by various author in their studies. Broadly on the basis of literature work we conclude that in heart disease prediction and diagnosis process every author used the feature extraction phase using any feature selection algorithm, prediction and diagnosis phase using any machine learning algorithm. Each author worked on any particular medical dataset [2, 10, 18, 24, 27], which contain various instances and attributes. 3.1

Feature Extraction Phase

Prediction of any diseases is completely based on the features selections. Feature selection is the process to define few important and critical features from available feature dataset. We also know the 80–20 rule means 80% effects comes from the 20% effect [29]. So it’s very important to selection of critical variables or features in the process of prediction and diagnosis. In the progress of diseases prediction expert system many authors used different optimization algorithm in feature selection process in last decades. Optimization algorithm provides the optimal feature set because minimal feature set provide efficient result and this set consist of no irrelevant attributes. In today’s scenario there are many optimization techniques available, all techniques work using heuristic based and gradient based search techniques in stochastic and deterministic contexts correspondingly. To make large acceptability of the optimization approach, medical expert systems are desired to develop using robust and efficient optimization algorithms. There are few example of optimization algorithm used in heart diseases prediction and diagnosis mainly are particle swarm optimization (PSO), Evolutionary algorithms simulated annealing and ant colony optimization. 3.1.1 Differential Evolution Optimization Differential Evolution (DE) algorithm is one type of heuristic approach; it has mainly 3 advantages [30]. First, whatever initial parameter values is it can find true global minimum value. Second, is its speed, robustness and third, is less no of control parameters. DE developed in 1997 by Price and Storm. It is the branch of evolutionary programming. The Coverage speed is the main issues in Evolutionary Algorithm, there are many studies have carried out to increase the speed with DE. It is mutation scheme that make self adaptive and self selection process. It is the stochastic optimization method that minimizes an objective function. DE Algorithm is population based same as genetic algorithm using same operators like crossover, mutation and selection. But DE mainly relies on mutation in search operation.

A Taxonomy on Machine Learning Based Techniques

19

3.1.2 Genetic Optimization Genetic Algorithm (GA) is also population based same as the DE algorithm. It deals with constrained and unconstrained optimization problem that is completely based on natural selection [30]. GA iteratively modify the population of individual solutions. While creating the next generation from current population GA use the following three Rules (a) Selection Rules (b) Crossover Rules (c) Mutation Rules The best point in the population approaches an optimal solution and GA follows the population approaches. The fundamental difference between GA and De is GA based on crossover whereas DE or Evolutionary Algorithm based on mutation for search mechanism. Generally Genetic Algorithm needs the fitness function for evaluation of solution domain and genetic representation for solution domain. 3.1.3 Particle Swarm Optimization PSO is also population based optimization technique. PSO developed in 1995 by Dr. Kennedy and Dr. Eberhart. Both Scientist were inspired with fish schooling or bird flocking. PSO has multiple advantages over GA like PSO is easy to implement and needs few parameter to adjust. Algorithm start with set of random particle i.e. solution and then searches for best particle by updating generations. PSO and GA both algorithm have many similarities like PSO and GA start with group of random generated population and search for best by using random technique and both have memory. Only difference is that PSO does not have operators like crossover and mutation. Particle updates them with internal velocity. PSO have been applied in many research and application area in past several years. The best part with PSO is it does require few parameters to adjust, which makes PSO is more attractive for researchers. If we compare with other optimization algorithm following 2 points make stronger to PSO (a) Fast (speed) (b) Less costly (cheaper) 3.2

Prediction and Diagnosis Phase

Various ML Techniques involved in prediction and diagnosis of CVDs in different studies with different dataset. Each ML techniques have own strength and limitations. Concept of Neural network is inspired from human brain concept. NN is based on a collection of connected nodes called artificial neurons. Nodes are called processing element. NN learn from itself from examples. There can be 3 layer input layer, hidden layer (not in all cases), output layer. In some simple cases only input and output layer can provide the result.

20

A. K. Srivastava et al.

K- Means clustering is one type of unsupervised learning technique [34]. It used for unlabeled data. It finds the different group on basis of similarities of objects or data. K shows the no of groups. Algorithm iteratively works and rearranges the attributes in matched groups at each time. Logistic Regression is used when the dependent variable (target) is categorical. Example, Suppose we want to predict email is spam or not means yes (1) or No (0), Whether the HF is malignant (1) or not (0). Multivariate Adaptive Regression Splines Jerome H. Friedman introduced MARS in 1991 in a form of regression analysis. It is also called non parametric regression technique. It helps to solve regression type problems because with the help of MARS we can predict the values of continuous dependent from a set of independent variable. Rough set theory Zdzislaw Pawlak proposed RS theory, with this theory we can approximate the lower and upper boundary of the original set. K-Nearest neighbor is also non parametric technique used in both regression and classification. KNN classify the input data into defined groups on the basis of prior data that is called training data. Support Vector Machine is works with labeled data means it is the example of supervised learning technique. SVM can perform efficiently with linear classification as well as non linear classification. Vapnik and Chervonenkis invented SVM in 1963. SVM performance is not good with large data set, when data set contain more noise. Classification and regression tree is one type of Decision tree algorithm, in which we use classification tree and regression tree technique. Decision Tree is the powerful tool used in ML. it is the graph based model classification tree and regression tree technique. It is the graph based model that represents possible consequences including all event outcomes. Tree consist of nodes always start with root node; mainly three types of node can be possible end node, chance node and decision nodes. Sometime in decision tree predictor values are continuous or infinite possible outcomes is called regression tree. Many authors used several of ensemble methods with DT to increase the performance of DT mainly are Boosted tree, Bagging and Random forest classifier. Bagging used in regression and statistical classification, It avoid the over fitting means reduces variance. Boosting helps in, to builds a series of trees is a sequential form in regression problem. Random Forest is just like Bagging only difference is that bagging takes 1 parameter that is no of tree with all possible features whereas RF takes 2 parameter one is no of tree and second is how many features participate to search to find optimal or best feature. Naïve Bayes is a one type of probabilistic classifier based on Bayes theorem. It is conditional probability model. In many situations in supervised learning with some type of probability model, NB classifier can train very optimally and efficiently.

A Taxonomy on Machine Learning Based Techniques

21

Classification by Clustering is the technique by which we can find homogeneous subgroup and it is the type unsupervised learning technique. When we identify clusters or group inside the classified group is called classification by clustering. CC model gives best result at compare with individual classification or clustering technique.

4 Analysis In order to further explore the relation between the heart disease prediction and techniques applied for prediction, efforts have been and same is mapped with the help of Table 1. It is clearly shown from Table 1 that Neural Network is the technique which is explored in maximum number of studies.

Table 1. Machine learning techniques used in different studies for heart disease identification Features

D E

L R

MA RS

√

√

G A

R S

SV M

D T

KN N

N N

K M

Fuz zy

DB S + NL P

CA RT

R F

√

√

N B

C C

√

√

√ √

√ √

ML P

SC RL

RB F

√

√

√

B T

A B

PS O

√

√

√

Referen ces [1] [3] [4]

√

[6] [7] [8] [9] [11] [13] [14] [15] [16] [19] [20]

√

√

√ √

√ √

√

√

√ √

√

√ √

√ √

√

√ √ √ √

√ √

√ √ √

Table 2. Performance of algorithm along with set of attributes used for heart disease prediction and diagnosis in different studies Reference No [1]

Data set name

[3]

SPECTF

UCI Repository

heart disease dataset [4] [6]

UCI datasets Cleveland Heart Disease Dataset

No of features used in experiment 13 09 44 19 10 4 13 13

Algorithm name NN

Algorithm accuracy (%) 83.0 85.0 SVM 75.0 78.0 81.0 85.0 MARS + NN 82.14 Fuzzy + NN 91.10 (continued)

22

A. K. Srivastava et al. Table 2. (continued)

Reference No [7] [8] [9] [11]

Data set name

[13]

Cleveland Heart Disease database

[14]

24

[15]

352 diagnosed heart disease dataset CAD data set

[16] [19]

Cleveland Data Set Heart Disease Dataset

76 14

[20]

MIMIC II database

32

Heart disease A.P i2b2 corpus UCI ML datasets Maria Nuova Hospital, Florence, Italy, (2001–2008)

No of features used in experiment 12 06 13 12

6

15

Algorithm name KNN + GA DBS + NLP NN + SVM NN SVM Fuzzy + GA CART RF NB DT CC GA

Algorithm accuracy (%) 100.00 91.5 87.5 77.8 80.3 69.9 81.8 83.3 96.5 99.2 88.3 92.3

NB + GA DT + GA CC + GA SVM + MLP BT + PSO RF + PSO AB + PSO RF

96.5 99.2 88.3 84.15 100% 90.37 88.89 82.1

In the current study we compared the accuracy measurement of different classifier KM, DT, LR, MARS, RS, KNN, GA, SVM, NN, DE, DBS, NLP, CART, RF, NB, CC, MLP, SCRL, RBF, BT, AB, PSO and combination of these. Table 1 show the various ML techniques used in various studies. Table 2 shows the accuracy of ML techniques in percentage along with associated no of features. On the basis of Table 2 we found, accuracy of algorithm is depends upon the number of features selected for heart disease analysis. When studies used the GA or PSO optimization techniques for feature selection, classifier worked efficiently. Figure 1 Represents the various algorithms along with the number of attributes used in several studies.

A Taxonomy on Machine Learning Based Techniques Table 3. Performance comparison of ML algorithm used in different studies Sr. no Algorithm name No of attributes Accuracy 01 NN 9 85.0 12 77.8 13 83.0 02 MARS + NN 13 82.14 03 Fuzzy + NN 13 91.10 04 SVM + NN 13 87.5 05 SVM 4 85.0 10 81.0 12 80.3 19 78.0 44 75.0 06 SVM + MLP 76 84.15 07 GA 24 92.3 08 NB 06 96.5 09 DT 06 99.2 10 CC 06 88.3 11 NB + GA 15 96.5 12 DT + GA 15 99.2 13 CC + GA 15 88.3 14 KNN + GA 12 100.0 15 RF 12 83.3 32 82.1 16 BT + PSO 14 100% 17 RF + PSO 14 90.37 18 AB + PSO 14 88.89 19 CART 12 81.8 20 DBS + NLP 06 91.5

A C C U R A C Y

ALGORITHM NAME

Fig. 1. Algorithm performance graph

23

24

A. K. Srivastava et al.

5 Conclusions Everyone knows if Heart is not functioning properly then how it can affects our whole body and. so Detection of heart diseases at early state is essential to prevent from other diseases affected from heart and save lives. In this study we are showing the comparatively analysis of various ML algorithm performances in the prediction and diagnosis of heart disease. Table 3 shows the comparatively study of all ML Techniques and on the basis of this table we analyze following conclusion. First Conclusion, NB, DT, CC along with GA has shown better performance as NB, DT and CC performed without GA. Second conclusion, BT, RF and AB along with PSO performed better as compared with performance of BT, RF, and AB without PSO. Third Conclusion, Accuracy of NN along with Fuzzy or SVM is optimal as compared to accuracy of NN algorithm alone. Summary of this study (Fig. 1) that classifier algorithm gives optimal result along with any feature optimization algorithm like GA or PSO.

References 1. Vivekanandan, T., Iyengar, N.C.S.N.: Optimal feature selection using a modified differential evolution algorithm and its effectiveness for prediction of heart disease. Comput. Biol. Med. 90, 125–136 (2017). ScienceDirect 2. Heart Disease Data set from UCI Repositories. http://archive.ics.uci.edu/ml/datasets/ HeartþDisease 3. Shilaskar, S., Ghatol, A.: Feature selection for medical diagnosis: evaluation for cardiovascular diseases. Exp. Syst. Appl. Sci. Direct 40(10), 4146–4153 (2014) 4. Shao, Y.E., Hou, C.D., Chiu, C.C.: Hybrid intelligent modeling schemes for heart disease classification. Appl. Soft Comput. 14(Part A), 47–52 (2014) 5. World Health Organization. http://www.who.int/mediacentre/factsheets/fs310.pdf. Accessed 03 Feb 2017 6. Samuel, O.W., Asogbon, G.M., Sangaiah, A.K., Fang, P., Li, G.: An integrated decision support system based on ANN and Fuzzy AHP for heart failure risk prediction. Int. J. Exp. Syst. Appl. 68, 163–172 (2017) 7. Jabbar, M.A., Deekshatulu, B.L, Chandra, P.: Classification of heart disease using K- nearest neighbor and genetic algorithm. In: International Conference on Computational Intelligence: Modeling Techniques and Applications, vol. 10, pp. 85 – 94 (2013) 8. Hui, Y., Jonathan, M.G.: A hybrid model for automatic identification of risk factors for heart disease. Int. J. Biomed. Inform. 58, S171–S182 (2015) 9. Olaniyi, E.O., Oyedotun, O.K., Khashman, A.: Heart diseases diagnosis using neural networks arbitration. Int. J. Intell. Syst. Appl. 12, 75–82 (2015) 10. Heart Disease Dataset from UCI Repositories. http://archive.ics.uci.edu/ml/datasets/ HeartþDisease 11. Guidi, G., Pettenati, M.C., Melillo, P., Iadanza, E.: A machine learning system to improve heart failure patient assistance. IEEE J. Biomed. Health Inform. 18, 1750–1756 (2014) 12. Anbarasi, M., Anupriya, E., Iyengar, NChSN: Enhanced Prediction of Heart Disease with Feature Subset Selection using Genetic Algorithm. Int. J. Eng. Sci. Technol. 2(10), 5370– 5376 (2010)

A Taxonomy on Machine Learning Based Techniques

25

13. Hongmei, Y., Jun, Z., Yingtao, J., Chenglin, P., Shouzhong, Xo: Selecting critical clinical features for heart diseases diagnosis with a real-coded genetic algorithm. Appl. Soft Comput. 8, 1105–1111 (2008). ScienceDirect 14. Soni, J., Ansari, U., Sharma, D., Soni, S.: Predictive data mining for medical diagnosis: an overview of heart disease prediction. Int. J. Comput. Appl. 17(8), 0975–8887 (2011) 15. Pouriyeh, S., Vahid, S., Sannino, G., De Pietro, G., Arabnia, H., Gutierrez, J.: A Comprehensive investigation and comparison of machine learning techniques in the domain of heart disease. In: 22nd IEEE Symposium on Computers and Communication (ISCC 2017), Workshops - ICTS4eHealth (2017) 16. Nahar, J., Imam, T., Tickle, K.S., Chen, Y.-P.P.: Computational intelligence for heart disease diagnosis: a medical knowledge driven approach. Exp. Syst. Appl. 40, 96–104 (2013) 17. http://archieve.ics.uci.edu/ml/datasets 18. Yekkala, I, Dixit, S., jabbar, M.A.: Prediction of heart disease using ensemble learning and particle swarm optimization. In: IEEE International Conference on Smart Technologies For Smart Nation (2017). Accession Number 17766551 19. Miao, F., Cai, Y.P., Zhang, Y.X., Fan, X.M., Li, Y.: Predictive modeling of hospital mortality for patients with heart failure by using an improved random survival forest. IEEE Access 6, 7244–7253 (2018). ISSN 2169-3536 20. Jin, B., Che, C., Liu, Z., Zhang, S., Yin, X., Wei, X.: Predicting the risk of heart failure with EHR sequential data modeling. IEEE Access 6, 9256–9261 (2018). ISSN 2169-3536 21. Pu, L.N., Zhao, Z., Zhang, Y.T.: Investigation on cardiovascular risk prediction using genetic information. IEEE Trans. Inf. Technol. Biomed. 16(5), 795–808 (2012) 22. Lafta, R., et al.: Coupling a fast fourier transformation with a machine learning ensemble model to support recommendations for heart disease patients in a tele health environment. IEEE Access. 5, 10674–10685 (2017). Accession Number 16950418 23. http://archive.ics.uci.edu/ml/datasets/Heart+Disease 24. Pecchia, L., Melillo, P., Bracale, M.: Remote health monitoring of heart failure with data mining via CART method on HRV features. IEEE Trans. Biomed. Eng. 58(3), 800–804 (2011) 25. Harvey, D.Y., Todd, M.D.: Automated feature design for numeric sequence classification by genetic programming. IEEE Trans. Evol. Comput. 19(4), 474–489 (2014) 26. http://dx.doi.org/10.1016/j.jbi.2015.09.006 27. Nahar, J., Imam, T., Tickle, K.S., Chen, Y.P.P.: Computational intelligence for heart disease diagnosis: a medical knowledge driven approach. Exp. Syst. Appl. 40(1), 96–104 (2013). ScienceDirect 28. Bunkley, N.: Joseph Juran, 103, pioneer in Quality Control, Dies. N. Y. Times (2008) 29. Water Resources Research Report, Optimization Using Differential Evolution, Report No: 060 Date: July 2008. [https://ir.lib.uwo.ca/cgi/viewcontent.cgi?article=1022&context=wrrr] 30. Blake, C.L., Mertz, C.J.: UCI Machine Learning Databases (2004). http://mlearn.ics.uci.edu/ databases/heartdisease/ 31. Beaglehole, R., Bonita, R.: Global public health: a scorecard. Lancet 372(9654), 1988–1996 (2008) 32. http://www.who.int/cardiovascular_diseases/world-heart-day-2017/en/ 33. https://www.datascience.com/blog/k-means-clustering 34. https://www.researchgate.net/publication/315800218_Rough_Sets_in_Machine_Learning_ A _Review [accessed Jul 27 2018]

Prediction and Analysis of Ecological Climatic Conditions (Humidity/Temperature) of Different Countries Fly-Ash Using Data Mining Techniques Divya Gupta(&) and Kanak Saxena Computer Applications Department, Samrat Ashok Technological Institute, Engineering College, Vidisha, MP, India [email protected]

Abstract. Basically coal is a deposited rock of biological origin which is produced by the diagenesis and metamorphism of flora at a certain compression and at defined temperature; the degree of alteration endured by coal as it develops from peat to anthracite, recognized as coalification. In recent past, there are several papers highlighted the concentration of heavy metals and metalloids in fly-ash, but there are still scanty of research which may highlight the climatic conditions especially temperature and humidity and particle size influence on metals in fly-ash. In compass of this issue, we focused our present manuscript on climatic variations especially humidity/temperature and different particle size of fly-ash and its influence on metal concentration. We used a formula for normalizing the values of generated fly-ash from coal and consequently calculated the metal concentrations in fly-ash using data mining statistical tools for two different natural variables like humidity and temperature against particle size. From our observations we noted that in all countries (India, Macedonia, Philippines and Spain) except Greece fly-ash, if we normalized the value (temperature/humidity) against particle size the lowest concentration of metals like Cr, Ni, Cu and Zn was at < 53 micron particle size and the highest metals like Pb and Zn was at 150–106 micron particle size. So, it is very clear from this normalized formula and values that particle size with normalized temperature/humidity plays role in metal concentrations in fly-ash. Keywords: Fly-ash concentration


Humidity
Temperature
Data mining tools
Metal

1 Introduction At present life is inconceivable without electricity because coal plays a dynamic role in electricity generation worldwide. Coal is the world’s most ample and extensively dispersed fossil fuel with reserves for all types of coal assessed to be about 990 billion tones [1]. At present coal fired power plants generates about 41% of the global electricity needs (World Coal Association, https://www.worldcoal.org/coal/uses-coal/coalelectricity, downloaded on 06.06.2017). In developed as well as in developing country © Springer Nature Singapore Pte Ltd. 2019 M. Prateek et al. (Eds.): NGCT 2018, CCIS 922, pp. 26–37, 2019. https://doi.org/10.1007/978-981-15-1718-1_3

Prediction and Analysis of Ecological Climatic Conditions

27

like India electricity generation till date is maximum based on coal based thermal power plant [2, 3]. Coal based thermal power stations are the main source of electricity generation in India in 21st century. According to Krishnan [4], fly-ash generation was 131 MT/Year in 2012–2013 after using different grades of coal and it is expected to increase by 300– 400 MT/Year by the end of 2016–2017. The toxic metals present in the ashes are initiated from the composition of the coal used in combustion, combustion conditions, and removal efficiency of air pollution control device and method of fly-ash disposal [2, 3]. Fly-ash generally consists of some radioactive (uranium, thorium, and their numerous decay products, including radium and radon) as well as highly toxic elements like arsenic, selenium and mercury. Other toxic metals which are found in fly-ash are V, B, Al, Cd, Pb, Ni and Cr [5, 6]. Fly-ash can be disposed of by two methods either wet or dry. The disposal of fly-ash in the environment can be a severe problem due to the leaching of toxic heavy metals in the ground water [7]. In fly-ash several factors controls the concentrations of essential and trace elements that comprise element sources, existence, combustion process, volatilization–condensation mechanism and last but not least particle size of the fly-ash [7, 8]. Chemical alignment of the coal combustion and stack released control device plays a significant role in chemical composition of fly-ash [9]. It was hypothesized that metals associated with essential minerals in natural coal have high volatility; and generally elements associated with unessential minerals have less volatility, Miller et al. [10]. It was also noticed that, trace elements and fly-ash particulate size are contrariwise related and the concentration of the trace elements at the fly-ash surfaces should be greater than that in the particle peripheral [11]. Now a day, data-mining procedures play a significant role in scientific fields like agriculture and environment science related research. In fact the performances prominence an importance of perceptive and using them to efficaciously accomplish newer findings in the fields, even though examining the unknown arrays is usually not a recent spectacle. Although in earlier days there are majority of explanations on data targeted to discover realistic laws that matters to different features of research Debeljak and Džeroski [12]. Data-mining is the method of pull out important and useful information’s from majority or large sets of data which in turn may be transformed into advantageous knowledge that aids to better thoughtful the subject and perceiving and eliminating the hitches related to its Debeljak and Džeroski [12]. In recent years there are several data mining algorithms has been established to grab a diversity of complications in the fields of environmental sciences, where data may be scant, incomplete, or heterogeneous. Several data mining methods are far more malleable than classical exhibiting tactics and might be practically applied to data-rich environmental complications Spate et al. [13]. Ecological modelling apprehends the data mining solicitations, with the aims for advancing the representations of associations between members of living populations and their abiotic atmosphere. In general modelling areas comprise population dynamics of cooperating species and territory suitability for an assumed class, demonstration comprise prognostic modeling, clustering and summarization. Most important feature of data mining procedures are termed as normalization, it plays a significant role in the management of data like scale down and scale up the range of data for its future dispensation. Normalization is scaling procedure or a

28

D. Gupta and K. Saxena

mapping method or a preprocessing platform, where any one might find new rang from are mining one. This tool is also very helpful for the extrapolation or forecasting purposes Patro and Sahu [14]. Contrary to our previous study [9], in the present research work we are analyzing the data with two different variables with initial concentration of chosen metals (Cr, Ni, Pb, Cd, Cu, Co and Zn) in fly-ash as well as with the normalized values (used formula) such as humidity and temperature of five different countries viz. India, Spain, Macedonia, Philippines and Greece at the same time with different particle size of fly-ash to see the differences in metal concentrations for the future analysis of biotic toxicity test on fly-ashes to predict the potential risk to the ecological systems.

2 Materials and Methods In this theoretical paper, we used five different countries (India, Philippines, Greece, Spain, and Macedonia) data of metals in fly-ash from Sijakova-Ivanova et al. [15]. Briefly, fly-ash from four different sites was taken Sijakova-Ivanova et al. [15] and after digestion of samples total metal concentration was measured by Inductively Coupled Plasma Atomic Emission Spectroscopy (AES-ICP) and later data mining tools were used for further analysis. For different particle size of fly-ash such as >150, 150–106, 106–75, 75–53 and 150 particles size i.e. 196 µg g−1 and the lowest at 106–75 particle size, i.e. 4.78 µg g−1. At average particle size Cr is highest at 40 °C and lowest at 5 °C temperature. In case of Pb, maximum Pb was noticed at 40 °C and at particle size 150– 106, i.e. 220.32 µg g−1 and lowest at 5 °C and particle size >150. At average particle size it again same like Cr. Cu also shows the same trend of increasing with temperature and humidity, but in case of Cu maximum Cu was noticed at average particle size rather than any other particle size i.e. 79.18 µg g−1 at 40 °C and the lowest at Cumulative energy per node RREP (B) So based on these three criteria and the route length, the Priority values are assigned, So Priority values based on C.E are as follows: Priority values based on C.E for RREP (A) = 2 Priority values based on C.E for RREP (B) = 1 Priority values based on C.E for RREP (C) = 3 Minimum Energy RREP (C) = Minimum Energy RREP (A) > Minimum Energy RREP (B) Route length of RREP (C) > RREP (A) So, priority values based on Minimum Energy are as follows: Priority values based on Minimum Energy for RREP (A) = 3 Priority values based on Minimum Energy for RREP (B) = 1 Priority values based on Minimum Energy for RREP (C) = 2 Maximum energy RREP (C) > Maximum energy RREP (B) > Maximum energy RREP (A)

328

S. Atri and S. Tyagi

So, Priority values based on Maximum Energy are as follows: Priority values based on Maximum Energy for RREP (A) = 1 Priority values based on Maximum Energy for RREP (B) = 2 Priority values based on Maximum Energy for RREP (C) = 3 Now Total Priority Value for each route request is calculated as follows (Table 4): For Total Priority value RREP (A) = (P.V. for C.E) + (P.V. for Max_E) + (P.V. for Min_E) = 2 + 1 + 3=6 For Total Priority value RREP (B) = (P.V. for C.E) + (P.V. for Max_E) + (P.V. for Min_E) = 1 + 2 + 1=4 For Total Priority value RREP (C) = (P.V. for C.E) + (P.V. for Max_E) + (P.V. for Min_E) = 3 + 3 + 2=8 Table 4. Priority values assigned to RREP (A), RREP (B) and RREP (C). RREP packets RREP(A) RREP(B) RREP(C)

P.V. (C.E.) P.V. (Min_E) P.V. (Max_E) Total priority value 2 3 1 6 1 1 2 4 3 2 3 8

Now Source Node S will select the path having highest Total Priority Value i.e. path C, having highest priority of 8 to send the data packet towards destination D. Next best path is path A having priority value of 6 and last comes the path B with priority value of 4.

5 Conclusion Multi-path routing is a reliable form of routing, which provides multiple paths between source node and destination node. Reliability ensures a dedicated path for consistent flow of data packet even in case of failure of one path. In this paper, priority value has been taken as a deciding factor for the selection of particular path for transmission of packets among several previously generated paths. Here RREP packets consist of three additional parameters namely, Cumulative Energy (C.E.), Minimum Energy (Min_E) and Maximum Energy (Max_E). Priority value is being assigned to each of the above parameters individually for every RREP received at the source node. Now total priority value is calculated by adding the priority value of all the three parameters and assigned to the RREP packets. Source node selects the path having highest value of total priority for sending the data packets. In case of route failure next best path can be utilized for further transmission of data packets.

Multi-path Priority Based Route Discovery Mechanism

329

References 1. Gulati, M.K., Kumar, K.: A review of QoS routing protocols in MANETs. In: International Conference on Computer Communication and Informatics (ICCCI), Coimbatore, pp. 1–6 (2013) 2. Asif, M., Khan, S., Ahmad, R., Sohail, M., Singh, D.: Quality of service of routing protocols in wireless sensor networks: a review. IEEE Access 5, 1846–1871 (2017) 3. Atri, S., Tyagi, S.: Simulating and analyzing the behavior of table-driven and on-demand routing protocol. Int. J. Comput. Sci. Eng. 6(2), 125–129 (2018) 4. Cidon, I., Rom, R., Shavitt, Y.: Analysis of multi-path routing. IEEE/ACM Trans. Network. 7, 885–896 (1999) 5. Xu, Y., Lee, W.-C., Xu, J., Mitchell, G.: PSGR: priority-based stateless geo-routing in wireless sensor networks. In: Mobile Adhoc and Sensor Systems Conference, Washington, DC, USA (2005) 6. Suthaputchakun, C., Sun, Z.: Priority based routing protocol with reliability enhancement in vehicular ad hoc network. In: International Conference on Communications and Information Technology (ICCIT), Hammamet, pp. 186–190 (2012) 7. Suthaputchakun, C., Sun, Z.: Priority based routing protocol in vehicular ad hoc network. In: 16th IEEE Symposium on Computers and Communications, Greece, pp. 723–728 (2011) 8. Naik, S.S., Bapat, A.U.: Message priority based routing protocol in MANETs. In: International Conference on Pervasive Computing (ICPC), Pune (2015) 9. Wei, D., Cao, H., Liu, Z.: Trust-based ad hoc on-demand multipath distance vector routing in MANETs. In: 16th International Symposium on Communications and Information Technologies (ISCIT), Qingdao, pp. 210–215 (2016) 10. Jayabarathan, J.K., Sivanantharaja, A., Robinson, S.: Quality of service enhancement in MANET using priority aware mechanism in AOMDV protocol. In: IEEE UP Section Conference on Electrical Computer and Electronics (UPCON), Allahabad (2015)

Secure and Trustworthy Cloud: Need of Digital India, an e-Governance Project Archana B. Saxena1(&) and Meenu Dave2 1

Jagan Institute of Management Studies (JIMS), Sector-5, Rohini, Delhi, India [email protected], [email protected] 2 JaganNath University, Chaksu, Jaipur, Rajasthan, India [email protected]

Abstract. Role of cloud computing in the execution of e-governance services has led to an ever-growing need for secure and trustworthy cloud services. Cloud Computing brings IT services in form of utilities that can be consumed as per demand. Cloud computing uses the internet to deliver differential services through geographically apart data centers. These data centers offer pooled infrastructure resources that can be utilized on pay as per use basis. All these properties of the cloud make it a perfect solution for E-Governance. The only concern that is making current users vigilant and future users dubious about this service is “Security”. Due to augmented security lapse incidences in recent years, consumers are apprehensive about adopting it and continuously losing trust in this computing paradigm. The prime concern of this research to find ways to overcome these challenges. The key intent of this research is to find a framework that can compute the trustworthiness of cloud provider based on security coverage. The framework evaluates trust value for a provider, on the basis of standards & certification attainment related to security components required for the services offered by him. Keywords: E-Governance
Digital India
Security Trustworthiness
OTF
Overall Trust Factor




Trust




Framework




1 Introduction E-Governance, a government initiative to deliver government services to citizens through ICT (Information and Communication Technologies). “Digital India” is an extended and transformed version of E-Governance project launched by the government of India with an intention to transform India into a digitally empowered society and knowledge economy [1]. This program worth more than lakh crore and kept three major targets: digital Infrastructure for every citizen as a utility, takes digital literacy to the next level and last but not least seamless integration across departments/jurisdiction and access of all government services to citizens by using information and communication technology [2, 20]. In order to ensure that this initiative is successfully implemented, the fundamental requirements must be put in place first. These include a strong and robust Information Technology infrastructure, provision for storing large © Springer Nature Singapore Pte Ltd. 2019 M. Prateek et al. (Eds.): NGCT 2018, CCIS 922, pp. 330–339, 2019. https://doi.org/10.1007/978-981-15-1718-1_28

Secure and Trustworthy Cloud: Need of Digital India, an e-Governance Project

331

amount of data which would be generated from the various operations and services of this initiative and above all it requires sustainable training, education, and awareness program to ensure that rural and urban citizens are able to adopt the benefits of being Digitally empowered efficiently and effectively. Cloud computing, a relatively new technology, provides requisite infrastructure along with the added benefits of cost savings to make Digital India, a dream come true. For, Cloud computing, operates on the backbone of the Internet and thereby links several Data Centers of the government departments and integrates them by network resource centers. In essence, Cloud computing has three major components: 1. Infrastructure as a service (IAAS), 2. Platform as a Service (PAAS) and 3. Software as a Service (SAAS) [3, 21]. The word service denotes the concept of pay per use structure. In other words, the customer pays to the cloud service provider the amount of money for the time which he uses. He is not required to incur extra cost for hardware-software or on the maintenance or any other cost such as licensing and procurement. According to E-tech India summit, (2015) the incorporation of appropriate encryption technology, Cloud Computing will be the next step in integrating the country for Digitalization [4]. Necessity and contribution of Cloud Computing in the success of Digital India scheme are well registered but security lapse and data leakage form cloud is raising few questions that must be answered before commonalities will adopt this scheme without any apprehension. Worth mentioning a recent data security lapse, where an IIT graduated student is arrested for unauthorized access of “Aadhar card database”, one of the important components of “Digital India” scheme [5]. With such incidents it quite obvious to have reluctances among service consumers towards “Digital India” schemes and services. Various queries related to security, trust, privacy, and data protection goes around in customer’s cognizance and resulted as the main hindrance to the adoption of cloud computing. The business market for cloud computing has many players that offer cloud services. some big giants like Google, Amazon, ICloud, Microsoft and some small players like Joyent, CenturyLink, salesForce, IBM, Rackspace, SpiderOak and many more. All of them are providing services using different parameters and conditions. which cloud provider is trustworthy? Is it offering the required security essentials defined by law or requirements of the services? These queries must be answered before selecting any provider for keeping confidential data related to any service of Digital India or individuals private data. Keeping the relevance of this technology in the success of Digital India projects and intention to solve the apprehensions of cloud consumers, the main focus of this paper is to build trust model that can compute trustworthiness of provider on the basis of security coverage provided by him through standards and certifications attainment. Paper is organized as follows: Sect. 2 discusses literature and define problem statement, Sect. 3 lists Research Methodology used in the development of Model, Sect. 4 explains proposed model for evaluation of Trust value, Sect. 5 lists conclusion and future scope. List of resources used in the development of this paper is listed in References.

332

A. B. Saxena and M. Dave

2 Literature Review and Problem Statement The role of cloud services in the development of Digital India is found not in the literature [6] but in real time media as well. Mobile applications and cloud services as a roadmap to achieve objectives of digital India [6]. 15 points vision of Narendra Modi which includes e-banking, design, e-commerce and many more can be materialized only with the support of Cloud Computing. Various advantages of cloud computing can serve various fields as per requirements through sharable resources and minimum finances [7]. Discussion of cloud challenges Like Privacy, Security, Trust from various aspects can be found in the literature [3, 8]. Not only challenges discussion but solutions of these challenge in the form of trust models can be found in the literature. Variety of trust models can be found in the literature that is based on differential parameters Quality of Service [9], SLA [10] and many more. It has been observed during the review that security is freely related to trust. The same notion is supported by various trust models that are based on security like [11–13]. Although a lot of literature is found related to security and model still author can observe gaps in the literature: – IT industry is based on Standards and certifications, still, there is no trust framework that can compute trust on the basis of these. – Security is a term that covers various components, no trust framework is evaluating on the basis of differential components covered under this head. – There are various independent bodies (like CSCC, CSIG and more) that regulate Cloud computing and Issues guidelines and standards to be implemented to have secure transactions through technology. No framework is based on these guidelines. – What should be the relevance (% value) of different factors in trust computation that depends on security provisions?

3 Methodology In order to achieve the mentioned objectives, initially, authors have defined the scope of the study, by selecting a cloud service that can be a support for “Digital India” operations. So the scope of our study is related to IAAS service of a cloud that is in use through the public deployed model. Once domain pertaining to the study is defined then the author have followed a structured approach for the development of this paper (Fig. 1):

Secure and Trustworthy Cloud: Need of Digital India, an e-Governance Project

333

Security

Diverse

Governance

Trust Components

Audit

SLA

Fig. 1. Components having an impact on Trust

3.1

Component Selection

The selection of the components for the evaluation of trust is based on the following basics: • IAAS security challenges for Public cloud [12, 14, 15]. • Authors own previous work related to security issues in IAAS [16]. • Guidelines issued by private bodies like CSIG, CSCC [17–19] (Fig. 2).

Security Data Security

Governance

SLA

Audit

Diverse

IT Governance

Performance Level Metrics

Periodical Audits

Technology Stack

Privacy Policy

Risk Compliance

Security Metrics

Security and Management Logs

Capacity

Authntication & Authorization

Exit Process

Data Management Metrics

CASB Implementations

Service and Helpdesk

Cloud Network and Internet Security

Personnel Data Protection Metrics

Availability of Audit data to Consumer

Cost and Benefit Equation

Cloud Application Security

Responsibilities and Penalities

External Auditing

Physical Infrastructure Security

Customers Feedback

Region and Domain Ceoncerns

Fig. 2. List of components and sub their elements that have an impact on Trust.

334

3.2

A. B. Saxena and M. Dave

The Relevance of Components in Trust Model

To find the relevance of various components of the trust model, a questionnaire is prepared and distributed online among free and pain cloud consumers. The objective of the questionnaire’s framed questions is to capture the respondent’s opinion about the trust and factors that can have an impact on trust. To acquire the consumer’s insight, responses are measured on a five-point scale. The Likert Scale is used to record responses on given questions. The Likert scale technique uses a scale that is assigned to series of statements Very Important (5), Important (4), Moderately Important (3), Slightly Important (2), Not Important (1). The idea is to select the Likert scale is to get the respondents opinion about the components and sub-elements of these components, that has an impact on trust. In order to get the meaningful analysis, received responses are filtered, edited and transformed as per SPSS (Statistical Package for Social Study) requirements (Fig. 3).

Relevance of Security Components for Various Cloud Comuting Models

3.4414 4.0667 3.8000 2.9000

Community 3.4276 3.9111 3.7200 2.7000

Hybrid

3.7379 3.6667 3.9600 2.8000

Security

Private

3.6552 4.0000 4.0000 3.0000

2.9167

4.1092 4.3519 4.2000

Public

Governance

SLA

Audit

Diverse

Fig. 3. Relevance of various components of trust, for different cloud computing models

335

3.5818 3.9600 3.2923

3.5000

3.7727 3.6200 3.0308

4.0000

3.7636 4.2400 3.2000

4.5000

3.8182 4.1667 3.2564

5.0000

4.1667 4.3667 3.8333

Secure and Trustworthy Cloud: Need of Digital India, an e-Governance Project

3.0000 2.5000

Large (>500) Mean

2.0000

Medium (100-500) Mean

1.5000

small (1-100) Mean

1.0000 0.5000 0.0000

Fig. 4. Relevance of various components in trust on the basis of company size: large, medium and small.

4.2 4.1 4 3.9 3.8 3.7 3.6 3.5 3.4 3.3 3.2

4.114

3.7337

3.7066 3.5244

3.5822 Series1

Fig. 5. Mean average: importance of various components as per collected data sets

From above Fig. 4 it can be analyzed the most of the respondents have given the maximum weight age to the security which comprises following components: (data security, privacy policy, authentication and authorization, network and internet security,

336

A. B. Saxena and M. Dave

cloud application security and security of physical infrastructure) in trust formation for cloud services. Respondents have given almost similar weight age to the next two components Governance (Mean: 3.7333) and SLA (3.7067). Governance has given only .266 more preference than SLA, which is because SLA is treated as contractual document while governance has more impact on trust because it includes subcomponents like RISK Policy, Risk Compliance, and Exit Process. Among the remaining two components Audit (Mean: 3.5244) and Diverse (Mean: 3.5822). Although the difference is very minute only .578 respondents have given more concern to diverse which includes subcomponents like Technology Stack, Capacity, support and help desk, cost-benefit equation and current customers feedback. This is because cloud consumers have given more importance to feedback, capacity, and technology stack as compare to internal, external audits and log details (Fig. 5). Proposed Trust Model: The analysis completed through SPSS on the online collected data, is considered as the base to assign a percentage to the various components selected for trust evaluation. These components will compute OTF (Overall Trust Value) for the CSP (Cloud Service Provider) (Fig. 6).

Security 32%

Audit

Governance

15%

20%

OTF 100%

Diverse

SLA

14%

19%

Fig. 6. Relevance of various components in Trust

Computation of trust value is done by on the basis of availability of standards and certifications related to the sub-elements of these components.

Secure and Trustworthy Cloud: Need of Digital India, an e-Governance Project

337

4 Execution of Proposed Model The sequence of activities in the trust model can be understood with the help of Fig. 7. Both the CSP (Cloud Service Provider) and CSC (Cloud Service Consumer) has to register with the Cloud repository with requested details. Compute Engine is responsible to compute the trust value for CSP on the basis of details entered by him. If CSC is already registered that is can use login details to view the list of CSP that are offering cloud services he is interested in along with trust value computed by the engine.

CSP(Login) Enter Details

Cloud Repository

Compute Engine

Contains details entered by CSP like: standards, Certifications

NO

CSS (Login)

IF (Reg)

yes

View OTF Values

Fig. 7. Execution of trust model

5 Conclusion and Future Scope The key focus of this work contributes to the execution of “Digital India” notion by offering a technical solution to address the Security & Trust challenges faced by Cloud Computing. Cloud computing is viewed as a perfect solution to store and process huge data collected by “Digital India” services. All the promising aspects of this technology becomes null if it can provide you with the required security of your data and information. This paper provides a comprehensive information about the list of components that has an impact on trust. The information about the elements that comprise the component in whole is also listed. a secondary survey like literature review is used to list the components. in order to define their relevance percentage in Trust model, Primary survey method: Questionnaire is used to collect data from free and paid cloud consumers. SPSS and Excel tools are used for analysis of online collected data. As per

338

A. B. Saxena and M. Dave

results components are assigned various percentage in computation and OTF is computed on the basis availability of standards and certifications related to the sub-elements of components. In future authors will try to convert this model into an application that can be practiced for the computation of trust value. The practical implementation of this model can be useful for our “Digital India” notion, where administrators can check the trust value of provider before enrolling the one for delivery of cloud services that will be used for storing and processing of data generated by digital India services.

References 1. Vats, D., Kumar, B.: Digital India: an initiative to transform India into a digitally empowered society. Int. J. Sci. Technol. Manag. 543–547 (2017) 2. Lakshmi, A., Vijayalakshmi, S.M., Ugrappa, E.: Evaluation of digital India programme. KHOJ: J. Indian Manag. Res. Pract. 277–282 (2016) 3. Armbrust, M., et al.: A view of cloud computing. Commun. ACM 53, 50–58 (2010) 4. Cloud computing crucial to Digital India, need safe practices’, say, experts. https:// indianexpress.com/article/technology/tech-news-technology/cloud-computing-crucial-to-dig ital-india-need-safe-practices-4463667/. Accessed 11 Sept 2018 5. IIT Kharagpur graduate hacked Aadhaar data through Digital India app: Police. https:// indianexpress.com/article/india/iit-grad-hacked-aadhaar-data-through-digital-india-app-cops4781447/. Accessed 20 Aug 2018 6. Yadav, J.: Digital India: a roadmap for the development of rural India. Int. J. Commer. Manag. Res. 2, 77–80 (2015) 7. Kulkarni, V.: Contribution of Cloud towards Digital India (2015). https://www.esds.co.in/ blog/contribution-of-cloud-towards-digital-india/#sthash.9Buk1zpZ.dpbs. Accessed 4 July 2018 8. Lee-Post, A., Pakath, R.: Cloud computing: a comprehensive introduction. In: Security, Trust and Regulatory Aspects of Cloud Computing in Business Environment, pp. 1–23. IGI Global, Texas (2014) 9. Manuel, P.: A trust model of cloud computing based on quality of services. Ann. Oper. Res. 233, 281–292 (2013) 10. Alhamad, M., Dillon, T., Chang, E.: SLA-based trust model for cloud computing. In International Conference on Network-Based Information System, Takayama (2010) 11. Li, W., Ping, L.: Trust model to enhance security and interoperability of cloud environment. In: First International Conference Cloud Com-2009, China, Beijing (2009) 12. Almulla, S., Yeun, C.Y.: Cloud computing security management. In: 2nd Engineering Systems Management and Its Applications (ICESMA), Sharjah, United Arab Emirates (2010) 13. Takabi, H., Joshi, J.B.D., Ahn, G.J.: Secure cloud: towards a comprehensive security framework for cloud computing environments. In: 34th Annual IEEE Computer Software and Applications Conference Workshops, Seoul, South Korea (2010) 14. Jadeja, Y., Modi, K.: Cloud computing - concepts, architecture and challenges. In: ICCEETInternational Conference on Computing, Electronics and Electrical Technologies, Kumaracoil, India (2012) 15. Robinson, N., et al.: The Cloud: Understanding the Security, Privacy and Trust Challenges, Cambridge, United Kingdom (2010)

Secure and Trustworthy Cloud: Need of Digital India, an e-Governance Project

339

16. Dawe, D.M., Saxena, B.A.: Loss of trust at IAAS: causing factor and mitigation techniques. In: International Conference on Computing and Communication Technologies for Smart Nation (IC3TSN 2017), Gurgaon, Haryana (2017) 17. Baudoin, C., Devlin, R.: Cloud Security Standards: What to Expect and what to Negotiate Version 2.0. Cloud Standards Customers Council (2016) 18. Baudoin, C., Cohen, E., Dotson, C., Gershater, J., Harris, D., Iyer, S.: Security for Cloud Computing Ten Steps to Ensure Success Version 3. Cloud Standards Customers Council (2017) 19. Cloud Service Level Agreement Standardisation Guidelines. CSIG Members, Brussels (2014) 20. Vision for Digital India. https://www.narendramodi.in/shri-narendra-modi-shares-his-visionfor-digital-india-5944. Accessed 8 Sept 2018 21. Qian, L., Luo, Z., Du, Y., Guo, L.: Cloud computing: an overview. In: Conference on Cloud Computing, Beijing, China (2009)

Hyperheuristic Framework with Evolutionary and Deterministic Algorithms for Virtual Machine Placement Problem Amol C. Adamuthe(&) and Akshayya Jadhav Rajarambapu Institute of Technology, Rajaramnagar, MS, India [email protected]

Abstract. Virtual machine placement in cloud computing requires to handle issues like energy efficiency, traffic optimization, load balancing, resource management, etc. VMP problem is constrained satisfaction problem belongs to category of NP problems. Hyperheuristic provides more general framework for range of problems and offer optimal solutions. In this paper, we proposed hyperheuristic framework for VMP problem with evolutionary algorithms and deterministic algorithms. Tabu search technique and Warm-up techniques are compared as higher level heuristics. Low level heuristics tested are first fit, best fit, Intelligent Water Drop and Simulated Annealing. Results of proposed hyperheuristic framework are compared with individual evolutionary algorithms for twelve instances. Results shows that hyperheuristic works better for all instances. Keywords: Virtual machine placement Tabu search


Hyperheuristic
IWD
SA


1 Introduction Cloud computing introduces a computing platform for users for on demand resource request using Internet. Today it is buzzword in the computing industry. Therefore, there is rapid growth in number of cloud service providers. Today lots of companies are adopting IaaS model of cloud computing. Increase in data centers needs more supply of energy. Increase in heat dissipation decrease computational efficiency and increases operating costs. It causes high burden of environmental and energy resources. Optimization in the number of servers in data centers has a significant benefit in reduction of the complexity of the infrastructure, improvement in availability of system, saving money and energy. Virtualization is backbone technology for cloud computing. Virtual Machine Monitor (VMM) or Hypervisor manages virtual machine related operations. Virtual machine placement problem (VMP) is allocation of virtual machines to physical machines to achieve specific objective with satisfying given constraints [1–5]. Minimizing the required physical machines is a main goal of this research. In literature, VMP studied with varying interest. Different authors used different technique to solve this problem. Different types of techniques such as deterministic, © Springer Nature Singapore Pte Ltd. 2019 M. Prateek et al. (Eds.): NGCT 2018, CCIS 922, pp. 340–350, 2019. https://doi.org/10.1007/978-981-15-1718-1_29

Hyperheuristic Framework with Evolutionary and Deterministic Algorithms

341

heuristic, meta-heuristic, approximation techniques experimented to solve this problem [6]. Heuristic optimization techniques provides reasonably good solution to combinatorial optimization problems. The main limitation of these techniques is that there performance varies with problem. Problem solving technique need more general framework which can apply to different problems of similar in nature. This is main inspiration behind hyperheuristic. Hyperheuristic has search space in heuristics rather than in solution. This framework designed for more generalized methodology. The main learning mechanism in hyperheuristic is to choose best heuristic at that decision point which highly affects hyperheuristic framework performance. Hyperheuristic framework consists of lower level, domain level and hyper level. Lower level contains set of heuristic algorithm and problem representation. Domain interface transfer problem-independent information to hyper level. High level operates on a set of heuristic algorithm and problem representation. Multidimensional bin packing problem is similar to virtual machine placement (VMP) where each dimension is for one of the resource type or attribute of a VM. This problem is NP in nature. In literature, many papers presented this problem as combinatorial optimization problem [7–9]. First fit and best fit are basic algorithms for solving bin packing problem. Objective of this paper is to allocate all virtual machines to physical machines without violating constraints by using traditional algorithms like first fit, best fit, meta-heuristics like simulated annealing, Intelligent water drop and hyperheuristic framework with tabu search and warm-up techniques. Section 2 is about related work. In Sect. 3, VMP formulation is discussed. Descriptions of lower level heuristics, hyperheuristic methodologies are in Sect. 4. Experimental results of hyperheuristic framework discussed in Sect. 5. Section 6 is conclusion of work.

2 Related Work Authors reported that many researchers contributed to solve virtual machine problem with different techniques [6]. In paper [10] authors reviewed virtual machine placement problem and presented comparative analysis of algorithmic techniques to solve the problem. Authors discussed strengths and challenges of these algorithms. Paper [11] presented review by considering different objectives, problem formulations and algorithmic techniques used for virtual machine placement problem. Paper [12] provides differential evolution algorithm for solving VMP where the objective is to minimize set of physical machines required to place all virtual machines. In [13], Jayshri Damodar Pagare et al. proposed an energy-efficient algorithm for VM consolidation it can be reduce or minimize energy consumption. In [14], Choudhary et al. proposed solution to VMP cloud data centers considering energy efficiency and quality of service. In [15], Hongjian Liet et al. developed virtual machine migration considering energy efficiency model. In [16], Sookhtsaraei et al. presented results of Genetic algorithm for virtual machine placement problem formulated as multi objective optimization. In [17], B. Benita Jacinth Suseela et al. proposed hybrid model of multi objective VMP problem. They proposed hybrid ACO-PSO algorithm. In [18], authors proposed bi objective simulated annealing to maximize profit and minimize power

342

A. C. Adamuthe and A. Jadhav

consumption. Paper [19] presented hybrid multiobjetcive algorithm for VM placement problem using genetic algorithm and fuzzy logic. Heuristic developments in literature provided good solution but same technique can’t apply to problem within same class. Heuristic technique needs extensive tuning for it. Whereas hyperheuristic tries to use same method to class of problems with large complexity. Chakhlevitch et al. introduces learning strategies to choose the fittest heuristic from the low level set. They tested two approaches- Warming up where during some warm up period, specific best fitted low level heuristics identified and then either continue execution (WU-C) till the stopping criteria or restart execution (WU-R) with identified heuristics. Step-by-step reduction where set of heuristics reduced till fittest lower heuristics remain in a set of lower heuristics [20]. Hyperheuristic is term introduced in Cowling et al. [21]. Previously same idea introduced by Fisher and Thompson [22, 23] and Crowston et al. [24]. All these paper handled job shop scheduling problem by probabilistic weighting to choose low level heuristics. Mockus et al. used Bayesian mechanism to select lower heuristics in same problem domain [25– 27]. Hart and Ross, solved job shop scheduling problem but by using Genetic algorithm where GA chromosomes represent method to solve problem [28]. Cowling et al. represent hyper-GA technique for personal scheduling problem where low level heuristics [29]. Low level heuristics are ranked based on their past performance. These ranks are input to low level heuristic selection function for current generation. Paper [30] reported improved performance of hyper-GA with adaptive length solution representation. Nareyek represents a weight adoption method to choose attractive lower heuristics [31]. Paper [32] used tabu search at a hyper level which selects lower level heuristics. Paper [33] and [34] provides good literature review of hyperheuristic methods.

3 Virtual Machine Placement Problem Virtual machine placement problem is most severe problem in cloud computing area. This paper worked on minimizing the required physical machines for allocating virtual machines with satisfying given constraints. The minimization of essential physical machines through maximize resource utilization which leads to minimization of electricity requirement and cost. The virtual machine considered as node which is attributed by CPU, memory and bandwidth utilization. The number of physical machines, known to be available and potential resources with the dimensions, such as CPU, memory and network bandwidth. The total number of virtual machines to host is known. The aim of the study is to find the correspondence between the virtual machines and PMs, meet the requirements for the virtual machine resources, in the course of the meeting the following objective. The problem is then formulated as follows. The problem formulation is similar to defined in [35], Min CðSÞ ¼

XX X¼1

A(x)

ð1Þ

Hyperheuristic Framework with Evolutionary and Deterministic Algorithms

Subject to;

XY y¼1

343

Bðy; YÞ; SðyÞ  AðxÞ; DðxÞ

For each dimension of a given PMj, all the available capacity will depend on the sum of the resources required by the VM placed. All virtual machines should be allocated (Table 1). Table 1. Symbols of objective function. Symbol C(S) X Y A(x) B(y, x) D(x) S(y)

Description Cost function s for active servers Total number of PMs Total number of VMs Binary variable, 1 if x PM is active otherwise 0 Binary variable, 1 if y VM is allocated on x PM Dimension vector of x PM Attribute vector of y VM

4 Proposed Hyperheuristic Method for VMP When problem with large dataset occurs, heuristics approach failed after some extent. It works good upto some level but at extreme point its performance decreases as problem complexity increases. Heuristic approaches unable to change their working behavior as complexity increases. Comparatively, hyperheuristic framework works effectively with complex problems and able to provide good optimal solution. 4.1

Low Level Heuristics

This paper presents use to two kinds of low level heuristics in hyperheuristic framework. Traditional Heuristic Algorithms. First fit algorithm allocates the first available physical machine satisfying the constraints. Best fit algorithm allocates the available smallest physical machine which is big enough for allocating virtual machine with satisfying all constraints. Metaheuristic Algorithms. Simulated Annealing Algorithm: Annealing is the process of cooling material in a heat bath. Due to heating of material, states of material changes frequently. Next step is cooling material gradually where states changes are rare. Simulated annealing (SA) emulates this physical process whereby the material is slowly cooled until a steady state is reached. Simulated annealing can be understood as an extension of the simple random gradient descent algorithm. In [36], author suggest that SA will be helpful to search for solutions in an optimization problem objective is to converge to an optimum state.

344

A. C. Adamuthe and A. Jadhav Algorithm: Simulated Annealing Algorithm Inputs : Problem Representation Output: solution 1. Randomly place VMs in PM 2. Set start_temp, end_temp 3. If (temp < (start_temp + end_temp)/2)) Select two VMs and swap out their locations Update temp 4. Else Move randomly selected VM to randomly selected PM Update temp 5. Do step 4 & 5 till temp < end_temp

Intelligent Water Drop Algorithm: IWD introduced by Shah-Hosseini in [37]. Inspiration for the development of the IWD is water drops flow into lakes, rivers and seas. This intelligence is more similar to river based scenario in which river find its path among many obstacles. The gravitational force provide tendency to water drop to flow towards destination. If there is no obstacle between path water drop directly flow to destination with shortest path. Due to obstacles in path, this traversing road seems to be optimal in terms of distance from last node that is destination and the limitations of the environmental conditions. When a drop of water is moving from a one point to the next point of front in the river, it is assumed that every water drop carry some amount of soil with it during its motion. The soil with water drop increases as passes long distance. After some distance water drop reload its soil to the riverbed. Also, some soil from riverbed also carried out by water drop. This is IWD’s property to carry and remove the soil during movement in the area. A drop of water has attribute like velocity, which plays important role in removal of soil from river bed. To remove soil from land value of velocity affects a lot. When water drop is faster means velocity is grater and soil with drop is less. 4.2

Hyper Heuristic Techniques

Tabu Search Hyper Heuristic: Hyperheuristic is seems as black box technique required only input and evolution function. Here input is set of low level techniques and evolution function is a tabu heuristic. Evolution function applied on Low level heuristics and solution is presented as output. At each decision point, evolution function determines next low level heuristic to choose. When stop condition appear output is presented. Tabu list is used to prevent selection of low ranked heuristic based on historical information. Hyperheuristic evolution function works as, initially it randomly select heuristic from low level set. First run set as best and further that if chosen heuristic which improve objective function value then its rank increase otherwise decreases. Low level

Hyperheuristic Framework with Evolutionary and Deterministic Algorithms

345

heuristics compete each other for better result. Reinforcement learning principles used for these competition rules. If stopping condition met, this process stops. Warm-Up Technique: In this technique, Hyperheuristic determine some good heuristics within some warm-up time. Set of low level heuristic is executed at warm up period. After that, we get some best heuristics list. We can continue execution till end called WU-C or we can restart execution only with this list of heuristic starting with initial solution (WU-R). Warm-up hyperheuristic framework is based on priority selection rule. Priority is assigned by number of physical machine used in warm up period. The other selection methods also present but often used.

5 Experimental Details, Results and Discussion Our hyperheuristic framework was coded in CodeBlock and all experiments were run on a Intel Pentium CPU B960 2.20 GHz with 2 GB installed RAM running under 32bit Microsoft Windows 7 ultimate. Dataset contain four attributes such as index number of physical machine or virtual ma-chine, CPU, memory and bandwidth of physical machine and virtual machine which are generated randomly. Both the problem specific information and set of low level heuristics provided with input to hyperheuristic framework. Hyperheuristic provides output in form of solution. Tabu search hyperheuristic technique and warm up technique are two different methodologies. Tabu search is heuristic technique which is local search methodology whereas warm-up technique is static technique. Both frameworks start execution with randomly generated feasible solution. For each dataset, program is executed 10 times and best result is presented. In tabu search hyperheuristic, initially all heuristics in low level has 0 rank. Rank is maintained during the range of [Rmin, Rmax] with lower and upper bond respectively. The variation between previous and new solution is measured by d variable. If (d > 0), rank of used heuristic is increased otherwise decreased by a (can be integer or real number) [20]. We choose a = 1 and Rmin = 0, Rmax = l, where l in number of low level heuristics. In Warm-up continuous approach, ratio for fittest and rejected heuristics is approximately set to 3:2. For fitness criteria, we use total improvement for low level heuristic. When a reduction occurs, the low level heuristics with the smallest total improvement over all previous iterations are discarded. For warm-up period we used 20% of total iterations. For experimentation, we use randomly generated dataset with 2000 VM instances. The performance of individual low level heuristics and hyperheuristics are evaluated. Initial solution generated by random approach. Metaheuristic algorithms result affected by number of iterations. Traditional algorithms not shown iteration wise improvement. They have fix execution structure at any iteration. Hence, first fit resulted with output value 568, best fit with 544 and worst fit with 675. Whereas metaheuristics, population based algorithm has iteration wise improvement.

346

A. C. Adamuthe and A. Jadhav Table 2. Iteration wise improvement in meta-heuristics and hyper-heuristics. No. of iterations SA IWD IS 850 850 100 589 578 500 582 563 1000 573 558 2000 568 542 5000 564 533 8000 551 528 10000 543 522 12000 540 518 15000 539 512 18000 537 509 20000 534 507 22000 534 505 25000 534 503

WU-C 850 522 517 511 503 492 491 489 487 482 478 473 473 472

Tabu HH 850 527 518 496 483 478 473 470 468 464 461 459 458 458

900

SA

800

IWD

700

WU-C HH Tabu HH2

600 500 400 300 200 100 0

Fig. 1. Heuristic algorithm performance (with same problem size).

Table 2 demonstrate comparison between individual meta-heuristics and hyperheuristic frameworks and best results obtained after number of iterations. Tabu search hyperheuristic gives best result compare to another WU-C framework and individual low level heuristics. In Table 3 and Fig. 1, it is observed that meta-heuristic techniques comparable with hyper heuristic. But after some iteration, they failed to improve solution whereas hyper heuristic framework able to provide more optimal solution.

Hyperheuristic Framework with Evolutionary and Deterministic Algorithms

347

Table 3. Compared result with hyperheuristic framework and individual heuristics. No. of VM SA IWD WU-C HH Tabu HH 20 3 3 3 3 40 8 8 8 8 60 11 11 11 11 80 14 13 13 13 100 22 22 22 21 150 30 31 29 29 200 40 40 40 38 250 48 47 48 43 500 195 158 137 107 1000 310 299 252 214 1500 396 389 367 334 2000 534 503 472 458 Note: Bold value denotes optimal solution 800

FF

700

BF WF

600

SA IWD

500

WU-C HH

400

Tabu HH

300 200 100 0 20

40

60

80

100

150

200

250

500 1000 1500 2000

Fig. 2. Heuristic algorithm performance (with different problem size 20 VMs to 2000 VMs).

We compared hyperheuristic results of tabu search framework and WU-C framework with individual low level heuristics. Every heuristic and framework starts with the initial solution obtained by random selection. Each algorithm algorithms has executed 10 times and best results given. Reduced sets of low level heuristics obtained by warm up period is same for different datasets. The stopping condition is 25000 iterations. Heuristic algorithm’s individual execution gives poor results compared to hyperheuristic framework. Above tabular experimentation clearly notify by following graph. We can observe heuristic and hyper heuristic framework behavior with small and large size problems. From above experimentation (Fig. 2), we can observe that a heuristic technique works well with relatively low complex problem. When problem size increases their performance reduces gradually. WF has always low performance. In our experiments,

348

A. C. Adamuthe and A. Jadhav

FF successfully compete with BF upto VM size 200. But after that limit of VM size FF fails to compete with BF algorithm and metaheuristics algorithms works better. Scientist required some general framework which can operate on same class problems with large dataset without any major change in code. Hyperheuristic works on same approach. We can observe above results where hyperheuristic works most effectively on large dataset whereas small dataset can be operated by individual heuristics also.

6 Conclusions This paper presents tabu search and WU-C technique based hyperheuristic framework for virtual machine placement problem in cloud computing. The low level heuristic used in this framework are first fit, best fit, simulated annealing and intelligent water drop algorithm. Results shows that hyperheuristic provides better results than individual heuristic. It is observed tabu hyperheuristic is better than WU-C hyperheuristic. Comparatively, hyperheuristic framework works effectively with complex problems and able to provide good optimal solution. Future scope: There is scope to test and enhance hyperheuristic framework to handle complexity in virtual machine placement problem with many conflicting objectives. This framework can be extended to make it domain independent, for similar class problems.

References 1. Kaur, A., Kalra, M.: Energy optimized VM placement in cloud environment. In: 2016 6th International Conference on Cloud System and Big Data Engineering (Confluence), pp. 141– 145. IEEE, January 2016 2. Mishra, M., Bellur, U.: Whither tightness of packing? The case for stable VM placement. IEEE Trans. Cloud Comput. 4(4), 481–494 (2016) 3. Zheng, Q., et al.: Virtual machine consolidated placement based on multi-objective biogeography-based optimization. Future Gener. Comput. Syst. 54, 95–122 (2016) 4. Tang, M., Pan, S.: A hybrid genetic algorithm for the energy-efficient virtual machine placement problem in data centers. Neural Process. Lett. 41(2), 211–221 (2015) 5. Satpathy, A., Addya, S.K., Turuk, A.K., Majhi, B., Sahoo, G.: Crow search based virtual machine placement strategy in cloud data centers with live migration. Comput. Electr. Eng. 69, 334–350 (2018) 6. Silva Filho, M.C., Monteiro, C.C., Inácio, P.R., Freire, M.M.: Approaches for optimizing virtual machine placement and migration in cloud environments: a survey. J. Parallel Distrib. Comput. 111, 222–250 (2018) 7. Bose, S.K., Sundarrajan, S.: Optimizing migration of virtual machines across data-centers. In: 2009 International Conference on Parallel Processing Workshops, ICPPW 2009, pp. 306–313. IEEE, September 2009 8. Feller, E., Morin, C., Esnault, A.: A case for fully decentralized dynamic VM consolidation in clouds. In: 2012 IEEE 4th International Conference on Cloud Computing Technology and Science (CloudCom), pp. 26–33. IEEE, December 2012

Hyperheuristic Framework with Evolutionary and Deterministic Algorithms

349

9. Feller, E., Rohr, C., Margery, D., Morin, C.: Energy management in IaaS clouds: a holistic approach. In: 2012 IEEE 5th International Conference on Cloud Computing (CLOUD), pp. 204–212. IEEE, June 2012 10. Masdari, M., Nabavi, S.S., Ahmadi, V.: An overview of virtual machine placement schemes in cloud computing. J. Netw. Comput. Appl. 66, 106–127 (2016) 11. Lopez-Pires, F., Baran, B.: Virtual machine placement literature review. arXiv preprint arXiv:1506.01509 (2015) 12. Adamuthe, A.C., Patil, J.T.: Differential evolution algorithm for optimizing virtual machine placement problem in cloud computing. Int. J. Intell. Syst. Appl. 10(7), 58 (2018) 13. Malekloo, M., Kara, N.: Multi-objective ACO virtual machine placement in cloud computing environments. In: Globecom Workshops (GC Wkshps), pp. 112–116. IEEE, December 2014 14. Choudhary, A., Rana, S., Matahai, K.J.: A critical analysis of energy efficient virtual machine placement techniques and its optimization in a cloud computing environment. Procedia Comput. Sci. 78, 132–138 (2016) 15. Dhanoa, I.S., Khurmi, S.S.: Power efficient hybrid VM allocation algorithm. Int. J. Comput. Appl. 127(17), 39–43 (2015) 16. Sookhtsaraei, R., Madani, M., Kavian, A.: A multi objective virtual machine placement method for reduce operational costs in cloud computing by genetic. Int. J. Comput. Netw. Commun. Secur. 2(8), 1–10 (2014) 17. Shi, K., Yu, H., Luo, F., Fan, G.: Multi-objective biogeography-based method to optimize virtual machine consolidation. In: SEKE, pp. 225–230 (2016) 18. Addya, S.K., Turuk, A.K., Sahoo, B., Sarkar, M., Biswash, S.K.: Simulated annealing based VM placement strategy to maximize the profit for cloud service providers. Eng. Sci. Technol. Int. J. 20(4), 1249–1259 (2017) 19. Xu, J., Fortes, J.A.: Multi-objective virtual machine placement in virtualized data center environments. In: 2010 IEEE/ACM International Conference on Green Computing and Communications (GreenCom) and International Conference on Cyber, Physical and Social Computing (CPSCom), pp. 179–188. IEEE, December 2010 20. Chakhlevitch, K., Cowling, P.: Choosing the fittest subset of low level heuristics in a hyperheuristic framework. In: Raidl, G.R., Gottlieb, J. (eds.) EvoCOP 2005. LNCS, vol. 3448, pp. 23–33. Springer, Heidelberg (2005). https://doi.org/10.1007/978-3-540-31996-2_3 21. Cowling, P., Kendall, G., Soubeiga, E.: A hyperheuristic approach to scheduling a sales summit. In: Burke, E., Erben, W. (eds.) PATAT 2000. LNCS, vol. 2079, pp. 176–190. Springer, Heidelberg (2000). https://doi.org/10.1007/3-540-44629-X_11 22. Fisher, H., Thompson, G.L.: Probabilistic learning combinations of local job-shop scheduling rules. In: Factory Scheduling Conference, Carnegie Institute of Technology (1961) 23. Fisher, H.: Probabilistic learning combinations of local job-shop scheduling rules. In: Industrial Scheduling, pp. 225–251 (1963) 24. Crowston, W.B., Glover, F., Trawick, J.D.: Probabilistic and parametric learning combinations of local job shop scheduling rules (No. ONR-RM117). Carnegie Institute of Technology, Pittsburgh, PA, Graduate School of Industrial Administration (1963) 25. Mockus, J.: A Set of Examples of Global and Discrete Optimization: Applications of Bayesian Heuristic Approach, vol. 41. Springer, Dordrecht (2000). https://doi.org/10.1007/ 978-1-4615-4671-9 26. Mockus, J.B., Mockus, L.J.: Bayesian approach to global optimization and application to multiobjective and constrained problems. J. Optim. Theory Appl. 70(1), 157–172 (1991)

350

A. C. Adamuthe and A. Jadhav

27. Mockus, J., Eddy, W., Reklaitis, G.: Bayesian Heuristic Approach to Discrete and Global Optimization: Algorithms, Visualization, Software, and Applications, vol. 17. Springer, Dordrecht (2013) 28. Ross, P., Schulenburg, S., Marín-Bläzquez, J.G., Hart, E.: Hyper-heuristics: learning to combine simple heuristics in bin-packing problems. In: Proceedings of the 4th Annual Conference on Genetic and Evolutionary Computation, pp. 942–948. Morgan Kaufmann Publishers Inc., July 2002 29. Cowling, P., Kendall, G., Han, L.: An investigation of a hyperheuristic genetic algorithm applied to a trainer scheduling problem. In: 2002 Proceedings of the 2002 Congress on Evolutionary Computation, CEC 2002, vol. 2, pp. 1185–1190. IEEE (2002) 30. Han, L., Kendall, G., Cowling, P.: An adaptive length chromosome hyper-heuristic genetic algorithm for a trainer scheduling problem. In: Recent Advances in Simulated Evolution and Learning, pp. 506–525 (2004) 31. Nareyek, A.: Choosing search heuristics by non-stationary reinforcement learning. In: Nareyek, A. (ed.) Metaheuristics: Computer Decision-Making, vol. 86, pp. 523–544. Springer, Boston (2003). https://doi.org/10.1007/978-1-4757-4137-7_25 32. Burke, E.K., Kendall, G., Soubeiga, E.: A tabu-search hyperheuristic for timetabling and rostering. J. Heuristics 9(6), 451–470 (2003) 33. Burke, E.K., et al.: Hyper-heuristics: a survey of the state of the art. J. Oper. Res. Soc. 64 (12), 1695–1724 (2013) 34. Özcan, E., Bilgin, B., Korkmaz, E.E.: A comprehensive analysis of hyper-heuristics. Intell. Data Anal. 12(1), 3–23 (2008) 35. Dong, J., Jin, X., Wang, H., Li, Y., Zhang, P., Cheng, S.: Energy-saving virtual machine placement in cloud data centers. In: 2013 13th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid), pp. 618–624. IEEE, May 2013 36. Qian, F., Ding, R.: Simulated annealing for the 0/1 multidimensional knapsack problem. Numer. Math.-Engl. Ser. 16(4), 320 (2007) 37. Shah-Hosseini, H.: Problem solving by intelligent water drops. In: 2007 IEEE Congress on Evolutionary Computation, CEC 2007, pp. 3226–3231. IEEE, September 2007

Efficient Directional Information Dissemination Approach in Vehicular Ad Hoc Networks Sandeep Kad1(&) and Vijay Kumar Banga2 1

2

IKG Punjab Technical University, Kapurthala, India [email protected] Amritsar College of Engineering and Technology, Amritsar, India [email protected]

Abstract. Vehicular Ad hoc Networks (VANETs) are attracting huge attention from research community these days. Timely delivery of information to the potential vehicle nodes is a challenging task. High mobility, frequent change in topology and uneven density of vehicles, makes the task of selecting vehicle as relay node for information dissemination difficult. Broadcasting is the simpler and common technique to serve multiple vehicles simultaneously. The conventional broadcasting approaches results in broadcast storm problem which leads to channel contention, and frequent collisions in the network. In this paper we are proposing an efficient information dissemination approach, where broadcast suppression is done by reducing the number of relay vehicle nodes and at the same time allocates higher probability to the most suitable vehicles for rebroadcasting the message in the direction of concern. Simulation of the proposed approach shows less number of collisions, low end to end delay and high message delivery ratio. Keywords: Vehicular Ad hoc Network
Intelligent Transport System Broadcast storm
Information dissemination




1 Introduction Smart cities have been the talk of the town from the last few years. Vehicular Adhoc Network being an important component of Intelligent Transport System (ITS) [1] is getting due attentions in the similar context as it leads to improvise the travelling experiences of the drivers and commuters. Huge number of vehicles ply on roads everyday with millions and millions of passengers travelling in them, so the safety of human lives is of utmost importance. A report from United States Department of Transportation, National Highway Traffic Safety Administration, reveals that in 2010, in United States alone, 32,999 people lost their lives in road mishaps with 3.9 million injured and 24 million vehicles damaged resulting in $242 billion of economic losses [2]. VANETs are somewhat similar to Mobile Ad hoc Networks (MANETs) and are self-organized and managed [3] with identical shared transmission conditions. The components of VANET comprises of Application Unit (AU), On Board Unit (OBU) and Road Side Unit (RSU’s) [4]. These vehicular nodes communicate by means of wireless © Springer Nature Singapore Pte Ltd. 2019 M. Prateek et al. (Eds.): NGCT 2018, CCIS 922, pp. 351–363, 2019. https://doi.org/10.1007/978-981-15-1718-1_30

352

S. Kad and V. K. Banga

medium called Wireless Access Vehicular Environment (WAVE), in which IEEE 802.11p is used that facilitates communication in the range of 1000 m, however, in practical upto 300 m is feasible [5]. In VANETs, applications can be categorized into safety applications and comfort applications. Safety applications include exchange of information (e.g. warnings about road accidents, low bridges, snow or slippery road, pollution and other road conditions etc.) [6] that creates an awareness amongst the commuters about the surrounding environment and the intentions of other vehicle drivers on the move. In general, in case of some emergency situation, on highways, vehicle movement is at very high speed and it is obvious that it will take time to move down to a safer speed. Therefore if drivers by some means can get the information well in time, they can definitely avert the miss happening which otherwise might be riskier. Therefore the role of data dissemination is quite important for efficient communication among the vehicular nodes in the network [7]. Data Dissemination can be classified basically into two models (i) pull model (ii) push model. A third category also exists which is the combination of the other two and is known as hybrid model [8]. Sometimes the information is to be disseminated to all the vehicles in a particular range or area of concern and broadcasting has been the most popular and widely used way for information dissemination [9]. For ex. Information about traffic jam or accident which is to be disseminated is broadcast to all the vehicles in the area of concern. Broadcast protocols have been classified by researchers as (i) Single hop and Multihop (ii) Deterministic and Probabilistic. In the single hop broadcast the source vehicle node broadcasts the message to all its one hop neighbor i.e. to all the vehicular nodes that are in communication range of source vehicular node and thereafter these one hop nodes do not rebroadcast the information further whereas in case of multihop broadcast protocols, source vehicle broadcast the packet to all the vehicles in its vicinity and these receivers act as relay nodes and rebroadcast this packet [10]. In deterministic protocols, few vehicular nodes from a group of nodes for ex. Multipoint Relays (MPR) are involved in the broadcast whereas in probabilistic broadcasting approaches, different nodes rebroadcast the packet on the basis of their probability which they acquire after following some criteria [11]. A simple broadcast by means of flooding results in huge redundancy of information leading to interference in shared medium of communication among neighboring vehicular nodes [12]. This leads to broadcast storm problem especially in dense scenarios which then results in delaying the delivery of information [13]. On the other hand in sparse environment, i.e. during off peak hours or at midnight, information dissemination becomes difficult due to nonavailability of relay nodes. So the two extreme traffic scenario poses different kind of problems which needs to be dealt with. In order to address the above mentioned issues, we propose a directional data dissemination approach that takes speed of the vehicles into consideration after determining the traffic density while gathering information about the neighboring vehicles instead of using beacons periodically that leads to network congestion and wastage of bandwidth. The rest of the paper is organized as: Sect. 2 provides a brief overview of broadcasting techniques in Vehicular Adhoc Network. Section 3 and 4 explains the

Efficient Directional Information Dissemination Approach in VAHNETs

353

objectives, the assumptions for proposed technique followed by proposed algorithm. Section 5 presents the performance analysis of the proposed approach and in Sect. 6 concludes the proposed work.

2 Related Work Keeping in view the density of vehicles, various solutions have been proposed by the researchers, to disseminate the information well in time to all the recipient vehicular nodes. Broadcast though, a most common way out to disseminate information, results in number of issues which needs to addressed. To mitigate the broadcast storm problem, number of broadcast suppression techniques have been proposed wherein least number of vehicles are selected as relay nodes so that issues like channel contention, redundancy, bandwidth wastage, collisions can be kept under control resulting in timely delivery of information to all the desired destination vehicles. Bachir et al. [14] proposed a distributed Inter-Vehicle Geocast (IVG) protocol for disseminating safety related information in highway scenario. Whenever the information is to be shared, firstly the area of concern is calculated, set of forwarding vehicular relay nodes (also called multicast group) are identified. Whenever a packet is received for rebroadcast, firstly its unique id is verified to check, if the packet is received for the first time, if it is so, it waits for specified time interval before it rebroadcast. Tonguz et al. [15] proposed three distance based broadcast suppression techniques where each node evaluates the rebroadcasting probability based upon the local topology information. Weighted p-persistence is a probabilistic scheme where the distance between the sender vehicle and the receiver vehicle decides the probabilistic value for rebroadcasting. If the node receives the message for the first time, it is rebroadcast with the evaluated probability, else it is discarded. Slotted 1-persistence and Slotted p-persistence are the other techniques where vehicle nodes upon checking message ID rebroadcasts with probability 1 or probability p respectively in the assigned slot in case no duplicate copy of message is received before the assigned time slot. Korkmaz et al. proposed Urban Multihop Broadcast (UMB) [16] protocol that addresses broadcast storm problem where the source vehicular node selects the farthest vehicular node in the direction of broadcast for disseminating information. At all the intersections, repeaters are used for directing data. These repeaters perform well in the environment where high rise buildings otherwise act as obstacles for vehicles. However, installation of repeaters at all the intersecting locations is a cumbersome task. Durresi et al. [17] proposed an emergency broadcast protocol where vehicles by means of sensor detect emergency situations and broadcast the information to the surrounding vehicles. In this approach, the highway is divided into number of dynamic virtual cells that moves along with the vehicles on the road. The vehicle nodes inside a virtual cell, selects cell reflector that behaves as a base station, receives messages from nodes inside a cell as well as from neighboring cells. For disseminating information the cell reflector does prioritization of all the messages received and accordingly decides the forwarding sequence. One of the major flaws in BROADCOMM protocol is that, it does not address network partition problem and performs in highway scenarios only.

354

S. Kad and V. K. Banga

Yang et al. [18] proposed Position based Adaptive Broadcast IVC for two-way multilane highway communication. The decision to disseminate information is done keeping in view position, direction and velocity of source node, relay nodes and destination nodes. Hung et al. [19] proposed Mobility Pattern Aware Routing Protocol where each vehicular node communicates with other vehicular nodes as well as with the base station. The base station keeps up to date information of all the vehicles in the network which includes geographical position, vehicle id and speed of the vehicle etc. Vehicles forward the route requests to the base station whenever they need to forward the information to any of the vehicles in the network. The base station calculates the path from source to the destination and forwards the reply to the source vehicular node. Thereafter, the source disseminates the information to the destined vehicle. In case there is broken link, the source vehicle forwards the information through base station to the destination. Distributed Vehicular broadcast (DV-Cast) protocol proposed by Tonguz et al. [20] operates both in dense environment by applying broadcast suppression techniques and sparse environment using store carry and forward concept thus mitigating broadcast storm problem and network partition issues respectively. It uses information about local topology to handle message rebroadcasting, which in turn adds overhead. In case of high mobility in highway scenario, maintaining optimal beacon frequency is quite difficult. Urban Vehicular Broadcast (UV-Cast) protocol proposed by Viriyasitavat et al. [21] addresses both broadcast storm problem and network partition in the urban vehicular environment. It performs both in well-connected and disconnected environment without the support of any infrastructure. Simple and Robust Dissemination (SRD) protocol [22] works for highway traffic environments. SRD proposes optimized 1-persistence broadcast suppression technique. Priorities are assigned to vehicles for rebroadcasting. Store and carry forward technique is used for network partition situations. The drawback of SRD is that it allocates higher priorities for rebroadcasting to the vehicles moving in direction of source vehicular node whereas a better intermediary vehicular node for rebroadcast might be available in the opposite direction. Bakhouya et al. [23] proposed an adaptive decentralized approach for disseminating information in VANETs. Every node dynamically adjusts the values of local parameters on the basis of information gathered from neighboring nodes and decides accordingly whether to rebroadcast the message or discard it. Vilas et al. [24] proposed DRIVE protocol that utilizes local one hop information to deliver messages, in dense environment, a vehicle inside a sweet spot is preferred to rebroadcast the message further to the vehicles and eliminate broadcast storm problem. Position of vehicle and timer are used, before a vehicle actually rebroadcasts. In case of partitioned network, vehicles outside area of interest is chosen for disseminating information. Chaqfeh et al. [25] proposed three variations for broadcast suppression: Speed Adaptive Broadcast (SAB), Slotted Speed Adaptive Broadcast (SSAB) and Grid Speed Adaptive Broadcast (GSAB). In case of SAB, traffic is dynamically detected using speed data of vehicles. Depending upon traffic conditions, every vehicle sets its probability for rebroadcast or delay. Further to implement broadcast suppression & reduce the amount of redundant data, separation in time slots among vehicles results in sufficient amount of time for vehicles to decide whether to rebroadcast or discard the rebroadcast after the message is received. In GSAB approach, to avoid same time slots

Efficient Directional Information Dissemination Approach in VAHNETs

355

to be assigned to vehicles moving in different lanes in the same direction, a dissemination delay is added among the vehicles of different lanes thus decreasing data redundancy and collision. Information about neighboring vehicles is not required by these protocols. Achour [26] proposed Simple and Efficient Adaptive Data dissemination (SEAD) protocol that addresses the broadcast storm problem by controlling excessive rebroadcasts in highway scenarios. It considers vehicle density and message direction to determine the probability of rebroadcast. Emergency Degree Broadcast (EDCast) protocol [27] takes into consideration the emergency degree of the packet as the decisive factor for prioritizing the packet broadcast, so that safety related information can be disseminated accurately and well in time. Higher broadcasting probability and smaller size contention window is assigned to the packet with higher emergency degree value. Latif et al. [28] proposed position based store and carry, broadcast storm mitigation approach where relay vehicular node for rebroadcasting is selected on the basis of direction, distance and position of nodes. The decision for best suitable vehicle is done on the basis of Analytical Network Process (ANP). Liu et al. [8] proposed Clustering and Probabilistic Broadcasting (CPB) scheme where cluster is constructed according to driving direction and forwarding probability is evaluated and adjusted as per varying traffic density. Chou et al. [29] proposed Appropriate Vehicular Emergency Dissemination (AVED) approach where vehicle’s crash sensors upon detecting impact signal broadcast an emergency message. Each vehicle shares Cooperative Awareness Message (CAM) consisting self-information about position, velocity and other information about the vehicle. This information is used to identify the best forwarder which is at a far location with relatively low velocity. The drawback of this approach is the presence of higher number of beacon messages to gather 1-hop information. Table 1 shows comparison of some broadcast based information dissemination techniques. It has been observed that in many of the existing approaches that periodic beacons are used which results in extra network overheads leading to congestion and further delays.

3 Objectives and Assumptions It has been observed that in dense traffic environment redundant broadcast leads to broadcast storm problem which largely impact the information delivery ratio and delay in delivery of information as well. Our objective in the proposed approach is to disseminate information with improved message delivery ratio and low delay. To suppress the broadcast storm in our approach, not all the relay nodes will broadcast, instead, all vehicles determine the probability of rebroadcast on the basis of additional communication range they can reach out in the area of concern in comparison to the sender node. To implement our proposed approach we have made few assumptions like: We are considering multilane highway where traffic movement is in both the directions. The information which is to be shared may have directional importance like

356

S. Kad and V. K. Banga

in case of traffic jam or a crash, vehicles moving in particular lane are the one which should be intimated about this situation. This information is to be shared in some limited geographical area say within the range of 3 to 4 km as well as the freshness of this information is also to be taken into consideration as the its importance is time specific. All the vehicles are equipped with devices as per IEEE 802.11p WAVE standards and Global Position System (GPS) to know the geographical position of the vehicles on the road segments. The communication range for all the vehicles is assumed to be same. There is no road side infrastructure available. The information that packet header contains include: source vehicle ID, source vehicle position coordinates, message ID, sender ID (relay nodes), source vehicle position coordinates, sender vehicle’s speed, timestamp (added by source vehicle), message dissemination direction.

Table 1. Comparison of some broadcast based data dissemination techniques in VANETs Protocol/approach Forwarding strategy

Architecture Environment GPS Mitigation support approach

Simulation

CPB [8] IVG [14]

V2V V2V

Highway Highway

Yes Yes

Probabilistic Delay based

NS2 GLOMOSIM

V2V

Urban

Yes

Probabilistic

OPNET

V2V

Urban

Yes

Delay based

OPNET

V2V

Urban

Yes

OPNET

V2V, V2I

Urban

Yes

Probabilistic/delay based Delay based

V2V

Highway

Yes

Delay based

–

V2V

Highway

Yes

Delay based

NS2

V2V

Urban

Yes

Delay based

SUMO

V2V

Highway

Yes

Delay based/probabilistic

OMNET++

V2V

Urban, Highway

Yes

Delay based

OMNET++

V2V

Urban

Yes

OMNET++

Hybrid V2V V2V

Highway Highway Highway

Yes Yes Yes

Probabilistic/delay based Hybrid Emergency degree Delay based

Weighted pPersistence [15] Slotted 1Persistence [15] Slotted pPersistence [15] UMB [16] BROADCOMM [17] DVCAST [20] UVCAST [21]

SRD [22]

DRIVE [24]

G-SAB [25] SEAD [26] ED-CAST [27] AVED [29]

Clustering based Position and Distance based Position and distance based Position and distance based Position and Distance based Position and distance based Position and distance based Position based, store and carry Position and distance based, store and carry Position and distance based, store and carry Position and distance based, store and carry Position and distance based Position based Distance based Position based

MATLAB

NS3, SUMO NS2, SUMO NS2

Efficient Directional Information Dissemination Approach in VAHNETs

357

4 Proposed Approach As our aim is to mitigate broadcast storm problem, we are suggesting a modification in optimized Slotted 1-persistence [22] where higher priority is given to vehicle in particular direction. In our approach we are assigning higher priorities to the vehicles whose communication range is farthest in the direction of message dissemination in comparison to current sending vehicle. This selection is done from all the vehicles who have got the message irrespective of their direction on movement. Further to reduce network overheads that arises because of beacons which are periodically disseminated to gather information about 1-hop neighbors, we are evaluating vehicle density in the neiging region on the basis of the speed of vehicle which at present are the part of the network. The flow of traffic (Tf) depends both on speed of vehicles (S) and traffic density (Td) [30]. It is given by Tf ¼ S  Td

ð1Þ

In general it is observed that vehicular movement will be at higher speeds if traffic density is less and vice versa. A linear relationship between vehicular speed and traffic density is given in [30] S Smax

¼1

D Dj

ð2Þ

Here Smax represents a maximum permissible speed on the road segment and is also referred as a state of free flow traffic, D refers to current traffic density and Dj is the S traffic density in extremely dense scenario i.e. a situation of traffic jam. If Smax approaches 1, it refers to more of a state of free flow. This happen in situations when traffic density is low. Whenever vehicle y receives a message from vehicle x, instead of rebroadcasting it immediately, it verifies from the message id to know if the message is received for the first time, thereafter probability of rebroadcast is evaluated. This is done by all the relay nodes which are candidates for rebroadcast. The probability of rebroadcast is given as PBxy ¼

minðdistxy ; RangeÞ Range

ð3Þ

In this dist refers to the distance between the sender vehicle x and receiver vehicle y. distxy ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðxb  xa Þ2 þ ðyb  ya Þ2

ð4Þ

In case PBxy comes out to be 1, it means that vehicle y at a location that is maximum possible of range of communication for vehicle x. If this is the scenario, it

358

S. Kad and V. K. Banga

implies that the probability of y to rebroadcast the message is highest. Timeslot allocation to each vehicle is done while keeping in consideration the traffic density at that moment. It indicates the waiting period before a vehicle can rebroadcast or can decide to ignore the message received. This period also addresses the channel contention and channel collision issues. Timeslot allocation Sxy is given by    Sxy ¼ N  1  PBxy

ð5Þ

Algorithm of the Proposed Approach Input:

(xa, ya) // Geographical Position of Sender Vehicle (xb, yb) // Geographical Position of Receiver Vehicle Output: Waiting Period // The time for which vehicular nodes waits before rebroadcast Start Compare the geographical coordinates of the sender vehicle and the receiver vehicle If (Sender is more in the direction of data dissemination in comparison to receiver) If (rebroadcast scheduled) cancel broadcast End If else If (message id already exists) If (message is scheduled) postpone rebroadcast If (message received within 1-hop delay) discard message End if End if else calculate Sxy and TSxy schedule rebroadcast End if End if End

N represents number of timeslots and is computed as [25] 

Sx N ¼ ð1  TSmax Þ  þ TSmax Smax

 ð6Þ

where Sx is speed of the sender vehicle which it shares with all the vehicles while disseminating the message, TSmax represent maximum timeslots whose value depends upon transmission range and width of single timeslot. A vehicle may receive same message from multiple senders, in that case it selects minimum probability of rebroadcast for itself. The decision for discarding an already received message is not

Efficient Directional Information Dissemination Approach in VAHNETs

359

taken immediately, instead the vehicle wait for a small period of time that is at the maximum 1-hop delay (comprising of transmission, propagation and medium access delay) in the scenario. In case during this period same message is received then the rebroadcast is finally avoided else it is rebroadcasted with probability 1. Though, allocation of timeslots addresses the issue of channel contention and collision to some extent, still there may be a situation where multiple vehicles are assigned similar timeslot and they may go for rebroadcast simultaneously. To avoid this situation a small random delay (delayrand of say upto 1 ms) is added to break this synchronization. So the final waiting period or the delay TSxy after which the vehicle can rebroadcast is given as   TSxy ¼ Sxy  a þ delayrand

ð7Þ

where a is the 1-hop delay. As discussed in the algorithm, upon receiving a message, receiver first of all verifies whether the sender is at far location in the direction of dissemination or not, if it is so, the message is discarded, else it is verified and if the same message was received earlier or not. If the message was received earlier, it is not immediately discarded, instead the vehicle node waits for some time (about 1-hop delay) and if during this period the same message is received again then the message is finally discarded otherwise the message is rebroadcasted with probability 1. The waiting period for a vehicle before it can disseminate a message is computed as discussed in Eq. 7.

5 Performance Analysis and Results In this section the performance of the proposed approach is discussed. Simulation of the proposed approach is done using NetSim1 and the traffic patterns are generated using SUMO [31]. To evaluate the results, comparison of proposed approach is done with the optimized slotted 1 persistence [22]. We are considering a 4-lane straight road segment of 5 km stretch with 2-lanes in each direction. The parameters for simulation environment are shown in Table 2. The simulation is done 10 times, and mean of these simulations is done to get final results. The evaluation is done on following metrics: • Message Delivery Ratio (MDR) – It defines the percentage of messages successfully delivered in the network. In an ideal scenario, message delivery should be almost 100%. • No. of Collisions: It defines the average number of packets collided for all vehicles at MAC layer. Lower number of collisions results in decline in broadcast storm. • Delay- The average amount of time taken by message disseminated from the source to all its destination. In safety applications this delay should be minimal.

1

https://www.tetcos.com/.

360

S. Kad and V. K. Banga Table 2. Simulation parameters Parameter Frequency band Bandwidth Transmission range Friis path loss exponent (n) Bit rate Highway length Maximum speed in lane Message size Inter-arrival time Minimum slot width Number of runs Vehicle density (vehicles/km/lane) Confidence level

Value 5.9 GHz 10 MHz 250 m 3 6 Mbps 5 km 100 Kmph 2312 bytes 2s 10 m 30 [20, 40, 60, 80, 100] 95%

Figure 1 depicts the message delivery ratio and the simulation results reveals that in both optimized slotted 1 persistence as well as in the proposed approach the message delivery ratio as desired is almost 100%. Figure 2 presents the end to end delay that happens when the messages are disseminated in the network. It is clearly visible that the proposed approach is performing quite better than the optimized slotted 1 persistence. It is because, no beacons are disseminated periodically as is the case in optimized slotted 1 persistence approach which adds to the networks overheads. Further it can be observed that as the vehicle density increases the gap in terms of delay for these approaches is further widening, it is so because, with rise in number of vehicles, there is rise in beacons as well.

Fig. 1. Message delivery ratio

Fig. 2. End to end delay

Figure 3 shows average collisions which are there when multiple vehicular nodes try to access medium simultaneously. The simulation results reveal the superiority of the proposed approach over optimized slotted 1 persistence approach. It is observed

Efficient Directional Information Dissemination Approach in VAHNETs

361

that with increase in flow of traffic the number of collisions increases in optimized slotted 1 persistence approach, one of the reasons for this might be use of periodic beacons for gathering information about the neighboring traffic. These beacon messages even though small in size, adds to the traffic, thus resulting in more collisions. It can also be noted that the difference in average number of collisions is increasing as the vehicle density is increasing.

Fig. 3. Number of collisions

6 Conclusion The timely dissemination of information to all the concerned vehicles is the key to the success of VANETs. Though broadcasting is commonly used approach for delivery of information to multiple vehicles simultaneously but a simple broadcast results in number of issues like contention for medium access, collisions, wastage of limited bandwidth leading to delay in delivery of information. These issues need to be addressed, so as to design an efficient information dissemination system. In our proposed approach speed of vehicles is utilized to gather information about traffic density instead of periodic beacons which are used in optimized slotted 1 persistence approach. In comparison to optimized slotted 1 persistence approach the proposed approach assigns priorities to relay vehicles on the basis of their communication range (i.e. vehicle that is farthest from the sender vehicle) in the direction of message dissemination irrespective of the direction in which the vehicle is moving. The simulation results reveals that the proposed approach reduces the end to end delay, average number of collisions while maintaining almost 100% message delivery ratio thus addressing broadcast storm to some extent. In future we would like to extend the proposed work by considering both urban as well as highway environments. Acknowledgements. The authors gratefully acknowledge IKG Punjab Technical University, Kapurthala, India and Amritsar College of Engineering and Technology, Amritsar, India to carry out this research work.

362

S. Kad and V. K. Banga

References 1. Chaqfeh, M., Lakas, A., Jawhar, I.: A survey on data dissemination in vehicular ad hoc networks. Veh. Commun. 1(4), 214–225 (2014) 2. The economic and societal impact of motor vehicle crashes, 2010 (Revised)1. Ann. Emerg. Med. 66, 194–196 (2015) 3. Wischhof, L., Ebner, A., Rohling, H.: Information dissemination in self-organizing intervehicle networks. IEEE Trans. Intell. Transp. Syst. 6(1), 90–101 (2005) 4. Al-Sultan, S., Al-Doori, M.M., Al-Bayatti, A.H., Zedan, H.: A comprehensive survey on vehicular ad hoc network. J. Netw. Comput. Appl. 37, 380–392 (2014) 5. Chakroun, O., Cherkaoui, S.: Overhead-free congestion control and data dissemination for 802.11p VANETs. Veh. Commun. 1, 123–133 (2014) 6. Cunha, F., et al.: Data communication in VANETs: protocols, applications and challenges. Ad Hoc Netw. 44, 90–103 (2016) 7. Ghebleh, R.: A comparative classification of information dissemination approaches in vehicular ad hoc networks from distinctive viewpoints: a survey. Comput. Netw. 131, 15–37 (2018) 8. Liu, L., Chen, C., Qiu, T., Zhang, M., Li, S., Zhou, B.: A data dissemination scheme based on clustering and probabilistic broadcasting in VANETs. Veh. Commun. 13, 78–88 (2018) 9. Dressler, F., Klingler, F., Sommer, C., Cohen, R.: Not all VANET broadcasts are the same: context-aware class based broadcast. IEEE/ACM Trans. Netw. 26(1), 17–30 (2018) 10. Panichpapiboon, S., Pattara-atikom, W.: A review of information dissemination protocols for vehicular ad hoc networks. IEEE Commun. Surv. Tutor. 14, 784–798 (2011) 11. Reina, D.G., Toral, S.L., Johnson, P., Barrero, F.: A survey on probabilistic broadcast schemes for wireless ad hoc networks. Ad Hoc Netw. 25, 263–292 (2015) 12. C., Hu, Hong, Y., Hou, J.: On mitigating the broadcast storm problem with directional antennas. In: International Conference on Communication, pp. 104–110. IEEE, USA (2003) 13. Ni, S.-Y., Tseng, Y.-C., Chen, Y.-S., Sheu, J.-P.: The broadcast storm problem in a mobile ad hoc network. In: 5th International Conference on Mobile Computing and Networking, pp. 151–162. ACM/IEEE, USA (1999) 14. Bachir, A., Benslimane, A.: A multicast protocol in ad hoc networks inter-vehicle geocast. In: 57th IEEE Semiannual Vehicular Technology Conference, pp. 2456–2460. IEEE, South Korea (2003) 15. Tonguz, O.K., Wisitpongphan, N., Parikh, J.S., Bai, F., Mudalige, P., Sadekar, V.K.: On the broadcast storm problem in ad hoc wireless networks. In: 3rd International Conference on Broadband Communications, Networks and Systems, pp. 1–11. IEEE, USA (2006) 16. Korkmaz, G., Ekici, E., Özgüner, F., Özgüner, Ü.: Urban multi-hop broadcast protocol for inter-vehicle communication systems. In: 1st International Workshop on Vehicular Adhoc Networks, pp. 76–85. ACM, USA (2004) 17. Durresi, M., Durresi, A., Barolli, L.: Emergency broadcast protocol for inter-vehicle communications. In: International Conference on Parallel and Distributed Systems, pp. 402– 406. IEEE, Japan (2005) 18. Yang, Y.-T., Chou, L.-D.: Position-based adaptive broadcast for inter-vehicle communications. In: International Conference on Communications Workshops, pp. 410–414. IEEE, China (2008) 19. Hung, C.-C., Chan, H., Wu, E.H.-K.: Mobility pattern aware routing for heterogeneous vehicular networks. In: International Conference on Wireless Communications and Networking, pp. 2200–2205. IEEE, USA (2008)

Efficient Directional Information Dissemination Approach in VAHNETs

363

20. Tonguz, O., Wisitpongphan, N., Bai, F.: DV-CAST: a distributed vehicular broadcast protocol for vehicular ad hoc networks. IEEE Wirel. Commun. 17, 47–57 (2010) 21. Viriyasitavat, W., Tonguz, O.K., Bai, F.: UV-CAST: an urban vehicular broadcast protocol. IEEE Commun. Mag. 9, 116–124 (2011) 22. Schwartz, R.S., Barbosa, R.R.R., Meratnia, N., Heijenk, G., Scholten, H.: A directional data dissemination protocol for vehicular environments. Comput. Commun. 34, 2057–2071 (2011) 23. Bakhouya, M., Gaber, J., Lorenz, P.: An adaptive approach for information dissemination in vehicular ad hoc networks. J. Netw. Comput. Appl. 34, 1971–1978 (2011) 24. Villas, L.A., Boukerche, A., Maia, G., Pazzi, R.W., Loureiro, A.A.F.: DRIVE: an efficient and robust data dissemination protocol for highway and urban vehicular ad hoc networks. Comput. Netw. 75, 381–394 (2014) 25. Chaqfeh, M., Lakas, A.: A novel approach for scalable multi-hop data dissemination in vehicular ad hoc networks. Ad Hoc Netw. 37, 228–239 (2016) 26. Achour, I., Bejaoui, T., Busson, A., Tabbane, S.: SEAD: a simple and efficient adaptive data dissemination protocol in vehicular ad-hoc networks. Wirel. Netw. 22, 1673–1683 (2016) 27. Wang, W., Luo, T., Hu, Y.: An adaptive information quantity-based broadcast protocol for safety services in VANET. Mob. Inf. Syst. 2016, 1–12 (2016) 28. Latif, S., et al.: Multicriteria based next forwarder selection for data dissemination in vehicular ad hoc networks using analytical network process. Math. Probl. Eng. 2017, 1–18 (2017) 29. Chou, Y.-H., Chu, T.-H., Kuo, S.-Y., Chen, C.-Y.: An adaptive emergency broadcast strategy for vehicular ad hoc networks. IEEE Sens. J. 18, 4814–4821 (2018) 30. Chowdhury, M.S., Sadek, A.: Fundamentals of Intelligent Transport Systems Planning. Artech House, Norwood (2013) 31. Krajzewicz, D., Erdmann, J., Behrisch, M., Bieker, L.: Recent development and applications of SUMO – Simulation of Urban MObility (2012)

Software Agents in Support of Scheduling Group Training Giorgi Mamatsashvili1, Konrad Gancarz1, Weronika Łajewska1, Maria Ganzha2, and Marcin Paprzycki3(&) 1

2

Faculty of Mathematics and Information Sciences, Warsaw University of Technology, Warsaw, Poland Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland 3 Faculty of Management and Technical Sciences, Warsaw Management Academy, Warsaw, Poland [email protected]

Abstract. Nowadays, being fit is becoming more and more popular. This includes eating habits – with, for instance, companies offering personalized boxdiets – as well as exercising. Obviously, one can go to a fitness club alone, but it is much more fun to go together with friends. Here, it should be obvious that reaching an agreement, between two persons, on the time and location of a fitness club, may be relatively easy. However, it becomes more complex with each additional person that would like to join the group. The aim of this contribution is to show how an agent-based application can be used to negotiate schedule of group training sessions. As a matter of fact, it will be shown that, under a limited number of “good-will assumptions”, the proposed application can fully eliminate human involvement in the scheduling process, as it can find the best suitable place and time for a training session, considering individual preferences. Keywords: Software agents


Scheduling
Individual preferences

1 Introduction The idea of scheduling something, as straightforward as a training session, isn’t very complicated. Therefore, one may question the need for a “software support” in a seemingly trivial task. As long as one has a training place, time for said physical activity, and a means to arrive there, there isn’t much work to be done. However, with fitness becoming so popular, forming groups to exercise together becomes more and more popular as well. Hence, the real problem arises when the number of potential participants increases. Finding an appropriate time and place to train becomes exponentially difficult with every new member interested in joining the group. This can be the case, due to multiple reasons. In modern days, most people have multiple things, “to which they must attend”. Moreover, these events are, often, daily occurrences. From educational institutes to jobs, whether it’s a typical nine to five job, or something with a more unorthodox schedule, it is highly unlikely that ones personal schedule matches well with schedules of friends, with whom (s)he may want to work out. This does not even include situations that occur unexpectedly. © Springer Nature Singapore Pte Ltd. 2019 M. Prateek et al. (Eds.): NGCT 2018, CCIS 922, pp. 364–375, 2019. https://doi.org/10.1007/978-981-15-1718-1_31

Software Agents in Support of Scheduling Group Training

365

This trend has contributed to the rise in popularity of scheduling applications such as Doodle or Appointy. To schedule an event “is first and foremost about finding a date and/or time. Doodle supports this by recognizing times and time spans, when creating a poll and by calendar import/export features, time zones, etc. when participating in one” [1]. Although Doodle allows users to vote on presented options, it does not account for preferences each user might have, nor does it have the important information about the user, or the event. Instead, it leaves the user no choice but to do the research her (him)self. Appointy [2], is a more business-oriented application, which allows users to schedule and to accept appointments with ease. It also offers many other services, such as sending reminders through email or SMS and its integration with social media makes it a very flexible tool. However, similarly to Doodle, it is not designed for unpredictable scenarios, and it does not have the “intelligence” to carry out the decision for the user. Here, let us note, that people have different preferences, e.g. time of the day when they would like to exercise, or whether the place should carry a late-evening football broadcasts. Attempting at satisfying such preferences can lead to direct conflicts. The aforementioned applications, and many similar ones, try to resolve these problems by having a vote – where individual preferences can be “represented through casting a vote”. Hence, they try to reach consensus by using a “democratic approach”. However, they require that users actively take place in “negotiations” (at least, by actually placing a vote). Here, let us note that daily schedules are more chaotic than well-structured (unlike meetings in a business environment), which can make involvement of democracy somewhat tedious and inefficient, when the number of preferences to be taken into account increases. The question thus arises – is it possible to reduce, if not eliminate, direct human participation in scheduling fitness session for a group of friends? The aim of our work was to show that this can be done (at least to some extent). Specifically, with enough preferences given, the proposed approach is expected to arrive at a conclusion that can be interpreted as a compromise between the preferences of various people without leaving anyone too upset. Let us now assume that the main use case that we are interested in, is as follows. A group of “busy friends” likes to exercise together. As many people today, they live their lives “according to a calendaring application”. In other words, we assume that, outside of extraordinary circumstances, all of their up-coming appointments/meetings are stored in such an application. However, we do not assume that the same calendaring application has to be used by each of them. Separately, we assume that they agree to trust the fitness scheduling application (that we describe in what follows). In other words, they let it facilitate group negotiations and specify place and time of an upcoming fitness session. More specifically, let us imagine that a group of three friends decides to have a training session together. They all work from Monday to Friday until 4 p.m., and one of them has a boxing session every Tuesday evening. Then, the program will try to book a session sometime after 4 p.m. on Monday, Wednesday, Thursday or Friday as, obviously, one of the members will not be able to join on Tuesday, due to the boxing class. Here, the exact time is to be decided by finding one matching preferred training time of all users. Moreover, if the application fails to find a suitable time during the week, in a more trivial case, it will attempt to book a training session during the weekend when the users, most-likely, have more free time.

366

G. Mamatsashvili et al.

Because of the nature of the problem, software agents seem to be the preferable resolution. Here, note that, according to classical references (e.g. [3, 4]) software agents are very good in negotiating and re-negotiating contracts/schedules/agreements/etc. Furthermore, they can act independently [5], while representing their “owners”. Our work is meant to illustrate, in the specific context, usability of software agents in finding consensus, while solving a, relatively simple, scheduling problem. Here, agents representing users need to be provided with the needed information (representing user preferences), and then they will communicate to establish the “consensus representing place and time”. Furthermore, it is assumed that if any of the preferences change, agents will adapt to the situation, and renegotiate the schedule, to reach the desirable outcome. To this effect we proceed as follows. In Sect. 2, we summarize some related work. We follow, in Sect. 3 with outline of the proposed approach. Next, in Sect. 4 we show how the implemented application works to solve the, above outlined, problem.

2 Related Work Agents are very versatile and can have many different applications. They can be trusted with different tasks, if they’re configured properly and have enough information, based on which they can represent interests of the user. If an agent has enough information, it can be expected to make a decision that is very similar to the one that would be made by the person it represents. Agents, of course, don’t necessarily have to represent people; given enough data, they will make choices that are appropriate in the given scenario [6]. Due to this fact, agents are often tasked with decision-making [7]. In this context, due to their independence and adaptiveness, they are capable of helping with the scheduling and rescheduling of certain events. It is not an uncommon practice to use agents for scheduling [8]. Agents are expected not only to formulate the original schedule, but also to adapt to the changes, depending on what information is available to them, reevaluate the situation, and adjust the schedule based on the newly obtained data. For example, system mentioned in [9] is used by Taxi companies for real-time vehicle scheduling. Multiagent system, developed by Magenta Corporation for Addison Lee, is one of many examples of this specific type of application for agent systems, in the real world. As stated in the document, with the advancements of the modern age, transportation networks have grown exponentially. The growth was accompanied by its difficulties, new problems that required new solutions. Industry giants such as DHL, UPS, TNT, DPD, and many others, have very large and complex networks, which the enterprise resource planning systems that are in use cannot handle well. This is particularly the case when response is needed to dynamically changing situations. The article claims, that the systems have failed to be very efficient due to the fact, that they’ve failed to keep up with the everevolving technological world and stuck to their older technologies. Solutions to these problems are outdated and limited. Multiagent systems can provide solutions to the problems that may arise with networks such as these. In the case of taxi companies, an agent can consider an order of a taxi at a specific time with a specific request. For example, a group of 7 people with luggage may not fit in a normal taxi car, and the

Software Agents in Support of Scheduling Group Training

367

current city traffic might make it difficult for certain drivers to reach the customer. The handling the order will need to consider all these requirements and match the customer with the most appropriate driver. This can be done by comparing what information the agent holds, concerning both the customer and the driver (and his vehicle). The solution to the problem is very practical as it can help save many resources for the company as well as it is scalable, which can make the concept future-proof in the commercial world. As mentioned before, the multi-agent system solutions appear in multiple contexts of scheduling. For instance, HOLOS multi-agent scheduling system is discussed in [8]. HOLOS was designed specifically for manufacturing enterprises. The system works by configuring a group of agents for a specific shop floor. These agents are tasked with exchanging information about the relative production orders; this information is used by the system to generate schedules. What is interesting about this technology, is that the scheduling system is derived from HOLOS-GA, is not fully automated, but rather semi-automated. Due to many challenges in the field of manufacturing, the system is interactive, it is implemented to assist the expert by automating most of the steps of the HOLOS methodology used by the system. This allows enterprises to tailor the system to their specific needs. Although, the system does not aim to eliminate the human involvement in the scheduling process, it still does a great deal of minimizing the amount of work an expert needs to do, which in turn can be both time and resource efficient. In [10], problems that manufacturing enterprises have faced as the world moves, more and more, towards a global economy are discussed. Due to the high competition in the field, for the enterprises to survive, they have to be flexible and agile. Similar to what was described in [9], many existing systems are outdated and can no longer be considered efficient. They lack many features a modern market requires such as flexibility and re-configurability. Although the systems are explicitly build to optimize production, they fail to respond to change. For that exact reason, the author suggests usage of software agents in the world of manufacturing. Agents are completely autonomous and can be tasked with dealing with the ever changing demands of the industry [11]. Their intelligence, paired with independence, can create an system that is resource-efficient. The author shows an example of an architecture that looks as follows. The agent tasked with supervision of the process, queries for an available agents that represent factory resources. The supervisor agent looks for an agent who possesses the skill that is required to complete the task. Here, each and every resource agent has to verify its skills and answer whether or not they’re capable of performing the task. In the specific example given in the paper, we have agent #1 refusing to do the task as it is out of service, and agent #2 being unable to accept the task as it is overloaded. The task, then, is given to agent #3 who needs to negotiate process of transportation with the agent that is tasked with transportation. In this paper, we can see an example of an multi-agent based infrastructure which schedules and negotiates tasks in a business environment. This system can help with the full automation of the tasks and doesn’t require any human-involvement. All these papers, have not only confirmed correctness of our decision to use software agents, but also provided valuable insights to the system that we have undertaken to develop.

368

G. Mamatsashvili et al.

3 Proposed Approach Taking into account what has been discussed thus far, let us now outline the proposed approach. We will start from the requirements for the considered application. 3.1

Top-Level Requirements and Architecture

As indicated above, we propose an agent-based architecture, in which each user will have a personal agent representing her/his interests. Moreover, each user will have all of her/his upcoming appointments/meetings/commitments stored in a calendaring application. In this way the proposed system will be able to know when the user is busy, and thus unable to exercise. Since we plan to use software agents, each user can easily use a different calendaring application. Each of them will be interfaced with her/his personal agent. In this way, while the calendaring applications will differ, personal agents, which belong to the same agent platform, will be able to communicate and negotiate meeting time. When the system is initialized (for the first time), after authenticating, users will be presented with a user interface where they will be able to specify their preferences. The interface will be split into two main sections, concerning time and the place (the gym). In each of these sections, user will be able to specify her/his ideal training preferences, so that the program will attempt to schedule a training session that isn’t too different from them. The first section, concerning time, will ask for preferred training hours and duration of the session. The second section will focus on the gym itself. The user will have to specify what is important in the gym; e.g. what “equipment” is necessary for them for their workout. They will be also able to suggest what they’d want to see in a gym (e.g. TV’s). Each agent will store this information and use it in the negotiations. This information will be stored within the application, for future use, and will be editable (in case if the preferences change). For a depiction of the interface, see Fig. 1. In the system, one of personal agents will be acting as group leader agent. Group leader is the personal agent that starts negotiations, and knows which other personal agents are to be involved in them. It represents a person who is “organizing the training session”. This person (her/his agent) is the one that starts the “scheduling process”. The role of group leader can be assumed by any personal agent. Negotiations are to be orchestrated by a separate agent. We will refer to it as the central agent. The negotiation process starts when the group leader agent instance is created by one of the users. This user also states who else should be in the group that will go out. This will result in instantiation of personal agents representing each user. The personal agents, after their creation, send, as a message, “their” preferences (retrieved from the preference repository) to the central agent. The central agent receives preferences from all personal agents representing users that are expected to exercise together. It also knows, from the group leader, how many/which personal agents are to send such preferences. The central agent waits for a limited time (which is a parameter of the system) for preferences to arrive. In case when some agents do not send preferences (e.g. when they are placed within mobile devices that are turned off/not connected to the Internet) in time, they are eliminated from the pool and will not take part in scheduling.

Software Agents in Support of Scheduling Group Training

369

Afterwards, the central agent will make decision, based on received preferences and sends messages back to the personal agents so that they can add the event to calendars of their users. The decision the central agent makes, is to represent a compromise between what all the users want. The decision making process is described in some detail in Sects. 3.2.1 and 3.2.2. 3.2

Technical Aspects

Let us now briefly describe key technical aspects of the developed application. Let us start from the way of dealing with time and activities stored in a calendaring application. Here, the schedule of the user has to be extracted from her/his calendar. As mentioned, thanks to use of agent infrastructure, we can “hide” calendaring applications behind personal agents, which have to “know” how to interface with their calendars. However, for the initial prototype we have selected the Google Calendar. Hence, we access user data using the Google Calendar API [12]. Note that, in case of other calendaring applications, we would use their respective interfaces. For instance One Calendar [13], Fantastical 2 [14], Lightning [15] (and many others) provide interfaces that allow easy integration with external software. In our program, we use the following method, which facilitates access all the user-events, in a specified time period (period that is pertinent to scheduling joint exercise). p u b l i c L i s t getEventsBetween ( DateTime from , DateTime to , S t r i n g u s e r I d ) throws IOException { CalendarConsumer consumer = ( e n t r y , c a l e n d a r , r e s u l t s ) −> { Events e v e n t s = c a l e n d a r . e v e n t s ( ) . l i s t ( entry . getId ( ) ) . s e t D e f a u l t B e t w e e n C r i t e r i a ( from , t o ) . e x e c u t e ( ) ; r e s u l t s . addAll ( events . getItems ( ) ) ; }; r e t u r n a b s t r a c t G e t ( consumer , u s e r I d ) ; }

Obviously, data can be read only after the user gives the program permission to access her/his calendar data. This method utilizes Google Calendar API to fetch the events from the user’s calendar. More specifically, we are explicitly interested in the upcoming week, when the program will attempts at scheduling a session and needs to know exactly what events the user might have within this time frame. We have limited our time-horizon to only one week as the purpose of the program is to attempt to find free time for a training session in the “near future”. As it can be seen in the code snippet, it is possible to fetch events from the user’s calendar from any specified time period. Although, the program, in its current form, is interested in the upcoming week only, this method is flexible enough to support different needs and ideas concerning calendar access. Once the application has access to the events from the calendar, it is time for the agents communicate. In our application, we have selected JADE (Java Agent

370

G. Mamatsashvili et al.

Development Framework), developed by Telecom Italia. JADE is considered to be one of the most advanced open source agent frameworks available. Apart from agent abstraction, JADE provides powerful task execution, peer to peer agent communication via asynchronous messages and many other features [16]. With the help of JADE, we start by creating the central agent, tasked with collecting users preferences. The central agent is the first agent created during the scheduling process. To create the agent, a createNewAgent method, found in the AgentController interface is called. When the agent has started, its state transitions from “INITIATED” to “ACTIVE”, and the controller is returned. This method is used to create every agent in the application. For every user, a personal agent is created. When one of the users is ready to start forming a group, its agent becomes a group leader and invites selected personal agents to participate in training scheduling. It also informs the central agent, which agents are in the group that is to be formed. While this may seem somewhat strange, as one could suggest that the group leader should be the one to play the role of central agent, this approach was not selected. Here, this would mean that each personal agent would have to have code for schedule negotiations. Furthermore, each one of then could have had “access” to other persons calendars (via messages that it would receive from the other personal agents). The latter is a very dubious choice (from the privacy perspective); even if the access would be only to a limited data-set. On the other hand, running a central agent in a secure and trusted environment and assuring users that their data will be deleted after being used, is more palatable. Here, we also have to distinguish between the current application setup, applied in the initial prototype, and the way that the program would have worked in real-life mobile scenario. Currently, all agents are created within a computer that emulates multi-user application. This was done to test the basic logic of preference representation, calendaring application access and, finally, the process of scheduling. In the case when the application was to run in a mobile world, agents would have been instantiated on smart phones (mobile devices, in general). This can be done, as in the most recent version of JADE it is possible to instantiate separate platforms on mobile devices and facilitate their communication. Hence, while currently running only a limited emulation, choice of agent platform assures that real-world deployment is possible. Personal agents provide the central agent with preferences of users they represent. To store user preferences, we use JpaRespository [17]. With the help of H2 database [20], we can persist the data in a local file stored on the machine (in the case of a mobile application, each agent would store data in a local instance of the database). Pertinent user preferences (we assume that there may be user preferences that are not related to a given scheduling session) are extracted form the database, and turned into string (by software called by the personal agent). If the agent fails to start for whatever reason, the error is logged. Next an ACL REQUEST message (see, [18]) is sent to the central agent. This message contains user preferences (in JSON format). When initiated, the central agent starts and is continuously, through CyclicBehaviour, waiting for messages from personal agents. For each REQUEST message, it tries to extract its content (with errors being logged). Once all expected messages are received (or the time-out happens), preferences extracted form JSON content are stored

Software Agents in Support of Scheduling Group Training

371

as objects and the scheduling algorithm is called. Let us now consider, in some detail, the way that user preferences are taken into account. 3.2.1 Gym Preferences Initially, during the selection process, the program tries to select the gym, based on user preferences. All gyms are stored in the application as objects and the information regarding them are stored as fields. Specifically, this is a single list, which is constant. Although this isn’t an efficient way of providing the application with information about gyms, this is a temporary solution. For the time being, data is simply mocked, while in the future stages, we could expect different solutions to the problem, such as a database containing all the necessary information about gyms. In the latter case, a separate interface to add/remove/modify gym information (manually or by extracting information from the Web) would have to be provided. During the gym selection process, based on user preferences, which are represented in form of weighs, gyms are scored depending on what they have to offer. There’s a minimum amount of points that is required to score by a given gym, for it to be considered further. This is a system parameter and it allows to reduce the number of gyms that are going to be taken into account. Currently, if any of the gyms scores below two points for any user, it will no longer be considered by the algorithm regardless of how high it’s score by other users might be. Currently available preferences can be seen on the right side in Fig. 1. Each preference is worth a specific number of points depending on their importance (again, a system parameter that can be adjusted). In the current version, the user is unable to express the importance of each preference and, instead, they are hardcoded in the program (obviously, this is only a temporary limitation to simplify the prototype). To determine the score of the gym, the program checks whether or not the gym can offer what the user is looking for (expressed in preferences). For every preference the gym can satisfy, the number of points that this preference was worth is added to the gym score, which is initially zero. Eventually, all the individual scored are summed up and the gym with the highest overall score is chosen. Currently, all the gyms along with the information regarding them, are also hardcoded in the program. This is subject to change as the program evolves. 3.2.2 Time Preferences Afterwards, the algorithm for the time selection begins. Firstly, all available times, during the upcoming week, are extracted from the calendar. After acquiring such data, the algorithm starts working to find the shared free time. For each user, the algorithm focuses on the preferred starting time of the workout and the preferred ending time of the workout. This is taken from the user inputs “Workout start lower bound” and “Workout start upper bound” that can be seen in Fig. 1. The algorithm then tries to choose a time for the workout from the shared preferred hours it just obtained, and if the chosen time happens to be available in all users’ schedules, the session can be scheduled. In other words, the program will only attempt to schedule the session in the shared preferred time. This way, the chance of the session

372

G. Mamatsashvili et al.

ending up scheduled either too early or too late is eliminated. From this time interval, the program checks different hours and compares it to the calendars of the users. If the program can find a time period where every user is free, then it will select that time for scheduling. In summary, the program ends up booking the session at a time which is relatively close to every user’s preference. Hence, it can be claimed that it arrives at a compromise. Afterwards, the central agent sends the message with performative AGREE, meaning it has agreed to do the request, to personal agents and it will send back time of the training session as a content of the message. In the final step, each personal agent calls a method, which adds the event to the calendar of the its, respective, user.

4 Experimental Verification We have implemented the initial prototype of the proposed application and tested it on a number of scenarios. As noted, the application has been run on a single computer, to test the main mechanisms (data representation, access, communication, etc.). Furthermore, a number of aspects of preference representation have been hard-coded to simplify the prototype. Here, let us report on two basic use cases. First, what happens when we try to schedule a simple training session and, second, what will happen when a problem occurs. 4.1

Successful Training Session Scheduling

In Fig. 1, we can see how the preference selection looks like. The preferences concerning time occupy the left half of the interface, while the ones concerning the gym can be seen on the right. The preferences shown in the picture are used as an example and are likely to be adjusted, as the application is being further developed. In the presented scenario, we have two users looking to book a training session (which is much easier to describe in a legible way). However, we have tested the code for multiple users, as well. Each user sets up a their preferences for the program. The first user chooses squat rack, yoga room and beactive membership support, as the gym preferences. The second user chooses squat rack and power bikes. All the gyms that are stored (hard-coded) in the application are considered. Here, three gyms that stand out the most and they are “Ancient”, “Olympus” and “Palace”, since the three of them satisfy more of the userspecified preferences than others. Olympus offers squat rack and a yoga room, and even supports the beactive membership, which leaves it with a score of 6 for the first user, and a score of 3 for the second user, therefore the overall score of the Olympus is 9. Palace, on the other hand, does not have a yoga room, but instead has a room with power bikes, which leaves it with an overall score of 9. Ancient offers everything users specified in their preferences and with the overall score of 10, it is chosen. It is worth noting that upcoming schedules of both users are relatively free, which gives the program a lot of different possibilities, as to when the training session can be scheduled.

Software Agents in Support of Scheduling Group Training

373

The program always chooses time period that is the closest to the “current moment”. In this particular case, in Fig. 2, we can see the upcoming gym event added from 18:00 to 20:00 as both involved users had set 2 h as a preferred training duration. The fact that the session will be held in the “Ancient” gym is also included in the description of the event. 4.2

Scheduling Failure

Recall that the program makes a decision based on what can be found in user’s calendars. If we have a scenario where one or more users’ schedule is full, the program may fail to schedule a training session. Unfortunately, the program cannot take care of such a problem as it tries to consider schedules of all the users, and every time period where there is an event scheduled automatically gets disqualified and will not be used for scheduling the training session. Although a possibility of something like this happening isn’t very high, it becomes more likely as the groups get bigger. Let us show a specific example and see how the failure is illustrated. In this experiment, we will, once again, have two users attempt to book a training schedule. The first user species that the time he’s willing to workout is from 9:00 to 12:00, while the second user’s choice is from 13:00 to 18:00. Since there is absolutely no overlapping between the two preferences, the program fails to schedule an event and the error is logged. 2018–10–21 19:32:56.618 ERROR 10896 —

Fig. 1. Example of an user going through their setup

374

G. Mamatsashvili et al.

Fig. 2. The program is done scheduling the session, and the event is added to the calendars of the users [ central ] p . g . jacked . s e r v i c e s . CalendarService S c h e d u l i n g F a i l u r e − Free time not found

:

The central agent sends back the message of performative FAILURE to individual agents to notify them, that the action failed. Note that for a group of, let us say, 9 people with busy schedules, it is possible for them not to have an overlapping time period when they all happen to be free in the upcoming week. Moreover, users may not be willing to wait an entire week for the workout. Here a number of scenarios is possible. (1) Program may fail and report this to the users (via their personal agents). In this case, which is the one currently implemented, it is advised that the program is only used by small groups of people, no more than 3 or 4 persons in each group. This way, we can maximize the chance of a desirable outcome. (2) Program may try to eliminate the most “problematic user(s)” and schedule the largest number of them during the closest time. Here, a multicriterial decision making process is needed. In this process number of people that can exercise together will be considered vis-a-vis other criteria. We plan to investigate this approach in the near future. Overall, it should also be noted that the decision carried out by the program isn’t perfect. The application considers preferences of all users, which might cause it to make a decision that is more likable to some than others. This will depend entirely on the algorithm in place.

5 Concluding Remarks In this work we have considered use of software agent infrastructure to schedule group exercises in a gym. Based on requirements analysis we have implemented an initial prototype and tested it in a number of scenarios. The developed application schedules a

Software Agents in Support of Scheduling Group Training

375

training session based on the needs and availability of the user. If the user is willing to trust the decision-making process of the program, then the human involvement can be minimized, when trying to find time for joint activities. In the text, above, we have indicated a number of immediate shortcomings of the developed prototype. These will be dealt with, first. Furthermore, in order to achieve better quality of scheduling, it is planned to perform an additional research concerning what is available in gyms and what can/should be represented in user preferences. Moreover, possibilities of, for instance: (1) migrating the application to mobile devices, (2) integrating with personal assistants like Alexa, Cortana or Google Assistant, (3) avoiding scheduling a session at a specific time period due to blacklisted events (see, [19]), (4) use of ontologies to represent gyms and user preferences, and semantic technologies, in general, (5) inclusion of diet control, are going to be considered to develop a complete fitness assistant. We will report on our progress in subsequent publications.

References 1. Doodle. https://help.doodle.com/customer/portal/articles/645363. Accessed 24 Sept 2018 2. Appointy. https://www.appointy.com. Accessed 24 Sept 2018 3. Hayzelden, A.L.G., Bigham, J.: Software Agents for Future Communication Systems. Springer, Heidelberg (1999). https://doi.org/10.1007/978-3-642-58418-3 4. Huhns, M.N., Singh, M.P.: Readings in Agents. Morgan Kaufmann, San Francisco (1998) 5. Lewis, F.L., Zhang, H., Hengster-Movric, K., Das, A.: Cooperative Control of Multi-Agent Systems: Optimal and Adaptive Design Ap-proaches. Springer, London (2014). https://doi. org/10.1007/978-1-4471-5574-4 6. Rabuzin, K., Malekovic, M., Baca, M.: A survey of the properties of agents, December 2005 7. Duan, Y., Ong, V.K., Xu, M., Mathews, B.: Supporting decision making process with “ideal” software agents–what do business executives want? Expert Syst. Appl. 39(5), 5534– 5547 (2012) 8. Rabelo, R.J., Camarinha-Matos, L.M., Afsarmanesh, H.: Multi-agent-based agile scheduling. Robot. Auton. Syst. 27, 15–28 (1999) 9. Glaschenko, A., Ivaschenko, Rzevski, G., Skobelev, P.: Multi-agent real time scheduling system for taxi companies. In: Proceedings of the 8th International Conference on Autonomous Agents and Multiagent Systems, pp. 29–36, May 2009 10. Verma, V.K.: Multi-agent based scheduling in manufacturing system. In: National Conference on Futuristic Approaches in Civil and Mechanical Engineering, March 2015 11. Leitao, P., Karnouskos, S.: Industrial Agents: Emerging Applications of Software Agents in Industry. Elsevier, Amsterdam (2015) 12. Google Calendar API. https://developers.google.com/calendar/. Accessed 02 Sept 2018 13. OneCalendar. https://www.onecalendar.nl/onecalendar/overview. Accessed 29 Nov 2018 14. Fantastical 2. https://flexibits.com/fantastical. Accessed 29 Oct 2018 15. Lightning Calendar. https://www.thunderbird.net/en-US/calendar/. Accessed 29 Oct 2018 16. JADE: Java agent development framework. http://jade.tilab.com/. Accessed 05 Sept 2018 17. JPA repository. https://docs.spring.io/spring-data/jpa/docs/current/api/org/springframework/ data/jpa/repository/JpaRepository.html. Accessed 05 Nov 2018 18. FIPA Peformatives. http://jmvidal.cse.sc.edu/talks/agentcommunication/performatives.html? style=White. Accessed 5 Sept 2018 19. Wordnet Database. https://wordnet.princeton.edu/. Accessed 24 Sept 2018 20. H2 Database. https://www.h2database.com/. Accessed 05 Nov 2018. Accessed 24 Sept 2018

Author Index

Adamuthe, Amol C. 340 Aggarwal, Rajeev 296 Agrawal, Saurabh 102 Allayear, Shaikh Muhammad Atri, Swati 321 Banga, Vijay Kumar 351 Bhowmik, Arka 281 Dave, Meenu 330 Dinanath 183 Doriya, Rajesh 64 Dubey, Ritika 152 Gancarz, Konrad 364 Ganesan, P. 92 Ganzha, Maria 364 Garg, Sanjay 140 Gupta, Divya 26 Harariya, Narendra

281

Jadhav, Akshayya 340 Jain, Saurabh 64 Kad, Sandeep 351 Khanna, Radhesh 239 Khare, Vijay 152 Khyatee 296 Kohli, Narendra 221 Korabu, Kirti 259 Kumar, Narander 311 Kumar, Naveen 183 Kumar, Ravin 163 Kumar, Sanjay 281 Kumar, Yugal 13 Kumari, Meena 116 Łajewska, Weronika

364

Mahalakshmi, G. S. 77 Mamatsashvili, Giorgi 364 Mankad, Sapan H. 140

Mehndiratta, Pulkit 49 Mehta, Rajesh 172 Munna, Md Tahsir Ahmed

38

38 Nagwani, Naresh Kumar Narang, Aditi 152

102

Pandey, M. K. 221 Paprzycki, Marcin 364 Potdar, Sonali 248 Prakash, Rishi 3 Prasad, Ritu 199 Puri, Vishal 248, 259 Rahman, Md. Habibur 38 Rahman, Md. Mushfiqur 38 Rahman, Sheikh Shah Mohammad Motiur 38 Rajpal, Navin 172 Ramesh Babu, A. 248, 259 Ranganayakulu, A. 92 Rani, Kritika 210 Rani, Manju 183 Ranjan, Rakesh 116 Rao, S. Jagan Mohan 92 Samriya, Jitendra Kumar 311 Sanjay, S. Dola 92 Sarkar, Aparna 296 Sathish, B. S. 92 Saurabh, Praneet 199 Saxena, Archana B. 330 Saxena, Kanak 26 Shakya, Amit Kumar 3 Sharma, Shashank 239 Shukla, Anurag 269 Siddique, Bushra 128 Singh, Pradeep Kumar 13 Singh, V. R. 116 Sisodia, Dilip Singh 102 Sohan, Md Fahimuzzman 38 Soni, Devpriya 49 Soni, Pramod Kumar 172 Sreeja, P. S. 77

378

Author Index

Srivastava, Anand Kumar 13 Srivastava, Rajshree 210 Sudalai Muthu T. 248, 259 Sufyan Beg, Mirza Mohd 128 Tiwari, Shalini 3 Tripathi, Sarsij 269 Tyagi, Sanjay 321

Upadhyay, Pragati

221

Verma, Bhupendra 199 Verma, Jai Prakash 140 Vidyarthi, Anurag 3 Yadav, Preeti 199