128 52 63MB
English Pages 611 [609] Year 2022
Lecture Notes in Networks and Systems 497
Leonard Barolli Editor
Complex, Intelligent and Software Intensive Systems Proceedings of the 16th International Conference on Complex, Intelligent and Software Intensive Systems (CISIS-2022)
Lecture Notes in Networks and Systems Volume 497
Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Advisory Editors Fernando Gomide, Department of Computer Engineering and Automation—DCA, School of Electrical and Computer Engineering—FEEC, University of Campinas— UNICAMP, São Paulo, Brazil Okyay Kaynak, Department of Electrical and Electronic Engineering, Bogazici University, Istanbul, Turkey Derong Liu, Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, USA Institute of Automation, Chinese Academy of Sciences, Beijing, China Witold Pedrycz, Department of Electrical and Computer Engineering, University of Alberta, Alberta, Canada Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Marios M. Polycarpou, Department of Electrical and Computer Engineering, KIOS Research Center for Intelligent Systems and Networks, University of Cyprus, Nicosia, Cyprus Imre J. Rudas, Óbuda University, Budapest, Hungary Jun Wang, Department of Computer Science, City University of Hong Kong, Kowloon, Hong Kong
The series “Lecture Notes in Networks and Systems” publishes the latest developments in Networks and Systems—quickly, informally and with high quality. Original research reported in proceedings and post-proceedings represents the core of LNNS. Volumes published in LNNS embrace all aspects and subfields of, as well as new challenges in, Networks and Systems. The series contains proceedings and edited volumes in systems and networks, spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor Networks, Control Systems, Energy Systems, Automotive Systems, Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems, Robotics, Social Systems, Economic Systems and other. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution and exposure which enable both a wide and rapid dissemination of research output. The series covers the theory, applications, and perspectives on the state of the art and future developments relevant to systems and networks, decision making, control, complex processes and related areas, as embedded in the fields of interdisciplinary and applied sciences, engineering, computer science, physics, economics, social, and life sciences, as well as the paradigms and methodologies behind them. Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago. All books published in the series are submitted for consideration in Web of Science. For proposals from Asia please contact Aninda Bose ([email protected]).
More information about this series at https://link.springer.com/bookseries/15179
Leonard Barolli Editor
Complex, Intelligent and Software Intensive Systems Proceedings of the 16th International Conference on Complex, Intelligent and Software Intensive Systems (CISIS-2022)
123
Editor Leonard Barolli Department of Information and Communication Engineering Fukuoka Institute of Technology Fukuoka, Japan
ISSN 2367-3370 ISSN 2367-3389 (electronic) Lecture Notes in Networks and Systems ISBN 978-3-031-08811-7 ISBN 978-3-031-08812-4 (eBook) https://doi.org/10.1007/978-3-031-08812-4 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Welcome Message of CISIS-2022 International Conference Organizers
Welcome to the 16th International Conference on Complex, Intelligent and Software Intensive Systems (CISIS-2022), which will be held from June 29 to July 1, 2022, in conjunction with the 16th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS-2022). The aim of the conference is to deliver a platform of scientific interaction between the three interwoven challenging areas of research and development of future ICT-enabled applications: software-intensive systems, complex systems and intelligent systems. Software-intensive systems are systems, which heavily interact with other systems, sensors, actuators, devices, other software systems and users. More and more domains are involved with software-intensive systems, e.g., automotive, telecommunication systems, embedded systems in general, industrial automation systems and business applications. Moreover, the outcome of web services delivers a new platform for enabling software-intensive systems. The conference is thus focused on tools, practically relevant and theoretical foundations for engineering software-intensive systems. Complex systems research is focused on the overall understanding of systems rather than its components. Complex systems are very much characterized by the changing environments in which they act by their multiple internal and external interactions. They evolve and adapt through internal and external dynamic interactions. The development of intelligent systems and agents, which is each time more characterized by the use of ontologies and their logical foundations, build a fruitful impulse for both software-intensive systems and complex systems. Recent research in the field of intelligent systems, robotics, neuroscience, artificial intelligence and cognitive sciences is very important factor for the future development and innovation of software-intensive and complex systems. This conference is aiming at delivering a forum for in-depth scientific discussions among the three communities. The papers included in the proceedings cover all aspects of theory, design and application of complex systems, intelligent systems and software-intensive systems. v
vi
Welcome Message of CISIS-2022 International Conference Organizers
We are very proud and honored to have two distinguished keynote talks by Prof. Keita Matsuo, Fukuoka Institute of Technology, Japan, and Dr. Anne Kayem, Hasso Plattner Institute, University of Potsdam, Germany, who will present their recent work and will give new insights and ideas to the conference participants. The organization of an international conference requires the support and help of many people. A lot of people have helped and worked hard to produce a successful technical program and conference proceedings. First, we would like to thank all authors for submitting their papers, the program committee members, and the reviewers who carried out the most difficult work by carefully evaluating the submitted papers. We are grateful to Honorary Chair Prof. Makoto Takizawa, Hosei University, Japan, for his guidance and support. Finally, we would like to thank Web Administrator Co-chairs for their excellent and timely work. We hope you will enjoy the conference proceedings.
CISIS-2022 Organizing Committee
Honorary Chair Makoto Takizawa
Hosei University, Japan
General Co-chairs Tomoya Enokido Marek Ogiela
Rissho University, Japan AGH University of Science and Technology, Poland
Program Committee Co-chairs Keita Matsuo Antonio Esposito Omar Hussain
Fukuoka Institute of Technology, Japan University of Campania “Luigi Vanvitelli”, Italy University of New South Wales, Australia
International Advisory Board David Taniar Minoru Uehara Arjan Durresi Beniamino Di Martino
Monash University, Australia Toyo University, Japan IUPUI, USA University of Campania “Luigi Vanvitelli”, Italy
Award Co-chairs Akio Koyama Kin Fun Li Olivier Terzo
Yamagata University, Japan University of Victoria, Canada LINKS Foundation, Italy
vii
viii
CISIS-2022 Organizing Committee
International Liaison Co-chairs Wenny Rahayu Fumiaki Sato Flora Amato
La Trobe University, Australia Toho University, Japan University of Naples Frederico II, Italy
Publicity Co-chairs Nadeem Javaid Takahiro Uchiya Markus Aleksy Farookh Hussain
COMSATS University Islamabad, Pakistan Nagoya Institute of Technology, Japan ABB AG Corporate Research Center, Germany University of Technology Sydney, Australia
Finance Chair Makoto Ikeda
Fukuoka Institute of Technology, Japan
Local Arrangement Co-chairs Tomoyuki Ishida Kevin Bylykbashi
Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan
Web Administrator Chairs Phudit Ampririt Ermioni Qafzezi
Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan
Steering Committee Chair Leonard Barolli
Fukuoka Institute of Technology, Japan
Track Areas and PC Members 1. Database and Data Mining Applications Track Co-chairs Kin Fun Li Pavel Krömer
University of Victoria, Canada Technical University of Ostrava, Czech Republic
PC Members Antonio Attanasio
Links Foundation, Italy
CISIS-2022 Organizing Committee
Tibebe Beshah Jana Heckenbergerova Konrad Jackowski Petr Musílek Aleš Zamuda Genoveva Vargas-Solar Xiaolan Sha Kosuke Takano Masahiro Ito Watheq ElKharashi Mohamed Elhaddad Wei Lu
ix
Addis Ababa University, Ethiopia University of Pardubice, Czech Republic Wroclaw University of Technology, Poland University of Alberta, Canada University of Maribor, Slovenia French Council of Scientific Research, LIG-LAFMIA, France Sky, UK Kanagawa Institute of Technology, Japan Toshiba Lab, Japan Ain Shams University, Egypt University of Victoria, Canada Keene State College, USA
2. Artificial Intelligence and Bio-Inspired Computing Track Co-chairs Hai Dong Salvatore Vitabile Urszula Ogiela
Royal Melbourne Institute of Technology, Australia University of Palermo, Italy AGH University of Science and Technology, Poland
PC Members Kit Yan Chan Shang-Pin Ma Pengcheng Zhang Le Sun Sajib Mistry Carmelo Militello Klodiana Goga Vincenzo Conti Minoru Uehara Philip Moore Mauro Migliardi Dario Bonino Andrea Tettamanzi Cornelius Weber Tim Niesen Rocco Raso Fulvio Corno
Curtin University, Australia National Taiwan Ocean University, Taiwan Hohai University, China Nanjing University of Information Science and Technology, China Curtin University, Australia Italian National Research Council, Italy Links Foundation, Italy University of Enna Kore, Italy Toyo University, Japan Lanzhou University, China University of Padua, Italy CHILI, Italy University of Nice, France Hamburg University, Germany German Research Center for Artificial Intelligence (DFKI), Germany German Research Center for Artificial Intelligence (DFKI), Germany Politecnico di Torino, Italy
x
CISIS-2022 Organizing Committee
3. Multimedia Systems and Virtual Reality Track Co-chairs Yoshinari Nomura Francesco Orciuoli Shinji Sugawara
Okayama University, Japan University of Salerno, Italy Chiba Institute of Technology, Japan
PC Members Shunsuke Mihara Shunsuke Oshima Yuuichi Teranishi Kazunori Ueda Hideaki Yanagisawa Kaoru Sugita Keita Matsuo Santi Caballé Nobuo Funabiki Yoshihiro Okada Tomoyuki Ishida Nicola Capuano Jordi Conesa Farzin Asadi David Gañan Le Hoang Son Jorge Miguel David Newell
Lockon Inc., Japan Kumamoto National College of Technology, Japan NICT, Japan Kochi University of Technology, Japan National Institute of Technology, Tokuyama College, Japan Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan Open University of Catalonia, Spain Okayama University, Japan Kyushu University, Japan Fukuoka Institute of Technology, Japan University of Basilicata, Italy Universitat Oberta de Catalunya, Spain Kocaeli University, Kocaeli, Turkey Universitat Oberta de Catalunya, Spain Vietnam National University, Vietnam Grupo San Valero, Spain Bournemouth University, UK
4. Next Generation Wireless Networks Track Co-chairs Marek Bolanowski Sriram Chellappan Kevin Bylykbashi
Rzeszow University of Technology, Poland Missouri University of Science and Technology, USA Fukuoka Institute of Technology, Japan
PC Members Yunfei Chen Elis Kulla Santi Caballé
University of Warwick, UK Fukuoka Institute of Technology, Japan Open University of Catalonia, Spain
CISIS-2022 Organizing Committee
Admir Barolli Makoto Ikeda Keita Matsuo Shinji Sakamoto Omer Wagar Zhibin Xie Jun Wang Vamsi Paruchuri Arjan Durresi Bhed Bista Tadeusz Czachórski Andrzej Paszkiewicz
xi
Aleksander Moisiu University, Albania Fukuoka Institute of Technology, Japan Fukuoka Institute of Technology, Japan Kanazawa Institute of Technology, Japan University of Engineering & Technology, Poland Jiangsu University of Science and Technology, China Nanjing University of Post and Telecommunication, China University of Central Arkansas, USA IUPUI, USA Iwate Prefectural University, Japan Polish Academy of Sciences, Poland Rzeszow University of Technology, Poland
5. Semantic Web and Web Services Track Co-chairs Antonio Messina Aneta Poniszewska-Maranda Salvatore Venticinque
Istituto di Calcolo e Reti ad Alte Prestazione CNR, Italy Lodz University of Technology, Poland University of Campania “Luigi Vanvitelli”, Italy
PC Members Alba Amato Nik Bessis Robert Bestak Ivan Demydov Marouane El Mabrouk Corinna Engelhardt-Nowitzki Michal Gregus Jozef Juhar Nikolay Kazantsev Manuele Kirsch Pinheiro Cristian Lai Michele Melchiori Giovanni Merlino Kamal Bashah Nor Shahniza Eric Pardede Pethuru Raj
Italian National Research Center (CNR), Italy Edge Hill University, UK Czech Technical University in Prague, Czech Republic Lviv Polytechnic National University, Ukraine Abdelmalek Essaadi University, Morocco University of Applied Sciences, Austria Comenius University in Bratislava, Slovakia Technical University of Košice, Slovakia National Research University Higher School of Economics, Russia Université Paris 1 Panthéon Sorbonne, France CRS4 Center for Advanced Studies, Research and Development in Sardinia, Italy Universita’ degli Studi di Brescia, Italy University of Messina, Italy University Technology MARA, Malaysia La Trobe University, Australia IBM Global Cloud Center of Excellence, India
xii
Jose Luis Vazquez Avila Anna Derezinska
CISIS-2022 Organizing Committee
University of Quintana Roo, México Warsaw University of Technology, Poland
6. Security and Trusted Computing Track Co-chairs Hiroaki Kikuchi Omar Khadeer Hussain Lidia Fotia
Meiji University, Japan University of New South Wales (UNSW) Canberra, Australia University of Calabria, Italy
PC Members Saqib Ali Zia Rehman Morteza Saberi Sazia Parvin Farookh Hussain Walayat Hussain Sabu Thampi
Sun Jingtao Anitta Patience Namanya Smita Rai Abhishek Saxena Ilias K. Savvas Fabrizio Messina Domenico Rosaci Alessandra De Benedictis
Sultan Qaboos University, Oman COMSATS Institute of Information Technology (CIIT), Pakistan UNSW Canberra, Australia UNSW Canberra, Australia University of Technology Sydney, Australia University of Technology Sydney, Australia Indian Institute of Information Technology and Management-Kerala (IIITM-K) Technopark Campus, India National Institute of Informatics, Japan University of Bradford, UK Uttarakhand Board of Technical Education Roorkee, India American Tower Corporation Limited, India University of Thessaly, Greece University of Catania, Italy University Mediterranea of Reggio Calabria University of Naples “Frederico II” Italy
7. HPC & Cloud Computing Services and Orchestration Tools Track Co-chairs Olivier Terzo Jan Martinovič
Jose Luis Vazquez-Poletti
Links Foundation, Italy IT4Innovations National Supercomputing Center, VSB Technical University of Ostrava, Czech Republic Universidad Complutense de Madrid, Spain
CISIS-2022 Organizing Committee
xiii
PC Members Alberto Scionti Antonio Attanasio Jan Platos Rustem Dautov Giovanni Merlino Francesco Longo Dario Bruneo Nik Bessis MingXue Wang Luciano Gaido Giacinto Donvito Andrea Tosatto Mario Cannataro Agustin C. Caminero Dana Petcu Marcin Paprzycki Rafael Tolosana
Links Foundation, Italy Links Foundation, Italy VŠB-Technical University of Ostrava, Czech Republic Kazan Federal University, Russia University of Messina, Italy University of Messina, Italy University of Messina, Italy Edge Hill University, UK Ericsson, Ireland Istituto Nazionale di Fisica Nucleare (INFN), Italy Istituto Nazionale di Fisica Nucleare (INFN), Italy Open-Xchange, Germany University “Magna Græcia” of Catanzaro, Italy Universidad Nacional de Educación a Distancia, Spain West University of Timisoara, Romania Systems Research Institute, Polish Academy of Sciences, Poland Universidad de Zaragoza, Spain
8. Parallel, Distributed and Multicore Computing Track Co-chairs Eduardo Alchieri Valentina Casola Lidia Ogiela
University of Brasilia, Brazil University of Naples “Federico II”, Italy AGH University of Science and Technology, Poland
PC Members Aldelir Luiz Edson Tavares Fernando Dotti Hylson Neto Jacir Bordim Lasaro Camargos
Catarinense Federal Institute, Brazil Federal University of Technology—Parana, Brazil Pontificia Universidade Catolica do Rio Grande do Sul, Brazil Catarinense Federal Institute, Brazil University of Brasilia, Brazil Federal University of Uberlandia, Brazil
xiv
Luiz Rodrigues Marcos Caetano Flora Amato Urszula Ogiela
CISIS-2022 Organizing Committee
Western Parana State University, Brazil University of Brasilia, Brazil University of Naples “Federico II”, Italy AGH University of Science and Technology, Poland
9. Energy Aware Computing and Systems Track Co-chairs Muzammil Behzad Zahoor Ali Khan Shigenari Nakamura
University of Oulu, Finland Higher Colleges of Technology, United Arab Emirates Tokyo Metropolitan Industrial Technology Research Institute, Japan
PC Members Naveed Ilyas Muhammad Sharjeel Javaid Muhammad Talal Hassan Waseem Raza Ayesha Hussain Umar Qasim Nadeem Javaid Yasir Javed Kashif Saleem Hai Wang
Gwangju Institute of Science and Technology, South Korea University of Hafr Al Batin, Saudi Arabia COMSATS University Islamabad, Pakistan University of Lahore, Pakistan COMSATS University Islamabad, Pakistan University of Engineering and Technology, Pakistan COMSATS University Islamabad, Pakistan Higher Colleges of Technology, UAE King Saud University, Saudi Arabia Saint Mary’s University, Canada
10. Multi-agent Systems, SLA Cloud and Social Computing Track Co-chairs Giuseppe Sarnè Douglas Macedo Takahiro Uchiya
Mediterranean University of Reggio Calabria, Italy Federal University of Santa Catarina, Brazil Nagoya Institute of Technology, Japan
PC Members Mario Dantas Luiz Bona Márcio Castro Fabrizio Messina
Federal University of Juiz de Fora, Brazil Federal University of Parana, Brazil Federal University of Santa Catarina, Brazil University of Catania, Italy
CISIS-2022 Organizing Committee
Hideyuki Takahashi Kazuto Sasai Satoru Izumi Domenico Rosaci Lidia Fotia
xv
Tohoku University, Japan Ibaraki University, Japan Tohoku University, Japan Mediterranean University of Reggio Calabria, Italy Mediterranean University of Reggio Calabria, Italy
11. Internet of Everything and Machine Learning Track Co-chairs Omid Ameri Sianaki Khandakar Ahmed Inmaculada Medina Bulo
Victoria University Sydney, Australia Victoria University, Australia Universidad de Cádiz, Spain
PC Members Farhad Daneshgar M. Reza Hoseiny F. Kamanashis Biswas (KB) Khaled Kourouche Huai Liu Mark A Gregory Nazmus Nafi Mashud Rana Farshid Hajati Ashkan Yousefi Nedal Ababneh Lorena Gutiérrez-Madroñal Juan Boubeta-Puig Guadalupe Ortiz Alfonso García del Prado Luis Llana
Victoria University Sydney, Australia University of Sydney, Australia Australian Catholic University, Australia Victoria University Sydney, Australia Victoria University, Australia RMIT University, Australia Victoria Institute of Technology, Australia CSIRO, Australia Victoria University Sydney, Australia Victoria University Sydney, Australia Abu Dhabi Polytechnic, Abu Dhabi, UAE University of Cádiz, Spain University of Cádiz, Spain University of Cádiz, Spain University of Cádiz, Spain Complutense University of Madrid, Spain
CISIS-2022 Reviewers Adhiatma Ardian Ali Khan Zahoor Amato Alba Amato Flora Barolli Admir Barolli Leonard Bista Bhed Chellappan Sriram
Chen Hsing-Chung Cui Baojiang Dantas Mario De Benedictis Alessandra Di Martino Beniamino Dong Hai Durresi Arjan Enokido Tomoya
xvi
Esposito Antonio Fachrunnisa Olivia Ficco Massimo Fotia Lidia Fun Li Kin Funabiki Nobuo Gotoh Yusuke Hussain Farookh Hussain Omar Javaid Nadeem Ikeda Makoto Ishida Tomoyuki Kikuchi Hiroaki Koyama Akio Kulla Elis Lee Kyungroul Matsuo Keita Mostarda Leonardo Oda Tetsuya Ogiela Lidia Ogiela Marek Okada Yoshihiro
CISIS-2022 Organizing Committee
Palmieri Francesco Park Hyunhee Paruchuri Vamsi Krishna Poniszewska-Maranda Aneta Rahayu Wenny Saito Takamichi Sakamoto Shinji Scionti Alberto Sianaki Omid Ameri Spaho Evjola Sudarti Ken Sugawara Shinji Takizawa Makoto Taniar David Terzo Olivier Uehara Minoru Venticinque Salvatore Vitabile Salvatore Woungang Isaac Xhafa Fatos Yim Kangbin Yoshihisa Tomoki
CISIS-2022 Keynote Talks
Design and Implementation Issues of Omnidirectional Robots and Their Applications for Different Environments Keita Matsuo Fukuoka Institute of Technology, Fukuoka, Japan
Abstract. Intelligent robotic systems are becoming essential for increasing Quality of Life (QoL) and keeping health for growing population of elderly people. In our research, in order to solve human health problems and support elderly people, we consider the design and implementation of omnidirectional robots. In this talk, I will introduce our results to show how omnidirectional wheelchair robots can support people with disabilities at home and at workplace. In our work, we also consider the use of the omnidirectional wheelchair robots for playing tennis and badminton. I also will present the application of omnidirectional robot as a mesh router in Wireless Mesh Networks (WMNs) in order to provide a good communication environment.
xix
Is Privacy the Same as Security, or Are They Just Two Sides of the Aame Coin? Anne Kayem Hasso-Plattner-Institute, University of Potsdam, Potsdam, Germany
Abstract. Almost every digital device either generates or consumes data in some form. The result is that the volumes of data collected grow exponentially each day. Data analytics proponents have mooted that it is now possible in some cases to actually predict future human behaviors based on data collected through tracking and various other means. On the other parallel, the question of privacy has become ever more important as users increasingly seek ways of guarding their personal data from exposure. This as such raises the question of what the distinction between privacy and security (data protection) is, and what the boundary between the two should be. For instance, the 2014 incident of a hacker faking the German minister of defense’s fingerprints was considered to be a security breach. However, a closer look at this issue highlights the fact that distinguishing between whether or not this was a privacy breach that enabled a security breach, or vice versa, does not have a straightforward answer. In this talk, I aim to explain why in my view privacy is different from security and, while though both privacy and security are mutually interdependent, why it is important to make the distinction. The talk will be supported by various examples to characterize privacy and distinguish it from security. At the same time, I will also explain why the two concepts are in fact two sides of the same coin.
xxi
Contents
Unnecessary Maneuvers as a Determinant of Driver Impatience in VANETs: Implementation and Evaluation of a Fuzzy-based System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kevin Bylykbashi, Ermioni Qafzezi, Phudit Ampririt, Elis Kulla, and Leonard Barolli Performance Evaluation of a Drone-Based Data Replication Method in Urban Disaster Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Makoto Ikeda, Seiya Sako, Masaya Azuma, Shota Uchimura, and Leonard Barolli
1
10
A Fast Convergence RDVM for Router Placement in WMNs: Performance Comparison of FC-RDVM with RDVM by WMNPSOHC Hybrid Intelligent System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shinji Sakamoto, Admir Barolli, Yi Liu, Elis Kulla, Leonard Barolli, and Makoto Takizawa
17
Energy-Efficient Two-Phase Locking Protocol by Omitting Meaningless Read and Write Methods . . . . . . . . . . . . . . . . . . . . . . . . . . Tomoya Enokido, Dilawaer Duolikun, and Makoto Takizawa
26
A New Method for Optimization of Number of Mesh Routers and Improving Cost Efficiency in Wireless Mesh Networks . . . . . . . . . . Aoto Hirata, Tetsuya Oda, Yuki Nagai, Tomoya Yasunaga, Nobuki Saito, Kengo Katayama, and Leonard Barolli A Wireless Sensor Network Testbed for Monitoring a Water Reservoir Tank: Experimental Results of Delay . . . . . . . . . . . . . . . . . . . Yuki Nagai, Tetsuya Oda, Chihiro Yukawa, Kyohei Toyoshima, Tomoya Yasunaga, Aoto Hirata, and Leonard Barolli
37
49
xxiii
xxiv
Contents
Remote Medical Assistance Vehicle in Covid-19 Quarantine Areas: A Case Study in Vietnam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Linh Thuy Thi Pham, Tan Phuc Nhan Bui, Ngoc Cam Thi Tran, Hai Thanh Nguyen, Khoi Tuan Huynh Nguyen, and Huong Hoang Luong Dynamic Job Allocation on Federated Cloud-HPC Environments . . . . . Giacomo Vitali, Alberto Scionti, Paolo Viviani, Chiara Vercellino, and Olivier Terzo Examination of Robot System Detecting Smoke Condition in the Event of a Fire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yang Kun and Takahiro Uchiya A Test Bed for Evaluating Graphene Filters in Indoor Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Federico Stirano, Fabrizio Bertone, Giuseppe Caragnano, and Olivier Terzo
59
71
83
94
An Overview of Emotion Recognition from Body Movement . . . . . . . . . 105 Laleh Ebdali Takalloo, Kin Fun Li, and Kosuke Takano Experimental Analysis and Verification of a Multi-modal-Biometrics Identity Verification Framework Based on the Dempster-Shafer Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Alfredo Cuzzocrea, Majid Abbasi Sisara, and Carmine Gallo Breast Ultrasound Image Classification Using EfficientNetV2 and Shallow Neural Network Architectures . . . . . . . . . . . . . . . . . . . . . . 130 Hai Thanh Nguyen, Linh Ngoc Le, Trang Minh Vo, Diem Ngoc Thi Pham, and Dien Thanh Tran Steganographic Approaches for Carrier Related Information Hiding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 Marek R. Ogiela and Urszula Ogiela A Bi-objective Genetic Algorithm for Wireless Sensor Network Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 Amit Dua, Pavel Krömer, Zbigniew J. Czech, and Tomasz Jastrząb 104 Fruits Classification Using Transfer Learning and DenseNet201 Fine-Tuning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 Khanh Vo Hong, Tin Tang Minh, Hoa Le Duc, Nam Truong Nhat, and Huong Luong Hoang Transfer Learning with Fine-Tuning on MobileNet and GRAD-CAM for Bones Abnormalities Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Huong Hoang Luong, Lan Thu Thi Le, Hai Thanh Nguyen, Vinh Quoc Hua, Khang Vu Nguyen, Thinh Nguyen Phuc Bach, Tu Ngoc Anh Nguyen, and Hien Tran Quang Nguyen
Contents
xxv
A Fuzzy-Based System for Handover in 5G Wireless Networks Considering Network Slicing Constraints . . . . . . . . . . . . . . . . . . . . . . . . 180 Phudit Ampririt, Ermioni Qafzezi, Kevin Bylykbashi, Makoto Ikeda, Keita Matsuo, and Leonard Barolli A Focused Beam Routing Protocol Considering Node Direction for Underwater Optical Wireless Communication in Delay Tolerant Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190 Keita Matsuo, Elis Kulla, and Leonard Barolli Taming Multi-node Accelerated Analytics: An Experience in Porting MATLAB to Scale with Python . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200 Paolo Viviani, Giacomo Vitali, Davide Lengani, Alberto Scionti, Chiara Vercellino, and Olivier Terzo An Adaptive Resource Allocation Protocol for Dynamic Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 Mojtaba Malek-Akhlagh and Jens H. Weber FPGA Implementation of an Object Recognition System with Low Power Consumption Using a YOLOv3-tiny-based CNN . . . . . . . . . . . . . 223 Yasutoshi Araki, Masatomo Matsuda, Taito Manabe, Yoichi Ishizuka, and Yuichiro Shibata Achieving Sustainable Competitive Advantage Through Green Innovation; the Moderating Effect of Islamic Environmental Ethics and Islamic Business Ethics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234 Budhi Cahyono and Marno Nugroho Digital Social Capital and Financial Inclusion for Small Medium Enterprises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249 Mutamimah and Pungky Lela Saputri Knowledge Absorptive Capacity Toward Sustainable Organizational Reputation in Digital Era . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260 Mulyana Mulyana and Erlinda Ramadhani Permata Putri The Role of Holistic Value Co-creation Capability in Improving Sustainable Relationship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 Ken Sudarti, Wasitowati, and Ari Pranaditya Conceptual Paper Tax Avoidance and Firm Value in Manufacturing Companies: A Case Study for Companies in Indonesia . . . . . . . . . . . . . 280 Chrisna Suhendi, Luluk Muhimatul Ifada, and Winarsih Customer Experience Management for ICT Industry Using SEM-PLS Analysis Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 Sri Safitri, Achmad Sudiro, Fatchur Rochman, and Mugiono Mugiono
xxvi
Contents
Increasing Repurchase Intention Through Personal Selling Capability, Customer Engagement, and Brand Trust . . . . . . . . . . . . . . . 303 Alifah Ratnawati, Erma Sri Hastuti, and Noor Kholis E-Impulse Buying Improvement with Product Knowledge, Shopping Lifestyle, and Positive Emotion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315 Lutfi Nurcholis and Nailus Sa’adah Transformational Performance of Police of the Republic of Indonesia Through Smart Working . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325 Siti Sumiati and Verri Dwi Prasetyo The Role of Tawhidic Paradigm in Knowledge Creation Process . . . . . . 337 Nurhidayati and Andhy Tri Adriyanto Islamic Human Values for Career Adaptability and Career Success of Millennial Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348 Ardian Adhiatma, Salsya Vivi Feronica Althof, and Meita Triantiani FAST: A Conceptual Framework for Reducing Fraud Financial Statement in Financial Business Practice . . . . . . . . . . . . . . . . . . . . . . . . 355 Ahmad Hijri Alfian, Verina Purnamasari, and Dian Essa Nugrahini Risk Management and Islamic Value: A Conceptual Development of Al-Adl Financing Risk Management . . . . . . . . . . . . . . . . . . . . . . . . . . . 364 Jumaizi, Widiyanto bin Mislan Cokrohadisumarto, and Eliya Tuzaka The Consumption Value and Value Congruity: A Conceptual Development of Hasanah Value Congruity . . . . . . . . . . . . . . . . . . . . . . . 373 Ari Pranaditya, Ken Sudarti, and Hendar Resilience of Companies Listed in Jakarta Islamic Index (JII) During the Pandemic COVID-19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382 Mutoharoh and Naila Najihah The Role of Institutional Investors in Lowering Information Asymmetry: Study on Mandatory Regulation of Integrated Reporting Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393 Naila Najihah and Mutoharoh Application of Business Process Semantic Annotation Techniques to Perform Pattern Recognition Activities Applied to the Generalized Civic Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404 Beniamino Di Martino, Mariangela Graziano, Luigi Colucci Cante, Antonio Esposito, and Maria Epifania A Semantic Representation for Public Calls Domain and Procedure: Housing Policies of Campania Region Case Study . . . . . . . . . . . . . . . . . 414 Beniamino Di Martino, Mariangela Graziano, Luigi Colucci Cante, Giuseppe Ferretti, and Valeria De Oto
Contents
xxvii
Machine Learning, Big Data Analytics and Natural Language Processing Techniques with Application to Social Media Analysis for Energy Communities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425 Beniamino Di Martino, Vincenzo Bombace, Luigi Colucci Cante, Antonio Esposito, Mariangela Graziano, Gennaro Junior Pezzullo, Alberto Tofani, and Gregorio D’Agostino Semantic Based Knowledge Management in e-Government Document Workflows: A Case Study for Judiciary Domain in Road Accident Trials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435 Beniamino Di Martino, Luigi Colucci Cante, Salvatore D’Angelo, Antonio Esposito, Mariangela Graziano, Rosario Ammendolia, and Pietro Lupi Towards the Identification of Architectural Patterns in Component Diagrams Through Semantic Techniques . . . . . . . . . . . . . . . . . . . . . . . . 446 Beniamino Di Martino, Piero Migliorato, and Antonio Esposito A Semantic Methodology for Security Controls Verification in Public Administration Business Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456 Massimiliano Rak, Daniele Granata, Beniamino Di Martino, and Luigi Colucci Cante Software Reuse Enabled by Machine Learning Based Source Code Analysis: A Case for Automated Classification of OpenSource Software with Respect to Requirements . . . . . . . . . . . . . . . . . . . . . . . . . 467 Dario Branco, Luigi Cuccaro, and Beniamino Di Martino Towards Machine Learning Enabled Analysis of Urban Mobility of Electric Motorbike: A Case Study for Improving Road Manteinance and Driver’s Safety in La Coruna City . . . . . . . . . . . . . . . 477 Dario Branco, Beniamino Di Martino, Salvatore Venticinque, Alberto León Fernández, and Valentin Porta Towards Semantic Description of Symbology and Heraldry Using Ontologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488 Alba Amato, Giuseppe Cirillo, and Francesco Moscato ECListener: A Platform for Monitoring Energy Communities . . . . . . . . 498 Gregorio D’Agostino, Alberto Tofani, Vincenzo Bombace, Luigi Colucci Cante, Antonio Esposito, Mariangela Graziano, Gennaro Junior Pezzullo, and Beniamino Di Martino Porting of Semantically Annotated and Geo-Located Images to an Interoperability Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 508 Alba Amato, Rocco Aversa, Dario Branco, Salvatore Venticinque, Giuseppina Renda, and Sabrina Mataluna
xxviii
Contents
Efficient Content Sharing Using Dynamic Fog in Cloud-Fog-Edge Three-Tiered Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517 Kohei Yoshikawa and Shinji Sugawara Study on the Comparison of Consumer Impression of E-commerce and Real Stores in the Fashion Tech Era, and the Effectiveness of VR Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528 Momoko Sakaguchi and Eiji Aoki Introducing Speaker Vectors for Child Speech Synthesis in Neural Vocoders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538 Satoshi Yoshida, Ken’ichi Furuya, and Hideyuki Mizuno Code Modification Problems for Multimedia Use in JavaScript-Based Web Client Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 548 Khaing Hsu Wai, Nobuo Funabiki, Huiyu Qi, Yanqi Xiao, Khin Thet Mon, and Yan Watequlis Syaifudin Design and Implementation of an Immersive Network Collaborative Environment Using OpenPose with 360VR Camera and WebXR . . . . . 557 Jingtao Xu, Wei Shi, and Yoshihiro Okada Dental Treatment Training System Using Haptic Device and Its User Evaluations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569 Masaki Nomi and Yoshihiro Okada Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
Unnecessary Maneuvers as a Determinant of Driver Impatience in VANETs: Implementation and Evaluation of a Fuzzy-based System Kevin Bylykbashi1(B) , Ermioni Qafzezi2 , Phudit Ampririt2 , Elis Kulla3 , and Leonard Barolli1 1
Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi,, Higashi-Ku Fukuoka 811-0295, Japan [email protected], [email protected] 2 Graduate School of Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi,, Higashi-Ku Fukuoka 811–0295, Japan {bd20101,bd21201}@bene.fit.ac.jp 3 Department of System Management, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi,, Higashi-Ku Fukuoka 811–0295, Japan [email protected]
Abstract. In our previous work, we implemented an intelligent system based on Fuzzy Logic (FL) for deciding the driver’s impatience in VANETs. The implemented system, called Fuzzy-based System for Deciding Driver Impatience (FSDDI), considered parameters that cause driver’s impatience such as their emotional condition, the time pressure, and the number of route stops. In this work, we implement a modified version of FSDDI, which considers the unnecessary maneuvers that drivers make while driving as an additional input. We show through simulations the effect that the unnecessary maneuvers and the other parameters have on the determination of the driver’s impatience and demonstrate some actions that can be performed when the driver shows high degrees of impatience.
1
Introduction
The highly competitive and rapidly advancing autonomous vehicle race has been on for several years now, and it is a matter of time until we have these vehicles on the roads. However, even if the automotive companies do all it takes to create fully automated cars, there will still be one big obstacle, the infrastructure. In addition, this could take decades, even in the most developed countries. Moreover, 93% of the world’s fatalities on the roads occur in low- and middleincome countries [10] and considering all these facts, Driver Assistance Systems (DASs) and Vehicular Ad hoc Networks (VANETs) should remain in focus for the foreseeable future. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 1–9, 2022. https://doi.org/10.1007/978-3-031-08812-4_1
2
K. Bylykbashi et al.
DASs are intelligent systems that are implemented in vehicles to increase driving safety by assisting drivers and can be very helpful in a variety of situations as they do not depend on the infrastructure as much as driverless vehicles do. Furthermore, DASs can provide driving support with very little cost, thus help the low- and middle-income countries in the long battle against car accidents. VANETs, on the other hand, aim not only at saving lives but also improving traffic mobility, increasing efficiency, and promoting travel convenience of drivers and passengers. In VANETs, network nodes (vehicles) are equipped with networking functions to exchange essential information with each other via vehicle-to-vehicle (V2V) and with roadside units (RSUs) through vehicle-toinfrastructure (V2I) communications. By leveraging the data acquired by other vehicles and infrastructure, DASs can make better decisions and offer more services, which provides drivers with enhanced applications and experience. These data range from simple information such as traffic and road condition messages to a complete perception of the surrounding environment obtained through cameras, thus improving road safety significantly. Nevertheless, there are other determinants that affect the driving operation, such as the drivers and their behaviors. In fact, according to some traffic safety facts provided by a survey of the U.S. Department of Transportation [9], the drivers are the immediate reason for more than 94% of the investigated car crashes. While most of the errors committed by drivers are involuntary, there are errors which come as a result of their behavior, and this must be utterly preventable. An indicator of the behavior of drivers is the patience they show when they are behind the wheel. Deciding the impatience of drivers is consequently a need that requires careful and immediate work. In [3], we have proposed an intelligent system based on Fuzzy Logic (FL) that determines the driver’s impatience in real-time based on factors such as driver’s emotional condition, time pressure and number of route stops. In this work, we present a modified version of our system that additionally considers the unnecessary maneuvers that drivers make as an input parameter. The concept of proposed system is given in Fig. 1. We evaluate the proposed system by computer simulations and see the effect that the unnecessary maneuvers and the other parameters have on the determination of the driver’s impatience. The structure of the paper is as follows. Section 2 presents a brief overview of VANETs. Section 3 describes the proposed fuzzy-based simulation system and its implementation. Section 4 discusses the simulation results. Finally, conclusions and future work are given in Sect. 5.
A Fuzzy-Based System for Deciding Driver’s Impatience in VANETs
3
Driver's Emotional Condition (DEC)
Driver's Impatience (DI)
Unnecessary Maneuvers (UM) Time Pressure (TP)
Actor Device
Number of Route Stops (NRS)
Fig. 1. Concept of proposed system.
2
Overview of VANETs
VANETs are a special case of Mobile Ad hoc Networks (MANETs) in which the mobile nodes are vehicles. In VANETs, nodes have high mobility and tend to follow organized routes instead of moving randomly. Moreover, vehicles offer attractive features such as higher computational capability and localization through GPS. As a key component of Intelligent Transportation Systems (ITSs), VANETs have huge potential to enable applications ranging from road safety, traffic optimization, infotainment, commercial to rural and disaster scenario connectivity. Among these, road safety and traffic optimization are considered the most important ones as they have the goal to reduce drastically the high number of accidents, guarantee road safety, make traffic management, and create new forms of inter-vehicle communications in ITSs. The ITSs manage the vehicle traffic, support drivers with safety and other information, and enable applications such as automated toll collection and DASs [4]. Despite the attractive features, VANETs are characterized by very large and dynamic topologies, variable capacity wireless links, bandwidth and hard delay constraints, and by short contact durations which are caused by the high mobility, high speed, and low density of vehicles. In addition, limited transmission ranges, physical obstacles, and interferences make these networks characterized by intermittent connectivity. Therefore, it is necessary to design proper systems, network architectures and applications that can overcome the problems that arise from vehicular environments.
4
K. Bylykbashi et al.
Fig. 2. Diagram of FSDDI.
3
Proposed Fuzzy-Based System
Our research work focuses on developing an intelligent non-complex driving support system that determines the driving risk level in real-time by considering different types of parameters. In previous works, we have considered different parameters, including in-car environment parameters such as the ambient temperature and noise, and driver’s vital signs, i.e., heart and respiratory rate, for which we implemented a testbed and conducted experiments in a real scenario [2]. The considered parameters include environmental factors and driver’s health condition, as these parameters affect the driver’s capability and vehicle performance. In [1], we presented an integrated fuzzy-based system, which in addition to those parameters, considers the following inputs: vehicle speed, weather and road condition, driver’s body temperature, and vehicle interior relative humidity. The inputs were categorized based on the way they affect the driving operation. In a more recent work [3], we proposed a system that decides the driver’s impatience since the impatient drivers are often an immediate cause of many road accidents. In this work, we consider the driver’s impatience because impatient drivers are with their behavior, an immediate cause of many accidents. We use FL to implement the proposed system as it can make a real-time decision based on the uncertainty and vagueness of the provided information [5–8,11,12]. The proposed system, called Fuzzy-based System for Deciding the Driver’s Impatience (FSDDI), is shown in Fig. 2. FSDDI has the following inputs: Driver’s Emotional Condition (DEC), Unnecessary Maneuvers (UM), Time Pres-
A Fuzzy-Based System for Deciding Driver’s Impatience in VANETs
5
Table 1. Parameters and their term sets for FSDDI. Parameters
Term set
Driver’s Emotional Condition (DEC) Unnecessary Maneuvers (UM) Time Pressure (TP) Number of Route Stops (NRS)
Very bad (VB), Bad (B), Good (G) Few (Fw), Moderate (Mr), Many (Mn) Low (Lo), Medium (Me), High (Hi) Few (Fe), Moderate (Mo), Many (Ma)
Driver’s Impatience (DI)
Low (Lw), Low-to-Moderate (LM), Moderate (Md), Moderate-to-High (MH), High (Hg), Very High (VH), Extremely High (EH)
sure (TP) and Number of Route Stops (NRS). The output of the system is the Driver’s Impatience (DI). The term set for each parameter is given in Table 1. Table 2. FRB of FSDDI. No DEC UM TP NRS DI
No DEC UM TP NRS DI
No DEC UM TP NRS DI
1
VB
Fw
Lo Fe
Md 28 B
Fw
Lo Fe
Lw
55 G
Fw
Lo Fe
Lw
2
VB
Fw
Lo Mo
MH 29 B
Fw
Lo Mo
LM 56 G
Fw
Lo Mo
Lw
3
VB
Fw
Lo Ma
Hg
30 B
Fw
Lo Ma
Md 57 G
Fw
Lo Ma
LM
4
VB
Fw
Me Fe
MH 31 B
Fw
Me Fe
LM 58 G
Fw
Me Fe
Lw
5
VB
Fw
Me Mo
Hg
32 B
Fw
Me Mo
Md 59 G
Fw
Me Mo
LM
6
VB
Fw
Me Ma
VH 33 B
Fw
Me Ma
MH 60 G
Fw
Me Ma
Md
7
VB
Fw
Hi
Fe
Hg
34 B
Fw
Hi
Fe
Md 61 G
Fw
Hi
Fe
LM
8
VB
Fw
Hi
Mo
VH 35 B
Fw
Hi
Mo
MH 62 G
Fw
Hi
Mo
Md
9
VB
Fw
Hi
Ma
EH 36 B
Fw
Hi
Ma
Hg
63 G
Fw
Hi
Ma
MH
10 VB
Mr
Lo Fe
MH 37 B
Mr
Lo Fe
LM 64 G
Mr
Lo Fe
Lw
11 VB
Mr
Lo Mo
Hg
38 B
Mr
Lo Mo
Md 65 G
Mr
Lo Mo
LM
12 VB
Mr
Lo Ma
VH 39 B
Mr
Lo Ma
MH 66 G
Mr
Lo Ma
Md
13 VB
Mr
Me Fe
Hg
40 B
Mr
Me Fe
Md 67 G
Mr
Me Fe
LM
14 VB
Mr
Me Mo
VH 41 B
Mr
Me Mo
MH 68 G
Mr
Me Mo
Md
15 VB
Mr
Me Ma
EH 42 B
Mr
Me Ma
Hg
69 G
Mr
Me Ma
MH
16 VB
Mr
Hi
Fe
VH 43 B
Mr
Hi
Fe
MH 70 G
Mr
Hi
Fe
Md
17 VB
Mr
Hi
Mo
EH 44 B
Mr
Hi
Mo
Hg
71 G
Mr
Hi
Mo
MH
18 VB
Mr
Hi
Ma
EH 45 B
Mr
Hi
Ma
VH 72 G
Mr
Hi
Ma
Hg
19 VB
Mn Lo Fe
Hg
Mn Lo Fe
Md 73 G
Mn Lo Fe
LM
20 VB
Mn Lo Mo
VH 47 B
Mn Lo Mo
MH 74 G
Mn Lo Mo
Md
21 VB
Mn Lo Ma
EH 48 B
Mn Lo Ma
Hg
Mn Lo Ma
MH
22 VB
Mn Me Fe
VH 49 B
Mn Me Fe
MH 76 G
Mn Me Fe
Md
23 VB
Mn Me Mo
EH 50 B
Mn Me Mo
Hg
77 G
Mn Me Mo
MH
24 VB
Mn Me Ma
EH 51 B
Mn Me Ma
VH 78 G
Mn Me Ma
Hg
25 VB
Mn Hi
Fe
EH 52 B
Mn Hi
Fe
Hg
79 G
Mn Hi
Fe
MH
26 VB
Mn Hi
Mo
EH 53 B
Mn Hi
Mo
VH 80 G
Mn Hi
Mo
Hg
27 VB
Mn Hi
Ma
EH 54 B
Mn Hi
Ma
EH 81 G
Mn Hi
Ma
VH
46 B
75 G
6
K. Bylykbashi et al.
Fig. 3. Membership functions.
Based on the linguistic description of input and output parameters, the Fuzzy Rule Base (FRB) of the proposed system forms a fuzzy set of dimensions | T (x1 ) | × | T (x2 ) | × · · · × | T (xn ) |, where | T (xi ) | is the number of terms on T (xi ) and n is the number of input parameters. FSDDI has four input parameters with three linguistic terms each, therefore there are 81 rules in the FRB. The FRB is shown in Table 2. The control rules of FRB have the form: IF “conditions” THEN “control action”. The membership functions are shown in Fig. 3. We use triangular and trapezoidal membership functions because these types of functions are more suitable for real-time operation.
4
Simulation Results
In this section, we present the simulation results for our proposed system. The simulation results are presented in Figs. 4, 5 and 6. We consider DEC and UM as constant parameters. We show the relation between DI and NRS for different TP values.
A Fuzzy-Based System for Deciding Driver’s Impatience in VANETs
7
Fig. 4. Simulation results for DEC = 0.1.
Fig. 5. Simulation results for DEC = 0.5.
In Fig. 4, we show the results for DEC = 0.1 and change UM from 0.1 to 0.9. We can see that almost all scenarios show high levels of impatience. The scenario with the lowest degree of impatience is when the driver makes only a few unnecessary maneuvers (e.g., unnecessary accelerations or lane changes) while driving along routes with few stops and under no pressure of time, and yet it is decided as a moderate impatience. On the other hand, when the driver makes many unnecessary maneuvers, the impatience is very high and extremely high in almost every scenario. When the driver is in somewhat better emotional condition (see Fig. 5), we see fewer scenarios with very high and extreme impatience. Such impatience is seen only when the driver is under the pressure of time and makes many unnecessary maneuvers. The drivers that make only a few of these maneuvers are more patient, which makes them less prone to accidents. The simulation results for drivers in good emotional condition are given in Fig. 6. We can see that the drivers are more patient, but tend to show impatience once they feel the pressure of time and make many unnecessary maneuvers. The DI increases even more when they are under maximum time pressure and there are many stops along the way, with the level of impatience in this scenario decided as very high.
8
K. Bylykbashi et al.
Fig. 6. Simulation results for DEC = 0.9.
In the cases when the driver show high levels of impatience, the system can take several actions that can improve the driving situations and therefore reduce the risks of accidents. For example, when the time is not a factor, the system can suggest the use of a longer route but with fewer stops, or adjust the interior environment of the car to improve the driver’s mood.
5
Conclusions
In this work, we presented the implementation of an FL approach that determines driver’s impatience by considering the driver’s emotional condition, the unnecessary maneuvers, the time pressure and the number of route stops. We showed through simulations the effect of the considered parameters on the determination of the degree of impatience. The simulations show that when drivers make many unnecessary maneuvers, they tend to show an increased degree of impatience, especially if they are under high time pressure. In addition, the impatience is even higher when the driver is not in good emotional condition and has taken a road that includes many stops. In the future, we would like to make extensive simulations and experiments to evaluate the proposed system and compare the performance with other systems.
References 1. Bylykbashi, K., Qafzezi, E., Ampririt, P., Ikeda, M., Matsuo, K., Barolli, L.: Performance evaluation of an integrated fuzzy-based driving-support system for realtime risk management in VANETs. Sensors 20(22), 6537 (2020). https://doi.org/ 10.3390/s20226537 2. Bylykbashi, K., Qafzezi, E., Ikeda, M., Matsuo, K., Barolli, L.: Fuzzy-based Driver Monitoring System (FDMS): Implementation of two intelligent FDMSs and a testbed for safe driving in VANETs. Futur. Gener. Comput. Syst. 105, 665–674 (2020). https://doi.org/10.1016/j.future.2019.12.030
A Fuzzy-Based System for Deciding Driver’s Impatience in VANETs
9
3. Bylykbashi, K., Qafzezi, E., Ampririt, P., Ikeda, M., Matsuo, K., Barolli, L.: A fuzzy-based system for deciding driver impatience in VANETs. In: Barolli, L. (ed.) 3PGCIC 2021. LNNS, vol. 343, pp. 129–137. Springer, Cham (2022). https://doi. org/10.1007/978-3-030-89899-1 13 4. Hartenstein, H., Laberteaux, L.: A tutorial survey on vehicular ad hoc networks. IEEE Commun. Mag. 46(6), 164–171 (2008). https://doi.org/10.1109/MCOM. 2008.4539481 5. Kandel, A.: Fuzzy Expert Systems. CRC Press Inc., Boca Raton (1992) 6. Klir, G.J., Folger, T.A.: Fuzzy Sets, Uncertainty, and Information. Prentice Hall Inc., Upper Saddle River (1987) 7. McNeill, F.M., Thro, E.: Fuzzy Logic: A Practical Approach. Academic Press Professional Inc, San Diego, CA, USA, (1994). https://doi.org/10.1016/C20130-11164-6 8. Munakata, T., Jani, Y.: Fuzzy systems: an overview. Commun. ACM 37(3), 69–77 (1994). https://doi.org/10.1145/175247.175254 9. Singh, S.: Critical Reasons for Crashes Investigated in the National Motor Vehicle Crash Causation Survey. Traffic Safety Facts: Crash Stats. Report No. DOT HS 812 506, Washington, DC: National Highway Traffic Safety Administration (NHTSA) (2018) 10. World Health Organization (2018) Global status report on road safety 2018: summary. World Health Organization, Geneva, Switzerland, (WHO/NMH/NVI/18.20). Licence: CC BY-NC-SA 3.0 IGO) 11. Zadeh, L.A., Kacprzyk, J.: Fuzzy Logic for the Management of Uncertainty. John Wiley & Sons Inc., New York (1992) 12. Zimmermann, H.J.: Fuzzy Set Theory and Its Applications. Springer Science & Business Media, New York (1996). https://doi.org/10.1007/978-94-015-8702-0
Performance Evaluation of a Drone-Based Data Replication Method in Urban Disaster Scenario Makoto Ikeda1(B) , Seiya Sako1 , Masaya Azuma2 , Shota Uchimura2 , and Leonard Barolli1 1
Department of Information and Communication Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-higashi, Higashi-ku, Fukuoka 811-0295, Japan [email protected], [email protected] 2 Graduate School of Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-higashi, Higashi-ku, Fukuoka 811-0295, Japan {mgm21101,mgm21102}@bene.fit.ac.jp Abstract. In this work, we focus on a drone-based message ferry between clusters in disaster situation. We consider Delay-Tolerant Networking (DTN) for our scenario and evaluate the message ferry method considering drones and vehicles. We use an urban grid road model in normal and disaster situations. From the simulation results, we found that using drones reduces delay time and improves delivery ratio. Keywords: Message ferry
1
· DTN · drone · Disaster
Introduction
In a Vehicle-to-Vehicle (V2V) network, if a node wants to send data to another node that does not have a neighboring node within communication range, there will be a delay or the message will be subsequently rejected. Delay-Tolerant Networking (DTN) [6] is a good approach for this network environment [2,10]. Various routing methods based on ad-hoc networks have been proposed, but they require consideration of unmanned aerial vehicles to cover the rural areas that are physically disconnected. In our previous work [3], we presented an adaptive anti-packet recovery method for shuttle buses and road-side units in urban environment. However, we did not consider flying terminals. In [13,14], the authors proposed a message ferry scheme. Design guidelines for message ferry systems were presented to help determine how many ferries are needed and how much performance is achieved. However, they did not consider mobile networks and networks with dynamic traffic. In this paper, we focus on the impact that drones have on the replicated bundle messages for urban disaster scenario. We consider Epidemic protocol as message replication. The rest of the paper is structured as follows. In Sect. 2, we give the overview of DTN. In Sect. 3, we provide the simulation settings. In Sect. 4, we provide the evaluation results. Finally, conclusions and future work are given in Sect. 5. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 10–16, 2022. https://doi.org/10.1007/978-3-031-08812-4_2
Performance Evaluation of a Drone-Based Data Replication Method
2
11
DTN
DTN is capable of providing a reliable internet connection for space-related operations [5,8,12]. Frequent link failures, disconnection and large delays are possible in space networks. In Vehicular DTN, the intermediate vehicles keep bundle messages in their storage and subsequently send them to other vehicles. RFC 4838 [4] specifies the network architecture. The well-known DTN protocol is Epidemic Routing [7,11], which is performed by using two control messages to replicate a bundle message. Each vehicle periodically transmits a Summary Vector (SV) to the network. The SV contains a list of stored messages for each vehicle. When the vehicles receive SVs, they compare them to their own. If the received SV contains an unknown bundle message, the vehicle sends a request message. In this method, the consumption of network resources and storage state become critical issues, because the vehicles replicate bundle messages to adjacent vehicles in their communication range. Then, the received bundle messages remain in the storage and the bundle messages are continuously replicated even if the end-point receives the messages. As a result, recovery mechanisms such as timers and anti-packets may delete the replicated bundle messages from the network. Due to the high latency, the anti-packet deletes the replicated messages too late. In the case of the conventional anti-packet, the end-point broadcasts the anti-packet, which contains the list of bundle messages that reach to the end-point. Vehicles delete messages and replicate the anti-packet to other vehicles based on the anti-packet. In the case of the conventional timer, the messages have a lifespan and they are deleted on a constant schedule when the lifespan of the bundle messages has expired.
3
Simulation Settings
In this section, we describe the message delivery method in urban disaster scenario on the Scenargie [9] simulator. In Fig. 1(a), we present an urban road model for normal scenario. For the disaster scenario (see Fig. 1(b)), there are some closed roads that vehicles can not move in that area. We considered the following conditions using a maximum of 90 regular vehicles on the road. • Both car and drone use the Epidemic with anti-packet as a message delivery terminal. • The message start-point and end-point are fixed. • Some roads are closed in a disaster situation. • Cars continue to move on the roads based on the GIS-based random way-point mobility model. • The drones periodically move along the air route in a rectangle way. When the drones reach the stop-point, they stay for 20 s. After that, the drones move to the next stop-point according to a trace-file-based schedule.
12
M. Ikeda et al.
(a) Normal scenario
(b) Disaster scenario
Fig. 1. Target environment for different situations.
• Radio waves in the 5.9 GHz band have strong linearity, and it is difficult to communicate with non-line-of-sight, such as behind large buildings and vehicles. Therefore, obstacles such as buildings may reduce the reachability of messages. We consider three drone altitudes: 20, 30 and 40 meters. Table 1 shows the simulation parameters used for the network simulator. The start-point replicates 40 kinds of bundle messages to the relay vehicles. Then, the replicated bundle messages are delivered to the end-point. The simu-
Performance Evaluation of a Drone-Based Data Replication Method
13
Table 1. Simulation parameters. Parameters
Values
Simulation time
500 [seconds]
Area dimensions
700 [m] × 700 [m]
Number of regular vehicles
30, 60 and 90 [vehicles]
Number of drones
0 or 2 [vehicles]
Building height
15 [m]
Drone flight altitudes
20, 30, 40 [m]
Drone stop duration
20 [s]
Mobility model: vehicles
GIS-based random way-point
Mobility model: drones
Trace-file based (scheduled)
Minimum speed
8.333 [m/s]
Maximum speed
16.666 [m/s]
Message start and end time 0 - 400 [seconds] Message generation interval 10 [seconds] Message size
1, 000 [bytes]
DTN protocol
Epidemic with anti-packet
PHY model
IEEE 802.11p
Propagation model
ITU-R P.1411
Radio frequency
5.9 [GHz]
Antenna model
Omni-directional
lation time is 500 s. The ITU-R P.1411 propagation model is used in this urban road scenario [1]. All buildings are 15-meter height. We evaluate the performance of overhead, delivery ratio and delay for different vehicles. • The overhead parameter indicates the number of replicated bundle messages in the network. • The delivery ratio indicates the value of the generated bundle messages divided by the delivered bundle messages to the end-point. • The delay indicates the transmission latency of the bundle message to reach the end-point.
4
Evaluation Results
We evaluate the drone-based message delivery method considering Epidemic with anti-packet in normal and disaster scenarios. The simulation results are shown in Fig. 2, Fig. 3 and Fig. 4. The distance represents the drones’ constant flight altitude during the simulation. “No drone” indicates that regular vehicles are the only ones that are considered as intermediate terminals.
14
M. Ikeda et al.
In Fig. 2, we present the simulation results for overhead from 30 vehicles to 90 vehicles. We observed that the overhead increases with increasing the number of vehicles. In addition, when drones fly at a high altitude in a sparse network, the overhead results are higher than those at a lower altitude. This is due to the increased probability of the vehicle making contact with the drone’s line-of-sight. We observed that when drones are utilized in a disaster scenario with a dense network, the number of replicas is increased.
(a) Normal scenario
(b) Disaster scenario
Fig. 2. Overhead.
In Fig. 3, we present the simulation results of the delivery ratio. The delivery ratio increases with increasing the number of vehicles for both scenarios. For 30 vehicles in normal scenario, delivery ratio is high when the drones’ flight altitude is low. On the other side, during disasters, it was observed that the altitude of the drone had a good effect on reachability. Additionally, we observed that the use of drones improved the delivery ratio in both scenarios.
(a) Normal scenario
(b) Disaster scenario
Fig. 3. Delivery ratio.
In Fig. 4, we present the simulation results of the delay. Our method improves the delay by around one second for 30 vehicles in a normal situation. For 60 and
Performance Evaluation of a Drone-Based Data Replication Method
(a) Normal scenario
15
(b) Disaster scenario
Fig. 4. Delay.
90 vehicles, the delay is almost the same. When the flight altitude is high in disaster scenario, the delay improves to around three seconds for 30 vehicles. In all cases, the usage of drones reduced the delay time.
5
Conclusions
In this paper, we evaluated the network performance of drone-based replicated method in an urban disaster scenario. From the simulation results, we observed that the deployment of drones can improve the delivery ratio while also reducing the delay time. In future work, we would like to compare other replication methods and consider other parameters.
References 1. Rec. ITU-R P.1411-7: Propagation data and prediction methods for the planning of short-range outdoor radiocommunication systems and radio local area networks in the frequency range 300 MHz to 100 GHz. ITU (2013) 2. Arafat, M.Y., Moh, S.: Location-aided delay tolerant routing protocol in UAV networks for post-disaster operation. IEEE Access 6, 59891–59906 (2018) 3. Azuma, M., Uchimura, S., Tada, Y., Ikeda, M., Barolli, L.: An adaptive antipacket recovery method for vehicular DTN: performance evaluation considering shuttle buses and roadside units scenario. In: Proceedings of the 16th International Conference on Broad-Band Wireless Computing, Communication and Applications (BWCCA-2021), pp. 234–241, October 2021 4. Cerf, V., et al.: Delay-tolerant networking architecture. IETF RFC 4838 (Informational), April 2007 5. Fall, K.: A delay-tolerant network architecture for challenged Internets. In: Proceedings of the International Conference on Applications, Technologies, Architectures, and Protocols for Computer Communications, SIGCOMM 2003, pp. 27–34 (2003)
16
M. Ikeda et al.
6. Kawabata, N., Yamasaki, Y., Ohsaki, H.: Hybrid cellular-DTN for vehicle volume data collection in rural areas. In: Proceedings of the IEEE 43rd Annual Computer Software and Applications Conference (COMPSAC-2019), vol. 2, pp. 276–284, July 2019 7. Ramanathan, R., Hansen, R., Basu, P., Hain, R.R., Krishnan, R.: Prioritized epidemic routing for opportunistic networks. In: Proceedings of the 1st International MobiSys Workshop on Mobile Opportunistic Networking (MobiOpp 2007) pp. 62– 66 (2007) 8. R¨ usch, S., Sch¨ urmann, D., Kapitza, R., Wolf, L.: Forward secure delay-tolerant networking. In: Proceedings of the 12th Workshop on Challenged Networks (CHANTS-2017), pp. 7–12, October 2017 9. Scenargie: Space-time engineering, LLC. http://www.spacetime-eng.com/ 10. Solpico, D., et al.: Application of the V-HUB standard using LoRa beacons, mobile cloud, UAVs, and DTN for disaster-resilient communications. In: Proceedings of the IEEE Global Humanitarian Technology Conference (GHTC-2019), pp. 1–8, October 2019 11. Vahdat, A., Becker, D.: Epidemic routing for partially-connected ad hoc networks. Duke University, Tech. rep. (2000) 12. Wyatt, J., Burleigh, S., Jones, R., Torgerson, L., Wissler, S.: Disruption tolerant networking flight validation experiment on NASA’s EPOXI mission. In: Proceedings of the 1st International Conference on Advances in Satellite and Space Communications (SPACOMM-2009), pp. 187–196, July 2009 13. Zhao, W., Ammar, M., Zegura, E.: Controlling the mobility of multiple data transport ferries in a delay-tolerant network. In: Proceedings IEEE 24th Annual Joint Conference of the IEEE Computer and Communications Societies, vol. 2, pp. 1407– 1418, March 2005 14. Zhao, W., Ammar, M.: Message ferrying: proactive routing in highly-partitioned wireless ad hoc networks. In: The Ninth IEEE Workshop on Future Trends of Distributed Computing Systems, 2003, FTDCS 2003. Proceedings, pp. 308–314, May 2003
A Fast Convergence RDVM for Router Placement in WMNs: Performance Comparison of FC-RDVM with RDVM by WMN-PSOHC Hybrid Intelligent System Shinji Sakamoto1(B) , Admir Barolli2 , Yi Liu3 , Elis Kulla4 , Leonard Barolli5 , and Makoto Takizawa6 1
2
5
Department of Information and Computer Science, Kanazawa Institute of Technology, 7-1 Ohgigaoka, Nonoichi, Ishikawa 921-8501, Japan [email protected] Department of Information Technology, Aleksander Moisiu University of Durres, L.1, Rruga e Currilave, Durres, Albania [email protected] 3 Department of Computer Science, National Institute of Technology, Oita College, 1666, Maki, Oita 870-0152, Japan [email protected] 4 Department of System Management, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected] Department of Information and Communication Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected] 6 Department of Advanced Sciences, Faculty of Science and Engineering, Hosei University, Kajino-Machi, Koganei-Shi, Tokyo 184-8584, Japan [email protected] Abstract. Wireless Mesh Networks (WMNs) have many advantages such as easy maintenance, low upfront cost and high robustness. However, WMNs have some problems such as node placement problem, security, transmission power and so on. In our previous work, we implemented a hybrid simulation system based on Particle Swarm Optimization (PSO) and Hill Climbing (HC) called WMN-PSOHC for solving the node placement problem in WMNs. We also proposed and impremented Rational Decrement of Vmax Method (RDVM). In this paper, we propose and implement a Fast Convergence RDVM (FC-RDVM). We compare the performance of FC-RDVM with RDVM. Simulation results show that FC-RDVM has better performance than RDVM.
1
Introduction
In this work, we deal with node placement problem in WMNs. We consider the version of the mesh router nodes placement problem in which we are given a c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 17–25, 2022. https://doi.org/10.1007/978-3-031-08812-4_3
18
S. Sakamoto et al.
grid area where to deploy a number of mesh router nodes and a number of mesh client nodes of fixed positions (of an arbitrary distribution) in the grid area. The objective is to find a location assignment for the mesh routers to the cells of the grid area that maximizes the network connectivity and client coverage. Network connectivity is measured by Size of Giant Component (SGC) of the resulting WMN graph, while the user coverage is simply the number of mesh client nodes that fall within the radio coverage of at least one mesh router node and is measured by Number of Covered Mesh Clients (NCMC). Node placement problems are known to be computationally hard to solve [13]. In some previous works, intelligent algorithms have been recently investigated [2,7,8]. We already implemented a Particle Swarm Optimization (PSO) based simulation system, called WMN-PSO [5]. Also, we implemented a simulation system based on Hill Climbing (HC) for solving node placement problem in WMNs, called WMNHC [4]. In our previous work, we presented a hybrid intelligent simulation system based on PSO and HC [6]. We called this system WMN-PSOHC. We also proposed and impremented Rational Decrement of Vmax Method (RDVM) replacement method [9]. In this paper, we propose and implement a Fast Convergence RDVM (FC-RDVM) for WMN-PSOHC. Then, we compare FC-RDVM with RDVM. The rest of the paper is organized as follows. In Sect. 2, we present intelligent algorithms. We present our designed and implemented hybrid simulation system in Sect. 3. The simulation results are given in Sect. 4. Finally, we give conclusions and future work in Sect. 5.
2 2.1
Intelligent Algorithms Particle Swarm Optimization
In Particle Swarm Optimization (PSO) algorithm, a number of simple entities (the particles) are placed in the search space of some problem or function and each evaluates the objective function at its current location. The objective function is often minimized and the exploration of the search space is not through evolution [3]. However, following a widespread practice of borrowing from the evolutionary computation field, in this work, we consider the bi-objective function and fitness function interchangeably. Each particle then determines its movement through the search space by combining some aspect of the history of its own current and best (best-fitness) locations with those of one or more members of the swarm, with some random perturbations. The next iteration takes place after all particles have been moved. Eventually the swarm as a whole, like a flock of birds collectively foraging for food, is likely to move close to an optimum of the fitness function. Each individual in the particle swarm is composed of three D-dimensional vectors, where D is the dimensionality of the search space. These are the current position xi , the previous best position pi and the velocity v i .
A Fast Convergence RDVM for Router Placement in WMNs
19
Algorithm 1. Pseudo code of WMN-PSOHC. /* Generate the initial solutions and parameters */ Computation maxtime:= Tmax , k := 1; Number of particle-patterns:= m, 2 ≤ m ∈ N 1 ; Particle-patterns initial solution:= P 0i ; Global initial solution:= G 0 ; Particle-patterns initial position:= x 0ij ; Particles initial velocity:= v 0ij ; PSO parameter:= ω, 0 < ω ∈ R1 ; PSO parameter:= C1 , 0 < C1 ∈ R1 ; PSO parameter:= C2 , 0 < C2 ∈ R1 ; /* Start PSO-HC */ Evaluate(G 0 , P 0 ); while k ≤ Tmax do /* Update velocities and positions */ = ω · v kij v k+1 ij +C1 · rand() · (best(Pijk ) − xkij ) +C2 · rand() · (best(Gk ) − xkij ); k+1 x ij = x kij + v k+1 ij ; /* if fitness value is increased, a new solution will be accepted. */ if Evaluate(G (k+1) , P (k+1) ) >= Evaluate(G (k) , P (k) ) then Update Solutions(G k , P k ); Evaluate(G (k+1) , P (k+1) ); else ReUpdate Solutions(G k+1 , P k+1 ); end if k = k + 1; end while Update Solutions(G k , P k ); return Best found pattern of particles as solution;
The particle swarm is more than just a collection of particles. A particle by itself has almost no power to solve any problem; progress occurs only when the particles interact. Problem solving is a population-wide phenomenon, emerging from the individual behaviors of the particles through their interactions. In any case, populations are organized according to some sort of communication structure or topology, often thought of as a social network. The topology typically consists of bidirectional edges connecting pairs of particles, so that if j is in i’s neighborhood, i is also in j’s. Each particle communicates with some other particles and is affected by the best point found by any member of its topological neighborhood. This is just the vector pi for that best neighbor, which we will denote with pg . The potential kinds of population “social networks” are hugely varied, but in practice certain types have been used more frequently. In the PSO process, the velocity of each particle is iteratively adjusted so that the particle stochastically oscillates around pi and pg locations.
20
2.2
S. Sakamoto et al.
Hill Climbing
Hill Climbing (HC) algorithm is a heuristic algorithm. The idea of HC is simple. In HC, the solution s is accepted as the new current solution if δ ≤ 0 holds, where δ = f (s ) − f (s). Here, the function f is called the fitness function. The fitness function gives points to a solution so that the system can evaluate the next solution s and the current solution s. The most important factor in HC is to define effectively the neighbor solution. The definition of the neighbor solution affects HC performance directly. In our WMN-PSOHC system, we use the next step of particle-pattern positions as the neighbor solutions for the HC part.
3
Proposed WMN-PSOHC System
We show the pseudo code of WMN-PSOHC in Algorithm 1. In following, we present the initialization, particle-pattern, fitness function and router replacement methods. Initialization Our proposed system starts by generating an initial solution randomly, by ad hoc methods [14]. We decide the velocity of particles by a random process considering the area size. For √ instance, when √ the area size is W × H, the velocity is decided randomly from − W 2 + H 2 to W 2 + H 2 . Our system can generate many client distributions. In this paper, we consider Normal distribution of mesh clients as shown in Fig. 1.
Fig. 1. Normal distribution of mesh clients.
A Fast Convergence RDVM for Router Placement in WMNs
21
Particle-pattern A particle is a mesh router. A fitness value of a particle-pattern is computed by combination of mesh routers and mesh clients positions. In other words, each particle-pattern is a solution as shown is Fig. 2. Therefore, the number of particle-patterns is the number of solutions.
Fig. 2. Relationship among global solution, particle-patterns and mesh routers.
Fitness function One of most important thing is to decide the determination of an appropriate objective function and its encoding. In our case, each particle-pattern has an own fitness value and compares other particle-patterns fitness value in order to share information of global solution. The fitness function follows a hierarchical approach in which the main objective is to maximize the SGC in WMN. Thus, we use α and β weight-coefficients for the fitness function and the fitness function of this scenario is defined as: Fitness = α × SGC(x ij , y ij ) + β × NCMC(x ij , y ij ). Router replacement methods A mesh router has x, y positions and velocity. Mesh routers are moved based on velocities. There are many router replacement methods in PSO field [1,10– 12]. In this paper, we compare two replacement methods: Rational Decrement of Vmax Method (RDVM) and Fast Convergence RDVM (FC-RDVM). In RDVM, PSO parameters are set to unstable region (ω = 0.9, C1 = C2 = 2.0). A value of Vmax which is maximum velocity of particles is considered. The Vmax is kept decreasing with the increasing of iterations as shown in Eq. (1). Vmax (k) =
W 2 + H2 ×
T −k k
(1)
where W and H are the width and the height of the considered area, respectively. Also, T and k are the total number of iterations and a current number of iteration, respectively. The k is a variable updating from 1 to T by increment with increasing the iteration.
22
S. Sakamoto et al.
Fig. 3. The difference of Vmax between RDVM and FC-RDVM. Table 1. Parameter settings. Parameters
Values
Clients distribution
Normal distribution
Area size
32 × 32
Number of mesh routers
16
Number of mesh clients
48
Total iterations
800
Iteration per phase
4
Number of particle-patterns
9
Radius of a mesh router
From 2.0 to 3.0
Fitness function weight-coefficients (α, β) 0.7, 0.3 Replacement methods
RDVM, FC-RDVM
In FC-RDVM, the Vmax decreases with the increasing of iterations as shown in Eq. (2). T −k (2) Vmax (k) = W 2 + H 2 × T + γk where γ is a curvature parameter. When the γ is larger, the curvature is larger as shown in Fig. 3. Other parameters are the same with RDVM.
A Fast Convergence RDVM for Router Placement in WMNs
23
Fig. 4. Simulation results of WMN-PSOHC for SGC.
4
Simulation Results
In this section, we show simulation results using WMN-PSOHC hybrid intelligent system. In this work, we consider Normal distribution of mesh clients. We consider the number of particle-patterns 9. We conducted simulations 100 times in order to avoid the effect of randomness and create a general view of results. The total number of iterations is considered 800 and the iterations per phase is considered 4. We show the parameter setting for WMN-PSOHC in Table 1. We show the simulation results in Fig. 4 and Fig. 5. Considering SGC parameter, the RDVM performance is not good until the number of phases is 100. Then, the curve increases and reaches 100% when the number of phases is 140. The FC-RDVM (γ = 1.0) is faster than RDVM. However, it needs more than 50 phases to reach 100%. The performance of FC-RDVM (γ = 10.0) is faster than FC-RDVM (γ = 1.0) and RDVM, and it reaches 100% for 10 phases. Also, for NCMC, the FC-RDVM (γ = 10.0) has better performance than FC-RDVM (γ = 1.0) and RDVM.
24
S. Sakamoto et al.
Fig. 5. Simulation results of WMN-PSOHC for NCMC.
5
Conclusions
In this work, we evaluated the performance of WMNs by using a hybrid simulation system based on PSO and HC (called WMN-PSOHC). In this paper, we proposed and implemented a Fast Convergence RDVM (FC-RDVM). Then, we compared the performance of FC-RDVM with RDVM. Simulation results show that WMN-PSOHC performs better when FCRDVM is used compared with RDVM. Also, we conclude that FC-RDVM can control the speed of convergence by using the curvature parameter. In our future work, we would like to evaluate the performance of the proposed system for different parameters and scenarios.
References 1. Clerc, M., Kennedy, J.: The particle swarm-explosion, stability, and convergence in a multidimensional complex space. IEEE Trans. Evol. Comput. 6(1), 58–73 (2002) 2. Ozera, K., Bylykbashi, K., Liu, Y., Barolli, L.: A fuzzy-based approach for cluster management in VANETs: performance evaluation for two fuzzy-based systems. Internet of Things 3, 120–133 (2018) 3. Poli, R., Kennedy, J., Blackwell, T.: Particle swarm optimization. Swarm Intell. 1(1), 33–57 (2007) 4. Sakamoto, S., Lala, A., Oda, T., Kolici, V., Barolli, L., Xhafa, F.: Analysis of WMN-HC simulation system data using friedman test. In: The Ninth International Conference on Complex, Intelligent, and Software Intensive Systems (CISIS-2015), pp 254–259. IEEE (2015)
A Fast Convergence RDVM for Router Placement in WMNs
25
5. Sakamoto, S., Oda, T., Ikeda, M., Barolli, L., Xhafa, F.: Implementation and evaluation of a simulation system based on particle swarm optimisation for node placement problem in wireless mesh networks. Int. J. Commun. Networks Distributed Syst. 17(1), 1–13 (2016) 6. Sakamoto, S., Ozera, K., Ikeda, M., Barolli, L.: Implementation of intelligent hybrid systems for node placement problem in WMNs considering particle swarm optimization, hill climbing and simulated annealing. Mob. Networks Appl. 23(1), 27–33 (2017). https://doi.org/10.1007/s11036-017-0897-7 7. Sakamoto, S., Barolli, A., Barolli, L., Okamoto, S.: Implementation of a web interface for hybrid intelligent systems. Int. J. Web Inf. Syst. 15(4), 420–431 (2019) 8. Sakamoto, S., Barolli, L., Okamoto, S.: WMN-PSOSA: an intelligent hybrid simulation system for WMNs and its performance evaluations. Int. J. Web Grid Serv. 15(4), 353–366 (2019) 9. Sakamoto, S., Barolli, L., Okamoto, S.: A comparison study of linearly decreasing inertia weight method and rational decrement of VMAX method for WMNs using WMN-PSOHC intelligent system considering normal distribution of mesh clients. In: Barolli, L., Natwichai, J., Enokido, T. (eds.) EIDWT 2021. LNDECT, vol. 65, pp. 104–113. Springer, Cham (2021). https://doi.org/10.1007/978-3-03070639-5 10 10. Schutte, J.F., Groenwold, A.A.: A study of global optimization using particle swarms. J. Global Optim. 31(1), 93–108 (2005) 11. Shi, Y.: Particle swarm optimization. IEEE Connections 2(1), 8–13 (2004) 12. Shi, Y., Eberhart, R.C.: Parameter selection in particle swarm optimization. In: Evolutionary Programming VII, pp. 591–600 (1998) 13. Wang, J., Xie, B., Cai, K., Agrawal, D.P.: Efficient mesh router placement in wireless mesh networks. In: Proceedings of IEEE International Conference on Mobile Adhoc and Sensor Systems (MASS-2007), pp. 1–9 (2007) 14. Xhafa, F., Sanchez, C., Barolli, L.: Ad hoc and neighborhood search methods for placement of mesh routers in wireless mesh networks. In: Proceedings of 29th IEEE International Conference on Distributed Computing Systems Workshops (ICDCS2009), pp. 400–405 (2009)
Energy-Efficient Two-Phase Locking Protocol by Omitting Meaningless Read and Write Methods Tomoya Enokido1(B) , Dilawaer Duolikun2 , and Makoto Takizawa3 1
Faculty of Business Administration, Rissho University, 4-2-16, Osaki, Shinagawa-ku, Tokyo 141-8602, Japan [email protected] 2 Department of Advanced Sciences, Faculty of Science and Engineering, Hosei University, 3-7-2, Kajino-cho, Koganei-shi, Tokyo 184-8584, Japan 3 Research Center for Computing and Multimedia Studies, Hosei University, 3-7-2, Kajino-cho, Koganei-shi, Tokyo 184-8584, Japan [email protected] Abstract. In current information systems like the IoT, a huge volume of data is gathered from various types of devices and is encapsulated along with methods as an object. An application is composed of multiple objects allocated to servers. Multiple conflicting transactions issued by clients have to be serialized to keep every object mutually consistent. However, the throughput of a system decreases since the overhead to serialize conflicting transactions increases. In addition, it is critical to reduce the electric energy consumption of the system to keep the transaction system consistent. In our previous studies, the Energy-Efficient TwoPhase Locking (EE2PL) protocol is proposed to reduce the total electric energy consumption of servers and the execution time of each transaction by omitting meaningless write methods. In this paper, the Improved EE2PL (IEE2PL) protocol is newly proposed to furthermore reduce the total electric energy consumption of servers and the execution time of each transaction than the EE2PL protocol by omitting both meaningless read and write methods. Evaluation results show the total electric energy consumption of servers and the execution time of each transaction can be more reduced in the IEE2PL protocol than the EE2PL protocol. Keywords: Improved EE2PL (IEE2PL) protocol · Energy consumption · Concurrency control · Transactions · Object-based systems
1
Introduction
In current information systems, a huge number of IoT (Internet of Things) [1,2] devices are interconnected with various types of networks like wired and wireless networks. A huge volume of data is gathered from these IoT devices and is manipulated to provide various types of application services like vehicle network c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 26–36, 2022. https://doi.org/10.1007/978-3-031-08812-4_4
IEE2PL Protocol
27
services [3]. Data gathered from IoT devices is encapsulated with methods to manipulate the data as an object [4,5]. In object-based systems [4–6], an object is a unit of computation resource like a database system [7] and an application is composed of multiple objects distributed in multiple physical servers. A transaction [7,8] is an atomic sequence of methods to manipulate objects. Each transaction created on a client issues methods supported by each object to utilize an application service. Here, conflicting methods issued by multiple transactions have to be serialized [5–8] to keep all the objects mutually consistent. The Two-Phase Locking (2PL) protocol [8–10] is widely used to serialize multiple conflicting transactions in object-based systems. In the 2PL protocol, each transaction locks target objects before manipulating the objects. By using the 2PL protocol, multiple conflicting transactions can be serialized and objects can be kept mutually consistent. On the other hand, the throughput of a system decreases in the 2PL protocol since the overhead to lock objects increases if the more number of transactions are concurrently performed in the system. In addition, the more number of transactions are issued in a system, the larger electric energy is consumed in servers since every method issued by each transaction is surely performed on every target object and the execution time of each transaction increases. Hence, an energy-efficient concurrency control mechanism is required to not only reduce the total electric energy consumption of servers but also increase the throughput of a system as discussed in Green computing [6,10–12]. In our previous studies, meaningless write methods [10] which are not required to be performed on each object is defined based on the precedent relation among transactions and semantics of methods. Then, the Energy-Efficient Two-Phase Locking (EE2P L) protocol [10] is proposed to reduce the total electric energy consumption of servers by omitting meaningless write methods on each object. In this paper, we newly introduce meaningless read methods which can be omitted on each object. Then, the Improved EE2PL (IEE2PL) protocol is newly proposed to furthermore reduce the total electric energy consumption of servers and the execution time of each transaction by omitting meaningless read and write methods as well as meaningless write methods. We evaluate the IEE2PL protocol in terms of the total electric energy consumption of servers and the average execution time of each transaction compared with the EE2PL protocol. Evaluation results show the total electric energy consumption of servers and the average execution time of each transaction in the IEE2PL protocol can be more reduced than the EE2PL protocol. In Sect. 2, we present the object-based system, data access model, and power consumption model of a server. In Sect. 3, we propose the IEE2PL protocol. In Sect. 4, we evaluate the IEE2PL protocol compared with the EE2PL protocol.
2 2.1
System Model Object-Based Systems
An object-based system is composed of a set S of multiple physical servers s1 , ..., sn (n ≥ 1) and clients. In this paper, a term server means a physical server.
28
T. Enokido et al.
An application service is composed of a set O of objects o1 , ..., om (m ≥ 1). Each object oh [4] is an encapsulation of data and methods to manipulate the data in the object oh like a relational database system. Each object oh is allocated to a server st in a set S of servers. Methods supported by each object oh are classified into read (r) and write (w) methods. Write methods are furthermore classified into full write (wf ) and partial write (wp ) methods, i.e. w = {wf , wp }. In this paper, we assume a whole data in an object oh is fully written by a full write method. A part of data in an object oh is written by a partial write method. Data in an object oh is fully read by a read method. Suppose a file object F supports create, drop, modify, and read methods. A pair of create and drop methods are full write methods. A modify method is a partial write method. Let op(oh ) be a state obtained by performing a method op (∈ {wf , wp , r}) on an object oh . Let opi ◦ opj (oh ) be a state of an object oh by performing a method opi before another method opj . A pair of methods opi and opj on an object oh are compatible if and only if (iff) opi ◦ opj (oh ) = opj ◦ opi (oh ). Otherwise, a method opi conf licts with another method opj on an object oh . In this paper, we assume conflicting relations among methods are given as shown in Table 1. Table 1. Conflicting relation among methods. Read (r)
Write (w) Full (wf ) Partial (wp )
×
×
Write (w) Full (wf ) × Partial (wp ) ×
× ×
× ×
Read (r)
2.2
Two-Phase Locking (2PL) Protocol
A transaction T i [8] is an atomic sequence of read (r) and write (w) methods to manipulate objects. Let T be a set {T 1 , ..., T k } (k ≥ 1) of transactions issued in a system. Multiple conflicting transactions are required to be serializable [7,8] to keep all the objects mutually consistent. Let H be a schedule of transactions in T. A transaction T i precedes another transaction T j (T i →H T j ) in a schedule H iff a method opi issued by the transaction T i is performed before another method opj issued by the transaction T j and the method opi conflicts with the method opj . A schedule H is serializable iff the precedent relation →H is acyclic [8]. The T wo-P hase Locking (2P L) protocol [5,7–9] is proposed to serialize multiple conflicting transactions. In this paper, multiple conflicting transactions are serialized based on the 2PL protocol. Let μ(op) be a lock mode of a method op. A pair of lock modes μ(opi ) and μ(opj ) are compatible iff a pair of methods opi and opj are compatible on an object oh . Otherwise, a lock mode μ(opi ) conflicts with another lock mode μ(opj ) on an object oh . In the 2PL protocol, a
IEE2PL Protocol
29
transaction T i locks and manipulates each object oh by the following procedure [5,7–9]: 1. If an object oh can be locked in a lock mode μ(op), the object oh is manipulated by the method op. 2. Once a transaction T i releases a lock on an object, the transaction T i does not lock any object until the transaction T i commits or aborts. When the transaction T i commits or aborts, the lock on the object oh is released. 2.3
Data Access Model
Methods which are being performed on an object are current at time τ . Let RPt (τ ) and W Pt (τ ) be sets of current read (r) and write (w) methods on a server st at time τ , respectively. A notation Pt (τ ) shows a set of current read and write methods on a server st at time τ , i.e. Pt (τ ) = RPt (τ ) ∪ W Pt (τ ). Let rti (oh ) and wti (oh ) be methods issued by a transaction T i to read and write data in an object oh on a server st , respectively. Each read method rti (oh ) in a set RPt (τ ) reads data in an object oh at rate RRti (τ ) [Byte/sec (B/sec)] at time τ . Each write method wti (oh ) in a set W Pt (τ ) writes data in an object oh at rate W Rti (τ ) [B/sec] at time τ . Let maxRRt and maxW Rt be the maximum read and write rates [B/sec] of a server st , respectively. The read rate RRti (τ ) (≤ maxRRt ) and write rate W Rti (τ ) (≤ maxW Rt ) are drt (τ ) · maxRRt and dwt (τ ) · maxW Rt , respectively. Here, drt (τ ) and dwt (τ ) are degradation ratios of the read rate RRti (τ ) and the write rate W Rti (τ ), respectively. The degradation ratio drt (τ ) is 1 / (|RPt (τ )| + rwt · |W Pt (τ )|) where 0 ≤ rwt ≤ 1. The degradation ratio dwt (τ ) is 1/(wrt · |RPt (τ )| + |W Pt (τ )|) where 0 ≤ wrt ≤ 1. Here, 0 ≤ drt (τ ) ≤ 1 and 0 ≤ dwt (τ ) ≤ 1. The read laxity rlti (τ ) [B] and write laxity wlti (τ ) [B] of methods rti (oh ) and i wt (oh ) show how much amount of data are to be read and written in an object oh by the methods rti (oh ) and wti (oh ), respectively, at time τ . Suppose that methods rti (oh ) and wti (oh ) start on a server st at time rstit and wstit , respectively. At time rstit , the read laxity rlti (τ ) = rbh [B] where rbh is the size of data in an object oh . The write laxity wlti (τ ) = wbh [B] where wbh is the size of data to be written in an object oh at time wstit . The read laxity rlti (τ ) and write laxity wlti (τ ) at time τ are rbh - Σττ=rsti RRti (τ ) and wbh - Σττ=wsti W Rti (τ ), respectively. t
2.4
t
Power Consumption Model of a Server
The Power Consumption model for a Storage server (PCS model) [13] to perform storage and computation processes is proposed in our previous studies. A notation Et (τ ) shows the electric power [W] of a server st at time τ . maxEt and minEt show the maximum and minimum electric power [W] of the server st , respectively. In this paper, we assume only read and write methods are performed on a server st . According to the PCS model [13], the electric power Et (τ ) [W] of a server st to perform multiple read and write methods at time τ is given as Eq. (1).
30
T. Enokido et al.
⎧ W Et ⎪ ⎪ ⎪ ⎨W RE (α) t Et (τ ) = ⎪ RE t ⎪ ⎪ ⎩ minEt
if if if if
|W Pt (τ )| ≥ 1 and |RPt (τ )| = 0. |W Pt (τ )| ≥ 1 and |RPt (τ )| ≥ 1. |W Pt (τ )| = 0 and |RPt (τ )| ≥ 1. |W Pt (τ )| = |RPt (τ )| = 0.
(1)
The electric power consumed in a server st is the minimum electric power minEt [W] if no method is performed on the server st . The server st consumes the electric power REt [W] iff only read methods are performed on the server st . The server st consumes the electric power W Et [W] iff only write methods are performed on the server st . The server st consumes the electric power W REt (α) [W] = α · REt + (1 - α) · W Et [W] where α = |RPt (τ )|/(|RPt (τ )| + |W Pt (τ )|) if both at least one read method and at least one write method are concurrently performed. Here, minEt ≤ REt ≤ W REt (α) ≤ W Et ≤ maxEt . The total electric energy T EEt (τ1 , τ2 ) [J] of a server st from time τ1 to τ2 is Σττ2=τ1 Et (τ ). The processing electric power P EPt (τ ) [W] of a server st at time τ is Et (τ ) minEt . The total processing electric energy T P EEt (τ1 , τ2 ) of a server st from time τ1 to τ2 is given as T P EEt (τ1 , τ2 ) = Σττ2=τ1 P EPt (τ ).
3 3.1
Improved EE2PL (IEE2PL) Protocol Meaningless Read and Write Methods
We define meaningless read and write methods which are not required to be performed on each object. A method op1 precedes op2 in a schedule H (op1 →H op2 ) iff (1) the method op1 is issued before op2 by a same transaction T i , (2) the method op1 issued by a transaction T i conflicts with the method op2 issued by a transaction T j and T i →H T j , or (3) op1 →H op3 →H op2 for some method op3 . Let Hh be a local schedule of methods which are performed on an object oh in a schedule H. A method op1 locally precedes another method op2 in a local schedule Hh (op1 →Hh op2 ) iff op1 →H op2 . Suppose a partial write method wp (oh ) locally precedes another full write method wf (oh ) in a local schedule Hh (wp (oh ) →Hh wf (oh )) on an object oh . Here, the partial write method wp (oh ) is not required to be performed on the object oh if the full write method wf (oh ) is surely performed on the object oh just after the partial write method wp (oh ), i.e. the full write method wf (oh ) can absorb the partial write method wp (oh ). [Absorption of Write Methods] A full write method op1 absorbs another partial or full write method op2 in a local subschedule Hh of an object oh iff one of the following conditions is hold: 1. op2 →Hh op1 and there is no read method op such that op2 →Hh op →Hh op1 . 2. op1 absorbs op and op absorbs op2 for some method op .
IEE2PL Protocol
31
[Absorption of Read Methods] A read method op1 absorbs another read method op2 in a local subschedule Hh of an object oh iff one of the following conditions is hold: 1. op1 →Hh op2 and there is no write method op such that op1 →Hh op →Hh op2 . 2. op1 absorbs op and op absorbs op2 for some method op . [Meaningless Methods]. A method op is meaningless iff the method op is absorbed by another method op in the local subschedule Hh on an object oh . 3.2
IEE2PL Protocol
In our previous studies, the Energy-Efficient Two-Phase Locking (EE2P L) protocol [10] is proposed to reduce not only the total electric energy consumption of servers but also the average execution time of each transaction by omitting only meaningless write methods on each object. In this paper, the Improved EE2PL (IEE2PL) protocol is proposed to furthermore reduce the total electric energy consumption of servers and the average execution time of each transaction by omitting meaningless read and write methods. Suppose an object oh hold by a server st is locked by a transaction T i with a lock mode μ(r) and a read method rti (oh ) is performed on the object oh as shown in Fig. 1. Another transaction T j locks the object oh with lock mode μ(r) and issues a read method rtj (oh ) to the object oh . Here, the read method rtj (oh ) is a meaningless method since the read method rti (oh ) issued by the transaction T i is being performed on the object oh and the read method rti (oh ) absorbs the read method rtj (oh ). Hence, the read method rtj (oh ) is not performed on the object oh and a result obtained by performing the read method rti (oh ) is sent to a pair of transactions Ti and Tj . Ti
Tj
oh lock
rti(oh)
lock j
rt (oh) result
the read method issues by j the transaction T is not performed.
result
time
Fig. 1. A meaningless read method.
Suppose a transaction T i locks an object oh allocated to a server st with a lock mode μ(w) and issues a partial write method wtpi (oh ) as shown in Fig. 2. The object oh sends a termination notification of the partial method wtpi (oh ) to the transaction T i as soon as the object oh receives the partial write method
32
T. Enokido et al.
wtpi (oh ) but the partial write method wtpi (oh ) is not performed until the next method is performed on the object oh . Suppose another transaction T j locks the object oh with a lock mode μ(w) after the transaction T i commits and issues a full write methods wtf j (oh ). Here, the partial write method wtpi (oh ) issued by the transaction T i is meaningless since the full write method wtf j (oh ) issued by the transaction T j absorbs the partial write method wtpi (oh ) on the object oh . Hence, the full write method wtf j (oh ) is performed on the object oh without performing the partial write method wtpi (oh ).
T
i
oh lock pi
notification commit
wt (oh) the partial write method is not performed until the next method is performed.
time
T
j
lock the full write method is performed without performing the previous partial write method.
fj
wt (oh) notification commit time
Fig. 2. A meaningless write method.
Suppose a transaction T j locks the object oh with a lock mode μ(r) after the transaction T i commits and issues a read method rtj (oh ) in Fig. 2. Here, the partial write method wtpi (oh ) issued by the transaction T i has to be performed before the read method rtj (oh ) is performed since the read method rtj (oh ) has to read data written by the partial write method wtpi (oh ). Let oh .Cr be a read method rti (oh ) issued by a transaction T i , which is being performed on a object oh . A notation oh .Dw shows a write method wti (oh ) issued by a transaction T i to write data of an object oh in a server st , which is waiting for the next method op to be performed on the object oh . Suppose a transaction T i issues a method op to an object oh . In the IEE2PL protocol, the method op is performed on the object oh by the following IEE2PL procedure:
IEE2PL Protocol
33
IEE2PL(op(oh )) { if op(oh ) = r, { /* op is a read method. */ if oh .Dw = φ, { if oh .Cr = φ, { oh .Cr = op(oh ); perform(op(oh )); oh .Cr = φ; } else a result of oh .Cr is sent to a transaction T i ; } else { perform(oh .Dw); oh .Dw = φ; oh .Cr = op(oh ); perform(op(oh )); oh .Cr = φ; } } else { /* op is a write method. */ if oh .Dw = φ, oh .Dw = op(oh ); else { /* oh .Dw = φ */ if op(oh ) absorbs oh .Dw, oh .Dw = op(oh ); /* oh .Dw is omitted. */ else { perform(oh .Dw); oh .Dw = op(oh ); } } } } In the IEE2PL protocol, the total electric energy consumption of servers and the execution time of each transaction can be more reduced than the EE2PL protocol since not only the meaningless write methods but also the meaningless read methods are omitted on each object.
4 4.1
Evaluation Environment
We evaluate the IEE2PL protocol in terms of the total processing electric energy consumption of a set S of servers and the average execution time of each transaction compared with the EE2PL protocol [10]. A homogeneous set S of servers is composed of ten servers s1 , ..., s10 (n = 10) and every server st (t = 1, ..., 10) follows the same data access model and power consumption model as shown in Table 2. Parameters of each server st are given based on the experimentations [13]. There are thirty objects o1 , ..., o30 and each object oh (h = 1, ..., 30) is randomly allocated to a server st in the set S of servers. The size of data in
34
T. Enokido et al. Table 2. A homogeneous set S of servers (t = 1, ..., 10). Server st maxRRt st
maxW Rt
rwt wrt minEt W Et
80 [MB/sec] 45 [MB/sec] 0.5
REt
0.5 39 [W] 53 [W] 43 [W]
each object oh is randomly selected between 50 and 100 [MByte]. Each object oh supports read (r), full write (wf ), and partial write (wp ) methods. The number m (0 ≤ m ≤ 1,000) of transactions are issues to manipulate objects. Each transaction issues three methods randomly selected from ninety methods on the thirty objects. The total amount of data of an object oh is fully read and written by each read (r) and full write (wf ) methods. On the other hand, a half size of data of an object oh is written by each partial write (wp ) method. The starting time of each transaction T i is randomly selected in a unit of one second between 1 and 360 [sec]. 4.2
Total Processing Electric Energy Consumption
Total electric energy consumption [KJ]
Figure 3 shows the total processing electric energy consumption [KJ] of the set S of servers to perform the number m of transactions in the EE2PL and IEE2PL protocols. For 0 ≤ m ≤ 1,000, the total processing electric energy consumption of the set S of servers can be reduced by about 18% in the IEE2PL protocol compared with the EE2PL protocol. In the IEE2PL protocol, not only meaningless write methods but also meaningless read methods are omitted on each object. As a result, the total processing electric energy consumption of the set S of servers can be more reduced in the IEE2PL protocol than the EE2PL protocol. 250 EE2PL IEE2PL
200 150 100 50 0
0
200
400 600 Number m of transactions
800
1000
Fig. 3. Total processing electric energy consumption [KJ] of a set S of servers.
IEE2PL Protocol
4.3
35
Average Execution Time of Each Transaction
Average execution time [sec] of each transaction
Figure 4 shows the average execution time [sec] of the m transactions in the EE2PL and IEE2PL protocols. In the EE2PL and IEE2PL protocols, the average execution time of each transaction increases as the total number m of transactions increases since more number of transactions are concurrently performed. For 0 < m ≤ 1,000, the average execution time of each transaction can be reduced by about 13% in the IEE2PL protocol compared with the EE2PL protocol. In the IEE2PL protocol, each transaction can commit without performing meaningless methods. Hence, the average execution time of each transaction can be more reduced in the IEE2PL protocol than the EE2PL protocol.
5.5 EE2PL IEE2PL
5 4.5 4 3.5 3 2.5
0
200
400 600 Number m of transactions
800
1000
Fig. 4. Average execution time [sec] of each transaction.
Following the evaluation, the total processing electric energy consumption of servers and the average execution time of each transaction in the IEE2PL protocol can be more reduced than the EE2PL protocol. Hence, the IEE2PL protocol is more useful than the EE2PL protocol.
5
Concluding Remarks
In this paper, we newly proposed the IEE2PL (Improved EE2PL) protocol to furthermore reduce the total processing electric energy consumption of servers and the average execution time of each transaction than the EE2PL protocol by omitting meaningless read and write methods. We evaluated the IEE2PL protocol compared with the EE2PL protocol. The evaluation results show the total processing electric energy consumption of servers and the average execution time of each transaction can be reduced by 18% and 13%, respectively, in the IEE2PL protocol compared with the EE2PL protocol. Following the evaluation, the IEE2PL protocol is more useful than the EE2PL protocol in a homogeneous set of servers.
36
T. Enokido et al.
References 1. Nakamura, S., Enokido, T., Takizawa, M.: Implementation and evaluation of the information flow control for the Internet of Things. Concurrency and Computation: Practice and Experience, vol. 33, no. 19 (2021) 2. Enokido, T. and Takizawa, M.: The redundant energy consumption laxity based algorithm to perform computation processes for IoT services. Internet of Things, vol. 9, (2020). https://doi.org/10.1016/j.iot.2020.100165 3. Bylykbashi, K., Qafzezi, E., Ampririt, P., Ikeda, M., Matsuo, K., and Barolli, L.: Effect of vehicle technical condition on real-time driving risk management in Internet of Vehicles: design and performance evaluation of an integrated fuzzybased system. Internet of Things, vol. 13 (2021). https://doi.org/10.1016/j.iot. 2021.100363 4. Object Management Group Inc.: Common object request broker architecture (CORBA) specification, version 3.3, part 1 – interfaces. (2012). http://www.omg. org/spec/CORBA/3.3/Interfaces/PDF 5. Tanaka, K., Hasegawa, K., Takizawa, M.: Quorum-based replication in objectbased systems. J. Inf. Sci. Eng. 16(3), 317–331 (2000) 6. Enokido, T., Duolikun, D., Takizawa, M.: Energy consumption laxity-based quorum selection for distributed object-based systems. Evol. Intel. 13, 71–82 (2020) 7. Gray, J. N.: Notes on database operating systems. In: Lecture Notes in Computer Science, vol. 60, pp. 393–481 (1978) 8. Bernstein, P.A., Hadzilacos, V., Goodman, N.: Concurrency control and recovery in database systems. Addison-Wesley (1987) 9. Garcia-Molina, H., Barbara, D.: How to assign votes in a distributed system. J. ACM 32(4), 814–860 (1985) 10. Enokido, T., Duolikun, D., Takizawa, M.: Energy-efficient concurrency control by omitting meaningless write methods in object-based systems. accepted for publication in Proceedings of the 36th International Conference on Advanced Information Networking and Applications (AINA-2022) (2022) 11. Natural Resources Defense Council (NRDS): Data center efficiency assessment - scaling up energy efficiency across the data center lndustry: Evaluating key drivers and barriers (2014). http://www.nrdc.org/energy/files/data-centerefficiency-assessment-IP.pdf 12. Enokido, T., Duolikun, D., Takizawa, M.: The improved redundant active timebased (IRATB) algorithm for process replication Proceedings of the 35th IEEE International Conference on Advanced Information Networking and Applications (AINA-2021), pp. 172–180 (2021) 13. Sawada, A., Kataoka, H., Duolikun, D., Enokido, T., and Takizawa, M.: Energyaware clusters of servers for storage and computation applications. In: Proceedings of the 30th IEEE International Conference on Advanced Information Networking and Applications (AINA-2016), pp. 400–407 (2016)
A New Method for Optimization of Number of Mesh Routers and Improving Cost Efficiency in Wireless Mesh Networks Aoto Hirata1 , Tetsuya Oda2(B) , Yuki Nagai1 , Tomoya Yasunaga1 , Nobuki Saito1 , Kengo Katayama2 , and Leonard Barolli3 1
3
Graduate School of Engineering, Okayama University of Science (OUS), Okayama, 1-1 Ridaicho, Kita-ku, Okayama 700–0005, Japan {t21jm02zr,t21jm01md,t22jm23rv,t22jm43sx}@ous.jp 2 Department of Information and Computer Engineering, Okayama University of Science (OUS), 1-1 Ridaicho, Kita-ku, Okayama 700–0005, Japan {oda,katayama}@ice.ous.ac.jp Department of Information and Communication Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected] Abstract. The Wireless Mesh Networks (WMNs) enable routers to communicate with each other wirelessly in order to create a stable network over a wide area at a low cost and it has attracted much attention in recent years. There are different methods for optimizing the placement of mesh routers. In our previous work, we proposed a Coverage Construction Method (CCM), CCM-based Hill Climbing (HC) and CCM-based Simulated Annealing (SA) system for mesh router placement problem considering normal and uniform distributions of mesh clients. We also proposed Delaunay edge and CCM-based SA. In this approach, we consider a realistic scenario for mesh client placement rather than randomly generated mesh clients with normal or uniform distributions. However, this approach required many mesh routers to cover mesh clients located over a wide area. In this paper, we propose a method for optimization of number of mesh routers in WMNs. For the simulations, we consider the evacuation areas in Okayama City, Japan, as the target to be covered by mesh routers. From the simulation results, we found that the proposed method was able to cover the evacuation area. The proposed method also reduced the number of mesh routers by an average of 28 [%].
1
Introduction
The Wireless Mesh Networks (WMNs) [1–4] are wireless network technologies that enables routers to communicate with each other wirelessly to create a stable network over a wide area at a low cost and it has attracted much attention in recent years. The placement of the mesh routers has a significant impact on cost, communication range and operational complexity. Therefore, there are c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 37–48, 2022. https://doi.org/10.1007/978-3-031-08812-4_5
38
A. Hirata et al.
many research works to optimize the placement of these mesh routers. In our previous work [5–16], we proposed and evaluated different meta-heuristics such as Genetic Algorithms (GA) [17], Hill Climbing (HC) [18], Simulated Annealing (SA) [19], Tabu Search (TS) [20] and Particle Swarm Optimization (PSO) [21] for mesh router placement optimization. Also, we proposed a Coverage Construction Method (CCM) [22], CCM-based Hill Climbing (HC) [23] and CCM-based Simulated Annealing (SA) system. The CCM is able to rapidly create a group of mesh routers with the radio communication ranges of the mesh routers linked to each other. The CCM-based HC system covered many mesh clients generated by normal and uniform distributions. We also showed that in the two islands model, the CCM-based HC system was able to find two islands and covered many mesh clients [24]. The CCM-based HC system adapted to varying number of mesh clients, number of mesh routers and area size [25,26]. The CCM-based SA system [27] was able to cover many mesh clients in normal distribution compared with CCM. We also proposed a Delaunay edge and CCM-based SA approach [28] that focuses on a more realistic mesh clients placement. However, this approach required many mesh routers to cover mesh clients located over a wide area. In this paper, we propose a method for optimization of number of mesh routers in WMNs. As evaluation metrics, we consider the Size of Giant Component (SGC), the Number of Covered Mesh Clients (NCMC) and Number of Mesh Routers (NMR). The structure of the paper is as follows. In Sect. 2, we give a short description of mesh router placement problem. In Sect. 3, we present the proposed method. In Sect. 4, we discuss the simulation results. Finally, in Sect. 5, we conclude the paper and give future research directions.
2
Mesh Router Placement Problem
We consider a two-dimensional continuous area to deploy a number of mesh routers and a number of mesh clients of fixed positions. The objective of the problem is to optimize a location assignment for the mesh routers to the twodimensional continuous area that maximizes the network connectivity and mesh clients coverage. Network connectivity is measured by the SGC, while the NCMC is the number of mesh clients that is within the radio communication range of at least one mesh router. An instance of the problem consists as follows. • An area W idth × Height which is the considered area for mesh router placement. Positions of mesh routers are not pre-determined and are to be computed. • The mesh router has radio communication range defining thus a vector of routers. • The mesh clients are located in arbitrary points of the considered area defining a matrix of clients.
A New Method for Optimization of Number of Mesh Routers in WMNs
39
Algorithm 1. Method for randomly generating mesh routers. Output: Placement list of mesh routers. 1: Set N umber of mesh routers. 2: Generate mesh router [0] randomly in considered area. 3: i ← 1. 4: while i < N umber of mesh routers do 5: Generate mesh router [i] randomly in considered area. 6: if SGC is maximized then 7: i ← i + 1. 8: else 9: Delete mesh router [i]. 10: end if 11: end while
3
Proposed Method
In this section, we describe the proposed method. In Algorithm 1, Algorithm 2 and Algorithm 3 are shown pseudo codes of CCM and CCM-based SA. 3.1
CCM for Mesh Router Placement Optimization
In our previous work, we proposed a CCM [22] for mesh router placement optimization problem. The pseudo code of randomly generating mesh routers method used in the CCM is shown in Algorithm 1 and the pseudo code of the CCM is shown in Algorithm 2. The CCM searches the solution with maximized SGC. Among the solutions generated, the mesh router placement with the highest NCMC is the final solution. We describe the operation of the CCM in follow. First, the mesh clients are generated in the considered area. Next, randomly is determined a single point coordinate to be mesh router 1. Once again, randomly determine a single point coordinate to be mesh router 2. Each mesh router has a radio communication range. If the radio communication ranges of the two routers do not overlap, delete router 2 and randomly determine a single point coordinate and make it as mesh router 2. This process is repeated until the radio communication ranges of two mesh routers overlaps. If the radio communication ranges of the two mesh routers overlap, generate next mesh routers. If there is no overlap in radio communication range with any mesh router, the mesh router is removed and generated randomly again. If any of the other mesh routers have overlapping radio communication ranges, generate next mesh routers. Continue this process until the setting number of mesh routers.
40
A. Hirata et al.
Algorithm 2. Coverage construction method. Input: Placement list of mesh clients. Output: Placement list of best mesh routers. 1: Set N umber of loop f or CCM . 2: i, Current N CM C, Best N CM C ← 0. 3: Current mesh routers ← Alg. 1. 4: Best mesh routers ← Current mesh routers. 5: while i < N umber of loop f or CCM do 6: Current N CM C ← N CM C of Current mesh routers. 7: if Current N CM C > Best N CM C then 8: Best N CM C ← Current N CM C. 9: Best mesh routers ← Current mesh routers. 10: end if 11: i ← i + 1. 12: Current mesh routers ← Alg. 1. 13: Current N CM C ← 0. 14: end while
By this procedure is created a group of mesh routers connected together without the derivation of connected component using Depth First Search (DFS) [29]. However, this method only creates a population of mesh routers at a considered area, but does not take into account the location of mesh clients. So, the procedure should be repeated for a setting number of loops. Then, determine how many mesh clients are included in the radio communication range group of the mesh router. The placement of the mesh router with the highest number of mesh clients covered during the iterative process is the solution of the CCM. 3.2
CCM-based SA for Mesh Router Placement Optimization
In this subsection, we describe the CCM-based SA. The pseudo code of the CCMbased SA for mesh router placement problem is shown in Algorithm 3. The SA is one of the local search algorithms, which is inspired by the cooling process of metals. In SA, local solutions are derived by transitioning states and repeating the search for neighboring solutions. SA also transitions states, according to the decided state transition probability if the current solution is worse than the previous one. SA requires solution evaluation and temperature values to decide state transition probabilities. The evaluation of placement (Eval), the temperature (T ) and the state transition probability (ST P ) in the proposed method are shown in Eq. (1), Eq. (2) and Eq. (3). Eval ← 10 × (N CM C with the best results so f ar − N CM C of current solution) (1)
A New Method for Optimization of Number of Mesh Routers in WMNs
41
Algorithm 3. CCM-based SA. Input: Placement list of mesh clients. Output: Placement list of best mesh routers. 1: Set N umber of loop f or SA, Initial T emperature, F inal T emperature. 2: Current number of loop ← 0. 3: Current mesh routers, Best mesh routers ← Alg. 2 (P lacement list of mesh clients). 4: Current N CM C, Best N CM C ← N CM C of Current mesh routers. 5: while Current number of loop < N umber of loop f or SA do 6: Randomly choose an index of Current mesh routers. 7: Randomly change coordinate of Current mesh router [choosed index]. 8: Current N CM C ← N CM C of Current mesh router. 9: if SGC is maximized then 10: r ← Randomly generate in (0.0, 100.0). 11: Eval ← 10 × (Best N CM C − Current N CM C) 12: T ← Initial T emperature + (F inal T emperature − Initial T emperature) × Current number of loops N umber of loops f or SA − Eval T
13: if e ≥ 1.0 then 14: Best N CM C ← Current N CM C. 15: Best mesh routers ← Current mesh routers. Eval 16: else if e− T > r then 17: Best N CM C ← Current N CM C. 18: Best mesh routers ← Current mesh routers. 19: else 20: Restore coordinate of Current mesh routers [choosed index]. 21: end if 22: else 23: Restore coordinate of Current mesh routers [choosed index]. 24: end if 25: Current number of loops ← Current number of loops + 1. 26: end while
T ← Initial T emperature + Current number of loops (F inal T emperature − Initial T emperature) × N umber of loops f or SA
ST P ← e−
Eval T
(2) (3)
Proposed method performs neighborhood search by changing the placement of one mesh router. The solution is basically updated when the SGC is maximized and the NCMC is larger than the previous one. In SA, the solution is also updated depending on the ST P when the SGC is maximized but the NCMC is decreased. We describe the operation of the proposed method in following. First, we randomly select one of the mesh routers in the group of mesh routers as the initial solution obtained by the CCM and change the placement of the chosen mesh router randomly. Then, we decide the NCMC for all mesh routers
42
A. Hirata et al.
and derive the SGC value. The SGC is derived by creating an adjacency list of mesh routers and using DFS to confirm the connected mesh routers. If the SGC is maximized and the NCMC is greater than the previous one, then the changed placement of mesh routers is the current solution. If the SGC is maximized but the NCMC is less than that of previous NCMC, then restore the placement of changed mesh router. But in this case, the changed placement of mesh routers is the current solution with a probability of ST P [%]. This process is repeated until the setting number of loops. 3.3
Delaunay Edge and CCM-based SA
In this subsection, we describe the Delaunay edge and CCM-based SA. In the previous methods, the purpose of simulations was to cover mesh clients that were randomly generated based on normal or uniform distributions. In the previous work, we also proposed Delaunay edge and CCM-based SA as a mesh router placement optimization method focusing on more realistic scenarios. In this method, Voronoi decomposition is performed to divide the regions where mesh clients in the problem area are close to each other before performing the CCM. Each region obtained by Voronoi decomposition is called a Voronoi cell and the mesh clients of each Voronoi cell are connected by lines based on the adjacency of these Voronoi cells. This line is called a Delaunay edge. To use this Delaunay edge in the proposed system, the coordinates of the Delaunay edge are derived based on the pixel information in the image. The coordinates of the listed Delaunay edges are considered as the possible regions in the CCM. By generating an initial solution with mesh routers placed in the middle of a groups of mesh clients by CCM, we can discover the distant mesh clients by the SA approach. 3.4
Number of Mesh Routers Optimization Method
In this subsection, we describe a method for optimization of number of mesh routers in the Delaunay edge and CCM-based SA. Delaunay edge and CCMbased SA systems are needed many mesh routers to cover mesh clients located over a wide area. The proposed system improves cost efficiency by reducing the number of non-essential mesh routers after mesh router placement optimization. The pseudo code of the NMR optimization method is shown in Algorithm 4. This method optimizes the NMR by removing the mesh routers that do not affect the NCMC and SGC. The NMR can also be reduced more efficiently by shuffling the mesh routers and repeating the operation.
A New Method for Optimization of Number of Mesh Routers in WMNs
43
Algorithm 4. NMR Optimization Method. Input: Placement list of mesh clients, Placement list of best mesh routers, Best NCMC. Output: Placement list of reduced mesh routers. 1: Set N umber of loop f or N M R Optimization M ethod. 2: Current N CM C, Current number of loop ← 0. 3: Current mesh routers ← P lacement list of best mesh routers. 4: while Current number of loop < N umber of loop f or N M R Optimization M ethod do 5: Shuffle the list of current mesh routers. 6: n ← 0. 7: while n < N M R of current mesh routers do 8: Remove the mesh router with array number n from the current mesh routers. 9: if SGC is maximized then 10: if Current N CM C = Best N CM C then 11: Best mesh routers ← Current mesh routers. 12: else 13: Restore the removed mesh router. 14: end if 15: else 16: Restore the removed mesh router. 17: end if 18: n ← n + 1. 19: end while 20: Current number of loop ← Current number of loop + 1. 21: end while
Table 1. Parameters and values for simulations. Width of considered area
260
Hight of considered area
180
Default number of mesh routers
256
Radius of radio communication range of mesh routers 4 Number of mesh clients
3089
Number of lop for CCM
3000
Number of loop for SA
100000
Number of loop for NMR Optimization Method
10
Initial temperature
100
Final temperature
1
44
A. Hirata et al.
(a) Original map.
(b) Evacuation area painted red.
(c) Voronoi edge.
(d) Delaunay edge.
Fig. 1. Visualization of evacuation area.
(a) Converted evacuation area.
(b) Converted Delaunay edge.
Fig. 2. Converted information for the proposed system. Table 2. Simulation results of Delaunay edge based CCM and Delaunay edge and CCM-based SA. Method
Best SGC
Average SGC
Best NCMC
Delaunay edge based CCM
256
256
2118
Average NCMC [%] 64.894
Delaunay edge and CCM-based SA
256
256
3089
94.201
A New Method for Optimization of Number of Mesh Routers in WMNs
45
Table 3. Simulation results of NMR optimization method. Method
Best NMR Average NMR
Proposed system 171
(a) Result of Delaunay edge based CCM.
184.710
(b) Result of Delaunay edge and CCM-based SA.
(c) Result of NMR optimization method.
Fig. 3. Visualization results.
4
Simulation Results
In this section, we present a simulation results. The parameters used for simulation are shown in Table 1. We deployed mesh clients based on geographic information. The placement area is around Okayama Station in Okayama City, Okayama Prefecture, Japan. The mesh clients are the buildings used as evacuation area. We used the GIS application, QGIS, to display the geographic information. We also used the shapefiles of buildings from Open Street Map and the information of evacuation areas from the open data released by Okayama City [30]. The original map image is shown in Fig. 1(a). The red points in Fig. 1(a) indicate the points where evacuation areas are located. Figure 1(b) shows buildings designated as evacuation areas painted in red. These red buildings are considered mesh clients. Figure 1(c) and Fig. 1(d) show the derived Voronoi edge and the Delaunay edge. The Delaunay edges used in the proposed system are con-
46
A. Hirata et al.
verted into information that can be used in the proposed system by extracting color pixels from images. In Fig. 2(a), we show the converted evacuation area. Figure 2(b) shows the converted Delaunay edge, which becomes placement area of the CCM in the proposed system. We performed the simulation 100 times for each method. The simulation results are shown in Table 2 and Table 3. We show the simulation results of the best SGC, avg. SGC, best NCMC, avg. NCMC, best NMR and avg. NMR. In all simulations, the SGC is maximized for each method and the method using SA covered most of the mesh clients. The proposed system was also able to reduce the NMR by 28 [%] on average using the NMR optimization method. The visualization results of each method are shown in Fig. 3.
5
Conclusions
In this paper, we proposed a method for optimization of number of mesh routers in WMNs. We evaluated the proposed method in simulations. The simulation results show that the proposed method was able to cover many mesh clients and reduce the number of mesh routers by an average of 28 [%]. In the future, we would like to consider other local search algorithms and Genetic Algorithms. Acknowledgement. This work was supported by JSPS KAKENHI Grant Number JP20K19793.
References 1. Akyildiz, I.F., et al.: Wireless mesh networks: a survey. Comput. Netw. 47(4), 445–487 (2005) 2. Oda, T., et al.: Implementation and experimental results of a WMN testbed in indoor environment considering LoS scenario. In: Proceedings of the IEEE 29-th International Conference on Advanced Information Networking and Applications (IEEE AINA-2015), pp. 37-42 (2015) 3. Jun, J., et al.: The nominal capacity of wireless mesh networks. IEEE Wirel. Commun. 10(5), 8–15 (2003) 4. Oyman, O., et al.: Multihop relaying for broadband wireless mesh networks: from theory to practice. IEEE Commun. Mag. 45(11), 116–122 (2007) 5. Oda, T., et al.: Evaluation of WMN-GA for different mutation operators. Int. J. Space-Based Situated Comput. 2(3) (2012) 6. Oda, T., et al.: Performance evaluation of WMN-GA for different mutation and crossover rates considering number of covered users parameter. Mob. Inf. Syst. 8(1), 1–16 (2012) 7. Oda, T., et al.: WMN-GA: a simulation system for WMNs and its evaluation considering selection operators. J. Ambient. Intell. Humaniz. Comput. 4(3), 323– 330 (2013) 8. Oda, T., et al.: Node placement in WMNs using WMN-GA system considering uniform and normal distribution of mesh clients. In: Proceedings of the IEEE 8-th International Conference on Complex, Intelligent and Software Intensive Systems (IEEE CISIS-2014), pp. 120-127 (2014)
A New Method for Optimization of Number of Mesh Routers in WMNs
47
9. Oda, T., et al.: A GA-based simulation system for WMNs: performance analysis for different WMN architectures considering TCP. In: Proceedings of the IEEE 9-th International Conference on Broadband and Wireless Computing, Communication and Applications (IEEE BWCCA-2014), pp. 120-126 (2014) 10. Oda, T., et al.: Effects of population size for location-aware node placement in WMNs: evaluation by a genetic algorithm-based approach. Pers. Ubiquit. Comput. 18(2), 261–269 (2014) 11. Ikeda, M., et al.: Analysis of WMN-GA simulation results: WMN performance considering stationary and mobile scenarios. In: Proceedings of the 28-th IEEE International Conference on Advanced Information Networking and Applications (IEEE AINA-2014), pp. 337-342 (2014) 12. Oda, T., et al.: Analysis of mesh router placement in wireless mesh networks using friedman test. In: Proceedings of the IEEE 28-th International Conference on Advanced Information Networking and Applications (IEEE AINA-2014), pp. 289-296 (2014) 13. Oda, T., et al.: Effect of different grid shapes in wireless mesh network-genetic algorithm system. Int. J. Web Grid Serv. 10(4), 371–395 (2014) 14. Oda, T., et al.: Analysis of mesh router placement in wireless mesh networks using friedman test considering different meta-heuristics. Int. J. Commun. Networks Distributed Syst. 15(1), 84–106 (2015) 15. Oda, T., et al.: A genetic algorithm-based system for wireless mesh networks: analysis of system data considering different routing protocols and architectures. Soft. Comput. 20(7), 2627–2640 (2016) 16. Sakamoto, S., et al.: Performance evaluation of intelligent hybrid systems for node placement in wireless mesh networks: a comparison study of WMN-PSOHC and WMN-PSOSA. In: Proceedings of the 11-th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS-2017), pp. 16-26 (2017) 17. Holland, J.H.: Genetic algorithms. Sci. Am. 267(1), 66–73 (1992) 18. Skalak, D.B.: Prototype and feature selection by sampling and random mutation hill climbing algorithms. In: Proceedings of the 11-th International Conference on Machine Learning (ICML-1994), pp. 293-301 (1994) 19. Kirkpatrick, S., et al.: Optimization by simulated annealing. Science 220(4598), 671–680 (1983) 20. Glover, F.: Tabu search: a tutorial. Interfaces 20(4), 74–94 (1990) 21. Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proceedings of the IEEE International Conference on Neural Networks (ICNN-1995), pp. 1942-1948 (1995) 22. Hirata, A., et al.: Approach of a solution construction method for mesh router placement optimization problem. In: Proceedings of the IEEE 9-th Global Conference on Consumer Electronics (IEEE GCCE-2020), pp. 467-468 (2020) 23. Hirata, A., et al.: A coverage construction method based hill climbing approach for mesh router placement optimization. In: Proceedings of the 15-th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2020), pp. 355–364 (2020) 24. Hirata, A., et al.: Simulation results of CCM based HC for mesh router placement optimization considering two Islands model of mesh clients distributions. In: Proceedings of the 9-th International Conference on Emerging Internet, Data & Web Technologies (EIDWT-2021), pp. 180–188 (2021)
48
A. Hirata et al.
25. Hirata, A., et al.: A coverage construction and hill climbing approach for mesh router placement optimization: simulation results for different number of mesh routers and instances considering normal distribution of mesh clients. In: Proceedings of the 15-th International Conference on Complex, Intelligent and Software Intensive Systems (CISIS-2021), pp. 161-171 (2021) 26. Hirata, A., et al.: A CCM-based HC system for mesh router placement optimization: a comparison study for different instances considering normal and uniform distributions of mesh clients. In: Proceedings of the 24-th International Conference on Network-Based Information Systems (NBiS-2021), pp. 329-340 (2021) 27. Hirata, A., et al.: A simulation system for mesh router placement in WMNs considering coverage construction method and simulated annealing. In: Proceedings of the 16-th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA-2021), pp. 78–87 (2021) 28. Hirata, A., et al.: Delaunay edge and CCM-based SA approach for mesh router placement optimization in WMN: a case study for evacuation area in Okayama City. In: Proceedings of the 10-th International Conference on Emerging Internet, Data & Web Technologies (EIDWT-2022), pp. 346–356 (2022) 29. Tarjan, R.: Depth-first search and linear graph algorithms. SIAM J. Comput. 1(2), 146–160 (1972) 30. Integrated GIS for all of Okayama Prefecture. http://www.gis.pref.okayama.jp/ pref-okayama/OpenData, ref. Nov. 16 2021
A Wireless Sensor Network Testbed for Monitoring a Water Reservoir Tank: Experimental Results of Delay Yuki Nagai1 , Tetsuya Oda2(B) , Chihiro Yukawa1 , Kyohei Toyoshima1 , Tomoya Yasunaga1 , Aoto Hirata1 , and Leonard Barolli3 1
Graduate School of Engineering, Okayama University of Science (OUS), 1-1 Ridaicho, Kita-ku, Okayama 700–0005, Japan {t22jm23rv,t22jm19st,t22jm24jd,t22jm43sx,t21jm02zr}@ous.jp 2 Department of Information and Computer Engineering, Okayama University of Science (OUS), 1-1 Ridaicho, Kita-ku, Okayama 700–0005, Japan [email protected] 3 Department of Information and Communication Engineering, Fukuoka Insitute of Technology, 3-30-1 Wajiro-Higashi-ku, Fukuoka 811-0295, Japan [email protected] Abstract. The water reservoir tank have various roles as septic tanks, agricultural water storage and also as fire protection tanks. The condition of the water reservoir tank changes with weather conditions. There is a risk of overtopping of the embankment and collapse of the surface of a wall during heavy rainfall. Therefore, by monitoring the water reservoir tank and predicting changes. The damages can be reduced by learning of hazards early stage. Wireless sensor fusion networks have the advantage of being able to collect and analyze a variety of information from a wide range of sources. They can be effective in monitoring water reservoir tank by predicting and preventing damages in water reservoir tank caused by various factors. In this paper, we developed sensing devices and propose a wireless sensor fusion network to monitor the water reservoir tank. For the experiment, we implemented the wireless sensor network testbed and analyze the delay of a wireless sensor network in an outdoor environment considering line of sight scenario.
1
Introduction
The water reservoir tank has a variety of uses, including septic tanks, industrial and agricultural water and fire protection tanks. The water reservoir tank is generally installed outdoors and the reservoir condition changes depending on weather conditions and the amount of water flowing. During heavy rains there is a risk of flooding due to overflow. In addition, although the water reservoir tank varies in construction, especially when soil is used for the walls or when they are topographically low and surrounded by slopes, there is a risk of sediment inflow due to the collapse of the walls or slopes. Such an event could have a significant impact on the surrounding area, depending on the use of the water reservoir c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 49–58, 2022. https://doi.org/10.1007/978-3-031-08812-4_6
50
Y. Nagai et al.
tank. Therefore, by monitoring the water reservoir tank and predicting changes. The damages can be reduced by learning of hazards early stage. However, monitoring the occurrence of disasters and changes must be done in real-time and in addition, it is difficult to grasp, monitor and predict all situations. Similar to the monitoring of water reservoirs tank is the monitoring of rivers. Since rivers extend over long distances and wide areas and condition changes depending on topography and weather conditions, it is difficult for humans to monitor the situation, predict disasters and provide information to the areas around rivers. Therefore, in [1], we developed some sensing devices and propose a wireless sensor fusion network [2–8] to monitor the river. The wireless sensor fusion network collects data as a batch to obtain various information from a wide area as real-time streaming, analyzing and predicting. In this paper, we developed sensing devices and propose a wireless sensor fusion network to monitor the water reservoir tank and implement the wireless sensor network testbed and analyze the delay in an outdoor environment considering line of sight scenario [9]. The paper is organized as follows. In Sect. 2, we present an overview of a water reservoir tank monitoring and predicting system. In Sect. 3, we show the description of the testbed. In Sect. 4, we discuss the experimental results. Finally, in Sect. 5, we conclude the paper.
2
Proposed System for Water Reservoir Tank Monitoring
We show the proposed system in Fig. 1. The wireless sensor fusion network collect sensing data obtained by multiple sensing devices from a wide area by means of the wireless sensor network. The wireless sensor network is based on a wireless mesh network where each sensor node is connected to multiple sensor nodes by wireless links, allowing information to be transmitted over multiple paths. Therefore, even if a failure occurs at one sensor node, the collection of sensing data can continue. The collected data can be analyzed and predicted by sensor fusion which processes multiple data at once, making it possible to extract information that cannot be obtained from a single sensing device. In the proposed system, water level, rainfall, soil moisture, temperature, humidity, atmospheric pressure, water temperature are measured by the sensing devices. The wireless sensor network is installed around the water reservoir tank and collects sensing data, which are used to predict water reservoir tank conditions and weather conditions based on the time-series analysis.
A Wireless Sensor Network Testbed for Monitoring a Water Reservoir Tank
51
Fig. 1. Proposed system.
2.1
Sensors for Water Reservoir Tank Monitoring
Figure 2 shows sensing devices used for the measurement and each sensing device is describe below. • Water Level Indicator In Fig. 2(a) is showing the water level indicater. From the water pressure w [M pa] measured by the water pressure sensor and the atmospheric pressure P [hP a] measured by the atmospheric pressure indicator on the ground, the water pressure W [M pa] is given by: W = w − P.
(1)
The water level h [m] at the water pressure W is given by: h = W × 0.0098.
(2)
• Indicator of Temperature, Humidity and Atmospheric Pressure In Fig. 2(b) is showing the indicator of temperature, humidity and atmospheric pressure. – The measurable range of temperature is –40.000 to 85.000 [◦ C]. – The measurable range of humidity is 0 to 100 [%]. – The measurable range of atmospheric pressure is 300.000 to 1100.000 [hpa]. • Soil Moisture Indicator In Fig. 2(c) is showing the soil moisture indicator. The relative permittivity of water (0.000 ≤ ◦ C ≤ 90.000) to obtain the soil moisture content (58.150 ≤ µ ≤ 88.150).
(3)
52
Y. Nagai et al.
Fig. 2. Sensing devices for the water reservoir tank monitoring.
• Tipping Bucket Type Raingauge In Fig. 2(c) is showing the tipping bucket type raingauge. From the number of times the tipping bucket tilts per hour N and the measurements per tilt 0.500 [mm], the measured rainfall Q [mm] is given by: Q = 0.500 × N.
(4)
We developed the waterproof mechanism and some parts of the sensing devices with a 3D printer in order to have a low cost [10–13]. 2.2
Sensor Fusion of Measurement Data
The collected sensing data are stored in the sink node. Since the water level in the water reservoir tank and the soil moisture surrounding the water reservoir tank change with time, it is necessary to analyze the time series in order to predict the changes in the water reservoir tank condition [14]. Therefore, we perform sensor fusion [15] with the accumulated sensing data and use Long Short-Term Memory (LSTM) [16–19] to predict the water reservoir tank fluctuation.
A Wireless Sensor Network Testbed for Monitoring a Water Reservoir Tank
53
Fig. 3. Location of sink node and sensor nodes and considered environment.
In order to improve the prediction accuracy, the LSTM is trained in advance by the observed data, taking into account the dataset of temperature, humidity, atmospheric pressure and rainfall of Automated Meteorological Data Acquisition System (AMeDAS) installed near the observation points provided by the Japan Meteorological Agency. Then, the proposed system learns the measured values collected by the sensing devices.
3
Description of Wireless Sensor Network Testbed
In this paper, XBee with ZigBee [20–25] is used as a device to build a wireless sensor network. The communication range of XBee is 30 [m] to 60 [m] in indoor and outdoor environments and the maximum communication range is approximately 120 [m] in a line-of-sight scenario. The maximum transmission speed of XBee is 250 [Kbps]. Jetson Nano for the sink node are used and Raspberry pi for the sensor node to control XBee. 3.1
Scenario Description
The experiment was conducted around a water reservoir tank at the Okayama University of Science considering line of sight and the locations of sink nodes and sensor nodes show in Fig. 3. The red dots in Fig. 3 show sink nodes and the black dots show sensor nodes, The distance between nodes is 10 [m]. The buildings around the experimental environment are made with reinforced concrete. Figure 4 shows the Snapshot of each node. The experimental parameters for the wireless sensor networ are show in Table 1.
54
Y. Nagai et al.
(a) Sink Node.
(b) Sensor Node 1.
(c) Sensor Node 2.
(d) Sensor Node 3.
Fig. 4. Snapshot of sink node and sensor nodes.
4
Experimental Results
For evaluation, we used a single flow from sensor node to sink node. In Fig. 5 is showing the average delay. Measurements are taken when the distance between the sink node and the sensor node is 10, 20 and 30 [m] and the source transmission Table 1. Experimental parameters. Functions
Values
Number of Trials
10
Duration
60 [sec]
Number of Sensor Nodes
3
Distance Between Sensor Nodes 10 [m] MAC
IEEE 802.15.4
Routing Protocol
AODV
Transport Protocol
UDP
Flow Type
CBR
Source
32, 64, 128, 256 and 512 [Kbps]
A Wireless Sensor Network Testbed for Monitoring a Water Reservoir Tank
Fig. 5. Experimental results for Avg. Delay.
55
56
Y. Nagai et al.
rate is 32, 64, 128, 256 and 512 [Kbps]. The delay time is the average value of 10 measurements for 60 [sec]. We can see that the delay increases with the increase of the source transmission rate and the increasing of communication distance. From the experimental results, at 32 and 64 [Kbps], the delay time is less than 1 [sec] and the average values of delay are 0.439 and 0.781 [sec] at a communication distance of 30 [m], respectively. Sensors on four sensing devices described in Subsect. 2.1 send up to 52 [bit] of data to the sink node via the network. Therefore, we think that the proposed system is possible to collect the sensing data and monitor the water reservoir tank in real time.
5
Conclusions
In this paper, we developed some sensing devices for water reservoir tank monitoring and implemented a wireless sensor networ testbed to investigate the performance of wireless sensor network in an outdoor environment considering line of sight scenario. From experimental results, we found that the delay is less than 1 [sec] for 64 [Kbps]. In the future, we would like to carry out experiments in various environments, including different weather conditions. Acknowledgement. This work was supported by JSPS KAKENHI Grant Number JP20K19793.
References 1. Nagai, Y., et al.: A river monitoring and predicting system considering a wireless sensor fusion network and LSTM. In: Barolli, L., Kulla, E., Ikeda, M. (eds.) Advances in Internet, Data & Web Technologies. EIDWT 2022. Lecture Notes on Data Engineering and Communications Technologies, vol. 118, pp. 283–290. Springer, Cham (2022). https://doi.org/10.1007/978-3-030-95903-6 30 2. Lewi, T., et al.: Aerial sensing system for wildfire detection. In: Proceedings of the 18-th ACM Conference on Embedded Networked Sensor Systems, pp. 595–596 (2020) 3. Mulukutla, G., et al.: Deployment of a large-scale soil monitoring geosensor network. SIGSPATIAL Spec. 7(2), 3–13 (2015) 4. Gayathri, M., et al.: A low cost wireless sensor network for water quality monitoring in natural water bodies. In: Proceedings of the IEEE Global Humanitarian Technology Conference, pp. 1–8 (2017) 5. Gellhaar, M., et al.: Design and evaluation of underground wireless sensor networks for reforestation monitoring. In: Proceedings of the 41-st International Conference on Embedded Wireless Systems and Networks, pp. 229–230 (2016) 6. Yu, A., et al.: Research of the factory sewage wireless monitoring system based on data fusion. In: Proceedings of the 3rd International Conference on Computer Science and Application Engineering, no. 65, pp. 1–6 (2019)
A Wireless Sensor Network Testbed for Monitoring a Water Reservoir Tank
57
7. Suzuki, M., et al.: A high-density earthquake monitoring system using wireless sensor networks. In: Proceedings of the 5-th International Conference on Embedded Networked Sensor Systems, pp. 373–374 (2007) 8. Oda, T., et al.: Design and implementation of a simulation system based on deep q-network for mobile actor node control in wireless sensor and actor networks. In: Proceedings of the IEEE 31st International Conference on Advanced Information Networking and Applications Workshops, pp. 195–200 (2017) 9. Oda, T., et al.: Implementation and experimental results of a WMN testbed in indoor environment considering LoS scenario. In: Proceedings of the IEEE 29-th International Conference on Advanced Information Networking and Applications, pp. 37–42 (2015) 10. Saito, N., et al.: Design and implementation of a DQN based AAV. In: Proceedings of the 15-th International Conference on Broad-Band and Wireless Computing, Communication and Applications, pp. 321–329 (2020) 11. Saito, N., et al.: Simulation results of a DQN based AAV testbed in corner environment: a comparison study for normal DQN and TLS-DQN. In: Proceedings of the 15-th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 156–167 (2021) 12. Saito, N., et al.: A tabu list strategy based DQN for AAV mobility in indoor singlepath environment: implementation and performance evaluation. Internet Things 14, 100394 (2021) 13. Saito, N., et al.: A LiDAR based mobile area decision method for TLS-DQN: improving control for AAV mobility. In: Proceedings of the 16-th International Conference on P2P, Parallel, Grid, Cloud and Internet Computing, pp. 30–42 (2021) 14. Sharma, P., et al.: A machine learning approach to flood severity classification and alerting. In: Proceedings of the 4-th ACM SIGSPATIAL International Workshop on Advances in Resilient and Intelligent, pp. 42–47 (2021) 15. Hang, C., et al.: Recursive truth estimation of time-varying sensing data from online open sources. In: Proceedings of the 14-th International Conference on Distributed Computing in Sensor Systems, pp. 25–34 (2018) 16. Hochreiter, S., et al.: Long short-term memory. Neural Comput. 9, 1735–1780 (1997) 17. Karevan, Z., et al.: Transductive LSTM for time-series prediction: an application to weather forecasting. Neural Netw. 125, 1–9 (2019) 18. Li, Y., et al.: Hydrological time series prediction model based on attention-LSTM neural network. In: Proceedings of the 2nd International Conference on Machine Learning and Machine, pp. 21–25 (2019) 19. Toyoshima, K., et al.: Proposal of a haptics and LSTM based soldering motion analysis system. In: Proceedings of the IEEE 10-th Global Conference on Consumer Electronics, pp. 1–2 (2021) 20. Oda, T., et al.: A genetic algorithm-based system for wireless mesh networks: analysis of system data considering different routing protocols and architectures. Soft Comput. 20(7), 2627–2640 (2016) 21. Oda, T., et al.: Evaluation of WMN-GA for different mutation operators. Int. J. Space-Based Situat. Comput. 2(3), 149–157 (2012) 22. Oda, T., et al.: WMN-GA: a simulation system for WMNs and its evaluation considering selection operators. J. Ambient Intell. Humaniz. Comput. 4(3), 323– 330 (2013) 23. Nishikawa, Y., et al.: Design of stable wireless sensor network for slope monitoring. In: Proceedings of the IEEE Topical Conference on Wireless Sensors and Sensor Networks, pp. 8–11 (2018)
58
Y. Nagai et al.
24. Hirata, A., et al.: A coverage construction method based hill climbing approach for mesh router placement optimization. In: Proceedings of the 15-th International Conference on Broad-Band and Wireless Computing, Communication and Applications, pp. 355–364 (2020) 25. Hirata, A., et al.: A coverage construction and hill climbing approach for mesh router placement optimization: simulation results for different number of mesh routers and instances considering normal distribution of mesh clients. In: Proceedings of the 15-th International Conference on Complex, Intelligent and Software Intensive Systems, pp. 161–171 (2021)
Remote Medical Assistance Vehicle in Covid-19 Quarantine Areas: A Case Study in Vietnam Linh Thuy Thi Pham1,5 , Tan Phuc Nhan Bui2,3 , Ngoc Cam Thi Tran2 , Hai Thanh Nguyen1(B) , Khoi Tuan Huynh Nguyen4 , and Huong Hoang Luong4 1
3
Can Tho University, Can Tho, Vietnam [email protected] 2 An Khanh High School, Can Tho, Vietnam [email protected] Saigon International University, Ho Chi Minh City, Vietnam 4 FPT University, Can Tho, Viet Nam [email protected] 5 FPT High School, FPT University, Can Tho, Vietnam
Abstract. The pain, namely “Covid-19 epidemic”, has caused many sacrifices, loss, and loneliness. Only those who have experienced traumatic losses can fully understand the pain that is hard to erase by the epidemic. This study focuses on designing a remote medical assistance vehicle used in quarantine areas in Vietnam to support epidemic prevention with simple, cheap, easy-to-use, and multi-function criteria. The proposed system includes a 3-layer vehicle for transporting supplies controlled remotely via Radio Frequency (RF) signals to help limit crossinfection for medical staff and volunteers. The main component is the RF transceiver circuit, which transmits and receives data wirelessly over 2.4 GHZ RF using IC Nrf24l01, Nordic standard SPI interface for remote control. DC motor driver circuit BTS7960 43A controls the motor to prevent overvoltage and current drop. Moreover, the vehicle integrates an electric sprayer to support disinfecting spray a Xiaomi camera to stream video and communicate directly with patients and healthy in isolation. Ultrasonic sensors and infrared sensors aim to scan obstacles through reflected waves. The reflected signals received from the barrier objects are used as input to the microcontroller. The microcontroller is then used to determine the distance of objects around the vehicle. If an obstacle is detected, the disinfectant sprayer can stop for several seconds to ensure the safety of medical staff if there is a pass. The system has a built-in light sensor that works at night. The system is deployed at a low cost and is evaluated through some experiments. It is expected to be easy to use and is an innovative solution for hospitals. Once the outbreak is over, the product can still be used in infectious disease areas. Keywords: Remote vehicle
· Medical assistance · Disinfectant sprayer
c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 59–70, 2022. https://doi.org/10.1007/978-3-031-08812-4_7
60
1
L. T. T. Pham et al.
Introduction
On 17 November 2019, the Chinese government announced that the first case of COVID-19 infection occurred in Hubei province, Wuhan city1 . Currently, more than 200 countries and territories are infected with the disease. According to the World Health Organization’s portal2 on the Covid-19 pandemic, over 450 million people worldwide have been infected, more than six million have died. In Vietnam alone, more than 2.4 million people have been infected with the disease, 38688 deaths have left a massive loss to the country, many problems occurred and faced such as the increasing number of quarantined people, leading to a shortage of medical staff. Bringing food, drinking water, household items, or going to the place to monitor the health of quarantined people poses a significant risk of crossinfection for medical staff. In addition, spraying disinfectants at medical facilities, schools, and companies takes time and effort. Currently, there are many robot products to support medical staff in the prevention of the Covid-19 epidemic. However, among those methods focusing on applying artificial intelligence AI, single-function but the very high cost is not suitable for small but overloaded medical facilities in epidemic prevention work. Wireless data transmission is a large array of information electronics, which the most effective is transmission by electromagnetic waves, also known as Radio, because of its advantages such as long-distance, omnidirectional, high operating frequency. On the other hand, Digital data transmission is widely applied, especially in digital control and information. This research focuses on designing and developing a remote medical assistance vehicle using RF signals and hardware components. In addition, the system integrates other features such as carrying isolation items, communicating directly with isolated people, spraying disinfectant, turning on the horn when encountering obstacles, and pedestrians when the vehicle is running on the road to track, display battery capacity and warn when battery capacity is about to run out. We have designed and proposed a medical assistance vehicle at the Can Tho Tuberculosis and Lung Disease Hospital, Vietnam, around September 2020, with positive results in supporting epidemic prevention, such as transporting isolation supplies and providing support. In addition, spraying disinfectant in the hospital area directly communicating with the quarantined person can limit cross-infection of medical staff, with a remote control feature within 100 m for 90% accuracy in tests.
2
Related Work
The Covid-19 pandemic is a fundamental challenge to the health and safety of the public, healthcare workers, and health systems worldwide. Therefore, many studies have been proposed globally, including using robots to improve 1 2
https://www.wincalendar.com/Calendar/Date/November-17-2019, accessed on 12 March 2022. https://covid19.who.int, accessed on 13 March 2022.
Remote Medical Assistance Vehicle in Covid-19 Quarantine Areas
61
patient treatment and boost health system load. Based on [1], the authors introduced medical examination robots, service robots, disinfection robots, and object transport robots with a variety of technologies used such as AI, 5G, wireless network sensors, HRC, tactile control, big data, and the cloud computing services. According to [2], the authors proposed a real-time robot-based backend system to assess the risk of COVID-19 infection. The robot combines real-time voice recognition, temperature measurement, keyword detection, cough detection, and other functions to convert live audio into structured data to assess the risk of COVID-19 infection. -19. It relies on accurate conversation data from the human-robot interactions, processes the voice signal to detect a cough, and classifies it if detected based on the cough’s length. At the same time, the robot will extract the vital information structure in the dialogue between the human and the robot and store it in the electronic medical record along with the cough recording and a dedicated epidemic map to remind. The user avoids high-risk areas based on the robot’s position. The product is placed in communities, hospitals, and supermarkets to support COVID-19 testing with a Top-1 accuracy of 76%. The authors [3] recommend the Robocov Robot, which is a low-cost robot that can perform several tasks to reduce the impact of COVID-19. It is a robot that integrates the features of S4 and AI, allowing the robot to complete enduser requests such as cleaning temperature monitoring to detect infected people’s detect masks. Remote navigation system with support and control of the obstacle avoidance system. The study in [4] proposed a robot and two main systems. The remote control system includes a wearable, original motion recorder, and a two-arm collaboration robot. The medical staff’s upper limb movement data can be obtained and used to remotely control the movement of the robotic arm using a motion capture device. The work on a five-door manual cleaning and maintenance robot by the authors [5] deployed AI to automate cleaning tasks. The overall cleaning process includes movable base movement, door handle detection, and HSR controller control to complete the cleaning task. The detection part exploits the deep learning technique to classify the image space and provide a set of coordinates to the robot. The cooperative control between spray and mop is developed in the Robot Operating System. The control module uses the information obtained from the detection module to create a task/functional space for the robot and assess the desired position for the operator’s action. Accuracy: The technique’s authenticity is tested offline and online through simulations and experiments. The detection accuracy is calculated to be more than 90% in both cases. The authors in [6] designed Smart Cleaner with an economical-cost autonomous indoor disinfection robot that integrated a wheeled mobile robot platform along with a hydrogen peroxide atomization device to perform in the indoor environment to combat the COVID-19 pandemic. The study in [7] gave a review on primary applications of cutting-edge industrial technologies against the pandemic, including unconventional applications of uncrewed vehicles. Another research in [8] highlighted the benefits of using heuristics in an emergency context for a vehicle routing problem like the one triggered by the COVID-19 pandemic.
62
L. T. T. Pham et al.
In [9], the researchers analyzed the critical Role of the Hercules Autonomous Vehicle during the COVID-19 pandemic avoiding person-to-person transmission aiming to prevent the pandemic. The above studies show that the use of RF signals and multifunctional design is suitable for the medical situation in Vietnam. However, there are very few studies on using RF signals to design Covid-19 prevention vehicles. Therefore, determine the use of RF signals in remote control in designing anti-epidemic support vehicles.
3 3.1
Materials and Methods Some Requirements for the System
The vehicle that does not obstruct are of moderate size does not obstruct the circulation of doctors and medical staff. Moderate weight should not exceed 60 kg. The vehicle consists of 4 wheels and multiple floors, with a high load to carry items easily. The floors are designed to be easy to clean and disinfect without affecting the internal circuit system. The Vehicles have installed a battery that can charge in 5 h and allow continuous use for 12 h. The vehicle can move both at night and in dark areas. Vehicles can spray disinfectants on utensils, wards, ambulances, and isolated areas. The vehicle can travel in many different terrains, including steep and uneven terrain. Remote control and detecting obstacles while spraying disinfectant must reach 90% or more. The Vehicles can communicate with quarantined people or other people through a video stream from the camera to avoid cross-infection as much as possible. In preparing devices, we chose electronic components with the price as low as possible to match the economic conditions of most small and medium hospitals in Vietnam. The device can be used after 10 min of instruction so that all medical staff can use it easily through the manual. 3.2
Components for the Designed Vehicle
Radio Frequency (RF) Signals. RF radio waves are repeated vibrations that can travel long distances in space and are characterized by frequency, wavelength, and amplitude. We use electromagnetic waves through the transmitter and receiver. The receiver will receive data from the transmitter. The decoder then decodes the data and activates the corresponding output pin. RF remote control is the first type of remote control that appeared in the world over many years but still plays a significant and widespread role in today’s life. RF control is applied to products such as electronic toys, garage door openers, smartphones, and laptop systems [10]. RF signals can travel even when there is an obstacle between the transmitter and receiver system, so it is preferred. RF transceiver circuit NRF24L01+ PA + LNA 2.4 Ghz EBYTE E01-ML01DP5 has an industry-standard design with the anti-interference iron case, size: 18 × 33.4 mm. The circuit is integrated with PA + LNA (Power + Low Noise Amplifiers)
Remote Medical Assistance Vehicle in Covid-19 Quarantine Areas
63
for long and stable transmission and reception distance. According to the manufacturer’s specifications, the transmission distance can be up to 2500 m. The component uses voltage 2 3.6 VDC, with maximum current consumption: 130 mA(Transmitter)/20 mA (Receiver), Operating frequency: 2.4 2.525 Ghz, High data transfer rate: 250k 2 Mbps, Communication standard: SPI.3 . Design for the Vehicle. The chassis is designed as shown in Fig. 1 including three floors lined with ceramic tiles with dimensions of 50 cm × 75 cm × 100 cm respectively, weight 50 kg, load 300 kg. The disinfectant spray module includes one electric sprayer with a capacity of 20 liters, one switch, one spray nozzle with two nozzles, one motor, and a travel switch to move the spray nozzle to the left and right with an angle of 150 ◦ C. The automatic hand washing machine module includes one electric box size of 20 cm × 20 cm × 10 cm. In addition, the inside includes one pump, nebulizer machine, Arduino Nano circuit, relay, switching circuit, and infrared sensor. The direct patient communication module includes 1 Xiaomi Mijia PTZ 360 ◦ C IP Camera, 1080P, and Mi Home software included with the camera. Finally, the UV sterilization module includes six 12 V UV lights designed above the vehicle with three balls on each side to help kill bacteria after spraying or for places with narrow spaces that cannot be sprayed with disinfectant. The solution design is shown in Fig. 2 in which the RF signal block will receive control signals from the Vehicle control handle to send data into the data processing block. The signal processing block will receive the signals from the RF signal block that the driver gives and process the signals to select the control method and then output the control signals to the ports through the DC motor driver. Electric cylinder, lights, signal horn. The dynamic block will receive the signals given from the signal processing block, execute the requests given by the control block, and move to the working position. The signal block will receive
Fig. 1. Design of vehicles to support in Covid-19 quarantine areas 3
https://arduinoinfo.mywikis.net/wiki/Nrf24L01-2.4GHz-HowTo, accessed on 16 March 2022.
64
L. T. T. Pham et al.
the signals given from the signal processing block and execute the command to control the moving signals whistle lights. The power block is responsible for supplying power to the whole system. The central circuit diagram of the medical assistance vehicle is designed as shown in Fig. 3 including the electric cylinder that will receive the signal from the 8-channel Relay Module and perform left and proper rotation of the steering bridge. 8-channel Opto Relay module selects high/Low trigger levels to turn on/off sprayers, UV lights, horns, etc., and antiinterference. Arduino Mega 2560 acts as a central processing unit, receiving signals from RF Modules, processing signals, and outputting signals (Digital, PWM) to implementing devices (Module Relay, DC 250W). The power supply for the vehicle to operate includes two batteries of 12 V–30 Ah. The 250 W motor driver circuit reverses the motor when receiving a HIGH or LOW signal from the Arduino Mega 2560, and receiving a pulse hash signal (PWM port) from the Arduino Mega 2560 adjusts the corresponding pulse to the motor at high power. NRF24L01+ PA + LNA 2.4 Ghz RF transceiver circuit NRF-E04 transmits and receives wireless data over 2.4 GHZ RF waves using genuine Nordic Nrf24l01+ IC with SPI communication standard. DC motor driver circuit BTS7960 43A controls the motor and prevents overvoltage and drop. The wiring diagram of the control handle of the vehicle is designed as shown in Fig. 4 including The joystick circuit is a knob that controls the Vehicle via RF signal; RF transceiver circuit NRF24L01+ PA + LNA 2.4 Ghz NRF-E04; used to transmit and receive data wirelessly via RF 2.4 GHZ; Arduino Nano board for central processing, receiving signals from joystick to control the vehicle forward, backward, left, right, on and off the aerosol switch; Power switch: toggle the power device of the handle; Aerosol switch to turn on and off the antibacterial spray system; The OLED screen is used to display the vehicle’s operating status.
Fig. 2. Block diagram
The wiring diagram of the handwashing machine placed on the vehicle to help the isolated person wash their hands automatically with alcohol solution directly on the vehicle is designed, including Arduino Nano Board with the task of receiving and processing information from the infrared sensor module. E3F-DS30C4 Adjustable IR Infrared Proximity Sensor: uses infrared light to identify obstacles in front of the sensor. The sensor emits infrared rays with a specialized frequency range for good anti-interference ability even in outdoor
Remote Medical Assistance Vehicle in Covid-19 Quarantine Areas
65
Fig. 3. Vehicle hub wiring diagram
Fig. 4. Wiring diagram of the controller
lighting conditions; Relay: switch on and off the pump device; Pump: Pumps disinfectant solution when someone puts their hand across the infrared sensor. The central algorithm diagram of Xe is designed as shown in Fig. 5. How Robot Center works: Use an RF signal to receive a real-time array of 5 elements from the handle and test the sensor. The five elements are: Command[0] reveals the relay of the aerosol while Command[1] denotes the relay of the two rear wheels. Command[2] specifies the relay of the two front wheels. Command[3] determines the relay of the buzzer. Finally, command[4] specifies the relay of the UV. PIR sensor (Passive InfraRed Sensor) returns a signal with an obstacle. It will directly affect the system, close the UV relay aerosol relay, turn on the horn relay, and disable the system for 7 s so pedestrians can move. The algorithm diagram of the control handle is designed as shown in Fig. 6. The controller uses RF signals to send a real-time array of 5 elements to the vehicle: Command[0] corresponds to the left switch state (aerosol); Command[1] corresponds to the left joystick state (forward and backward); Command[2]
66
L. T. T. Pham et al.
Fig. 5. The workflow for the designed vehicle
corresponds to the state of the right joystick (spin left and right); Command[3] controls the state of the proper button (horn); Command[4] corresponds to the state of the left button (UV light).
4 4.1
Experiments and Evaluation Operating Principles
To get started, we attach two batteries to the bottom floor of the vehicle and turn on the control switch the vehicle. Remote control and carrying items: We have a handlebar control box to control the vehicle. It includes two switches and two joysticks. First, we turn on the control power switch. Then, when the vehicle moves, the Oled screen displays the vehicle’s operating status as forwarding, Backwarding, turning Left, turning Right. To speed up, we can press the right joysticks forward and stop pressing if we do not want to continue accelerating. The vehicle is designed with three floors, with a maximum load of 300 kg for medical equipment or personal things for people in quarantine areas. Disinfectant spray: To perform this function, we need to pour the disinfectant solution into the spray bottle not to exceed 20 L. To start spraying, turn on the left switch, and the Oled screen will display the status of Disinfectant Spray. The spray nozzle will automatically spray left and right with a rotation angle of 150 ◦ C. If an obstacle is detected in front of the vehicle (using the ultrasonic sensor located in the automatic hand wash box), the automatic nozzle can stop (in seven seconds) and automatically start spraying again if no object is detected (Fig. 7).
Remote Medical Assistance Vehicle in Covid-19 Quarantine Areas
67
Fig. 6. The workflow for the designed remote medical assistance vehicle
Fig. 7. Components of Spraying and nozzle and UV light system (left) and automatic hand washing machine (right)
Turning on the spray UV light: In places where there is a narrow space, the vehicle can not enter to spray, or the areas need maximum cleaning to kill both viruses and bacteria, UV rays are the optimal choice. To use this function, we press the left joysticks. The Oled screen will display the status of On UV. Note because UV rays will affect the user’s eyes, so only turn on when that area is empty. The operator, if standing nearby, needs to wear eye protection. Automatic hand washing machine: When medical staff or volunteers need to wash their hands after contacting sick people or after transporting items without
68
L. T. T. Pham et al.
alcohol, they can wash them directly on the vehicle through the automatic hand washing machine. This function includes a bottle of hand sanitizer or alcohol placed outside the control box to increase the amount of solution used many times. And one box with a mini pump, ultrasonic sensor, and washing faucet. The solution can be pumped out automatically with the right amount. Communicating with quarantined people and patients via Camera: In some cases, the vehicle entering the isolation area gets stuck, and the medical staff cannot enter that area. The camera can be a valuable measure to know whether the vehicle is lost and necessary support. Alternatively, medical staff can communicate with them via camera to avoid cross-contamination. You need to download and install the MiHome application on your phone or computer to use this function. When needed, we turn on the Camera power button and turn it off to save power. In addition, the system can automatically turn on the horn when it encounters an obstacle through the ultrasonic sensor, it is dark, or a dimly lit area will automatically turn on the light based on the light sensor. 4.2
Evaluation
Carrying utensils: With a design of 3 floors lined with ceramic tiles for easy cleaning, 300 kg load, carrying many different utensils. However, due to the ceramic tile design, the weight increases. The vehicle can move and carry belongings on various terrains, including uneven terrain on the sand, rocks, wood, and rough surfaces. The vehicle can climb 45 ◦ C easily and 50 ◦ C if not accompanied by items on the vehicle. The detailed results are as follows: If you carry bags of rice (10 kg/bag), you will get 20 bags. If carrying trays of food, each floor has 12 trays. If carrying bulky items(fans, styrofoam boxes), it is about 50 kg. Spraying disinfectant: Using an electric sprayer with a large spray area, the capacity of 20 L, can spray continuously for 6–8 h, pressure 0.15–0.6 MPa, 80 PSI, saving electricity due to automatic shutdown relay when non-spray bottle. It was designed with a plastic housing to help resist chemical corrosion. The sprayer can be adjusted to the left and right quickly. The ability to communicate with people in quarantine areas via camera: We deployed a Camera1080P, equipped with 2-way talk capability, allowing users to make video calls. Many people monitor the same camera or multiple cameras through the phone anywhere while they need a wifi or 3G/4G network. They are equipped with a new generation intelligent motion recognition system, with the intervention of artificial intelligence4 . Besides, Remote control by RF signal can be up to 2000 m with no obstructions by trees or other electromagnetic waves. We make a statistical table of data when remote control (30 measurements/time) like the Table 8. The results showed that the vehicle worked well within the first 100 m. However, the number of signal loss increases gradually, especially in the environment with trees. Evaluation of the power: The vehicle is used with two batteries, 12 V–13 Ah. The current consumption of the sprayer is 12 A, Xiaomi 4
https://www.mi.com/vn/mi-home-security-camera-360-1080P, March 2022.
accessed
on
16
Remote Medical Assistance Vehicle in Covid-19 Quarantine Areas
69
Fig. 8. The experiment on the number of signal loss with the distances
Camera 2 A, RF circuit 0.13 A, and other components range from 40 mA to 2 A. The data table calculates the amount of electricity consumed in continuous use. Vehicles do not carry belongings, only move in 3 h, vehicle moving and spraying in 2.5 h, Utensils (no spraying) in 2.75 h. Assessment of mobility of the vehicle: is exhibited in Table 1. Vehicles supporting epidemic prevention have two mobility: normal movement and acceleration (Fig. 8). Table 1. Test cases of vehicle movements Test case
Velocity
Normal movement (flat terrain)
0.83 m/s
Fast movement (acceleration, flat terrain)
2.7 m/s
Moving with heavy objects (>100 kg)
0.63 m/s
Moving with bulky items
2.5 m/s
Accelerated movement, climbing slopes without carrying equipment 2.6 m/s Moving uphill with luggage (86% from 70000
Emilya [40]
8026
2015
18
Camera
Face and body
Video, sound
GEMEP
2014
8
Xsens MVN motion capture
Motion capture+ audio
Video, sound
-
PACO [41] UCLIC [42]
4080
2006
4
Motion capture
Body
PTDCSM
PACO
183
2006
4
VICON (3D)
Body
Avatar
UCLIC
Emotional body motion database [43]
1451
2014
11
Camerasensor
Body
bvhmvnx
-
OMG emotion [44]
2400
2018
7
Camera
Facebodysound
-
-
EmoPain [35]
35
2020
3
18 sensors
Face and body
26 point in 3Dvideo
EmoPain
No name [34]
118
2016
3
-
Body
Excel
Link
Action database [45]
2783
2014
5
Camera
Body
HD video
Action Database
BEAST [46]
254
2011
4
Cameramat face
-
Image (bmp)
BEAST
No name [47]
560
2018
6
Kinect
Posture, video, vocal
3D sckeleton (XEF)
-
EWalk [17]
1384
2019
4
Camera
Body
Skeleton
-
IEMOCAP [48]
1384
2008
9
Camera
Face, upper-body, sound
Video, sound, face, skeleton
IEMOCAP
MPI [49]
1447
2014
11
Motion capture system
Body
Bvh (3D)
MPI
datasets are not ideal. Many research work use self-generated data and are not available for review. Some multimodal datasets are not currently available [34, 35]. The most referenced datasets in the literature, for the recognition of emotion based on the state of the human body, is shown in Table 1. The various data features and access links are also listed.
4
Conclusions
In this overview and brief survey, recognizing emotions based on human body movement is presented and discussed. The basic concepts of body language, emotions, and their relationship are covered. The framework and components of emotion recognition systems are introduced. The datasets often cited in the literature are reviewed and the most referenced ones are presented with their characteristics.
An Overview of Emotion Recognition from Body Movement
115
References 1. Schindler, K., Van Gool, L., De Gelder, B.: Recognizing emotions expressed by body pose: a biologically inspired neural model. Neural Netw. 21, 1238–1246 (2008) 2. Elman, J.L.: Encyclopedia of Language and Linguistics, 2nd edn. Elsevier, Oxford (2005) 3. De Silva, P., Bianchi-Berthouze, N.: Modeling human affective postures: an information theoretic characterization of posture features. Comput. Anim. Virtual Worlds 15, 269–276 (2004) 4. Thanh Nguyen, D., Li, W., Ogunbona, P.: Human detection from images and videos: a survey. Pattern Recogn. 51, 148–175 (2016) 5. Kachouane, M., Sahki, S., Lakrouf, M., Ouadah, N.: Hog based fast human detection. In: 24th International Conference on Microelectronics (ICM), pp. 1–4 (2012) 6. Dalal, N., Triggs, B., Schmid, C.: Human detection using oriented histograms of flow and appearance. In: European Conference on Computer Vision, pp. 428–441 (2006) 7. Noroozi, F., Adrian Corneanu, C., Kami´ nska, D., Sapi´ nski, T., Escalera, S., Anbarjafari, G.: Survey on emotional body gesture recognition. IEEE Trans. Affect. Comput. 12, 505–523 (2018) 8. Ansari, M., Kumar Singh, D.: Human detection techniques for real time surveillance: a comprehensive survey. Multimedia Tools Appl. 80, 8759–8808 (2021) 9. Chen, Y., Tian, Y., He, M.: Monocular human pose estimation: a survey of deep learning-based methods. Comput. Vis. Image Underst. 192, 102897 (2020) 10. Toshev, A., Szegedy, C.: Deeppose: human pose estimation via deep neural networks. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 1653–1660 (2014) 11. Charles, J., Pfister, T., Magee, D., Hogg, D., Zisserman, A.: Personalizing human video pose estimation. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 3063–3072 (2016) 12. Pishchulin, L., et al.: Deepcut: joint subset partition and labeling for multi person pose estimation. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 4929–4937 (2016) 13. Alldieck, T., Magnor, M., Xu, W., Theobalt, C., Pons-Moll, G.: Video based reconstruction of 3d people models. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 8387–8397 (2018) 14. Crane, E., Gross, M.: Motion capture and emotion: affect detection in whole body movement. In: International Conference on Affective Computing and Intelligent Interaction, pp. 95–101 (2007) 15. Gunes, H., Piccardi, M.: Affect recognition from face and body: early fusion vs. late fusion. In: IEEE International Conference on Systems, Man and Cybernetics, vol. 4, pp. 3437–3443 (2005) 16. Maret, Y., Oberson, D., Gavrilova, M.: Identifying an emotional state from body movements using genetic-based algorithms. In: International Conference on Artificial Intelligence and Soft Computing, pp. 474–485 (2018) 17. Randhavane, T., Bhattacharya, U., Kapsaskis, K., Gray, K., Bera, A., Manocha, D.: Identifying emotions from walking using affective and deep features. arXiv preprint arXiv:1906.11884 (2019) 18. Ferdous, A., Hossain, B.A.S.M., Marina, G.: Emotion recognition from body movement. IEEE Access 8, 11761–11781 (2019)
116
L. Ebdali Takalloo et al.
19. Karg, M., K¨ uhnlenz, K., Buss, M.: Recognition of affect based on gait patterns. IEEE Trans. Syst. Man Cybernet. Part B (Cybernet.) 40, 1050–1061 (2010) 20. Kapur, A., Kapur, A., Virji-Babul, N., Tzanetakis, G., Driessen, P.: Gesture-based affective computing on motion capture data. In: International Conference on Affective Computing and Intelligent Interaction, pp. 1–7 (2005) 21. Sapi´ nski, T., Kami´ nska, D., Pelikant, A., Anbarjafari, G.: Emotion recognition from skeletal movement. Entropy 21, 646 (2019) 22. Kaza, K., et al.: Body motion analysis for emotion recognition in serious games. In: International Conference on Universal Access in Human-Computer Interaction, pp. 33–42 (2016) 23. Coulson, M.: Attributing emotion to static body postures: recognition accuracy, confusions, and viewpoint dependence. Nonverbal Behav. 28, 117–139 (2004) 24. Posner, J., Russell, J., Peterson, B.: The circumplex model of affect: an integrative approach to affective neuroscience, cognitive development, and psychopathology. Dev. Psychopathol. 17, 715–734 (2005) 25. Bernhardt, D.: Emotion inference from human body motion. Technical report, University of Cambridge, Computer Laboratory (2010) 26. Viegas, C.: Two stage emotion recognition using frame-level and video-level features. In: 15th IEEE International Conference on Automatic Face and Gesture Recognition, pp. 912–915 (2020) 27. Alghowinem, S., Goecke, R., Cohn, J., Wagner, M., Parker, G., Breakspear, M.: Cross-cultural detection of depression from nonverbal behaviour. In: 11th IEEE International Conference and Workshops on Automatic Face and Gesture Recognition, vol. 1, pp. 1–8 (2015) 28. Shi, J., Liu, C., Toshinori Ishi, C., Ishiguro, H.: Skeleton-based emotion recognition based on two-stream self-attention enhanced spatial-temporal graph convolutional network. Sensors 21, 205 (2021) 29. Arunnehru, J., Geetha, K.: Automatic human emotion recognition in surveillance video. In: Intelligent Techniques in Signal Processing for Multimedia Security, pp. 321–342 (2017) 30. Sebastian-Kaltwang, M., Romera-Paredes, B., Martinez, B., Singh, A., Cella, M., Valstar, M.: The automatic detection of chronic pain-related expression: requirements, challenges and the multimodal emopain dataset. IEEE Trans. Affect. Comput. 7, 435–451 (2015) 31. Wang, W., Enescu, V., Sahli, H.: Adaptive real-time emotion recognition from body movements. ACM Trans. Interact. Intell. Syst. 5, 1–21 (2015) 32. Ahmed, F., Sieu, B., Gavrilova, M.: Score and rank-level fusion for emotion recognition using genetic algorithm. In: IEEE 17th International Conference on Cognitive Informatics & Cognitive Computing, pp. 46–53 (2018) 33. Wei, G., Jian, L., Mo, S.: Multimodal (audio, facial and gesture) based emotion recognition challenge. In: 15th IEEE International Conference on Automatic Face and Gesture Recognition, pp. 908–911 (2020) 34. Yuan, X., Mahmoud, M.: Alanet: autoencoder-lstm for pain and protective behaviour detection. In: 15th IEEE International Conference on Automatic Face and Gesture Recognition, pp. 824–828 (2020) 35. Egede, J., et al.: Emopain challenge 2020: multimodal pain evaluation from facial and bodily expressions. In: 15th IEEE International Conference on Automatic Face and Gesture Recognition, pp. 849–856 (2020) 36. Gunes, H., Piccard, M.: A bimodal face and body gesture database for automatic analysis of human nonverbal affective behavior. In: 18th IEEE International Conference on Pattern Recognition, vol. 1, pp. 1148–115 (2006)
An Overview of Emotion Recognition from Body Movement
117
37. Douglas-Cowie, E., et al.:‘The humaine database. Emotion-Oriented Systems, pp. 243–284 (2011) 38. Baveye, Y., Dellandrea, E., Chamaret, C., Chen, L.: Liris-accede: a video database for affective content analysis. IEEE Trans. Affect. Comput. 6, 43–55 (2015) 39. B¨ anziger, T., Mortillaro, M., Scherer, K.: Introducing the Geneva multimodal expression corpus for experimental research on emotion perception. Emotion 12, 1161 (2012) 40. Fourati, N., Pelachaud, C.: Emilya: emotional body expression in daily actions database. In: Proceedings of the Ninth International Conference on Language Resources and Evaluation, pp. 3486–3493 (2014) 41. Ma, Y., Paterson, H., Pollick, F.: A motion capture library for the study of identity, gender, and emotion perception from biological motion. Behav. Res. Methods 38, 134–141 (2006) 42. Kleinsmith, A., De Silva, R., Bianchi-Berthouze, N.: Cross-cultural differences in recognizing affect from body posture. Interact. Comput. 18, 1371–1389 (2006) 43. Emotional Body Motion Database. http://ebmdb.tuebingen.mpg.de/ 44. Barros, P., Churamani, N., Lakomkin, E., Siqueira, H., Sutherland, A., Wermter, S.: The omg-emotion behavior dataset. IEEE International Joint Conference on Neural Networks, pp. 1–7 (2018) 45. Keefe, B., Villing, M., Racey, C., Strong, S., Wincenciak, J., Barraclough, N.: A database of whole-body action videos for the study of action, emotion, and untrustworthiness. Behav. Res. Methods 46, 1042–1051 (2014) 46. De Gelder, B., Van den Stock, J.: The bodily expressive action stimulus test (beast). Construction and validation of a stimulus basis for measuring perception of whole body expression of emotions. Front. Psychol. 2, 181 (2011) 47. Sapi´ nski, T., Kami´ nska, D., Pelikant, A., Ozcinar, C., Avots, E., Anbarjafari, G.: Multimodal database of emotional speech, video and gestures. In: International Conference on Pattern Recognition, pp. 153–163 (2018) 48. Busso, C., Bulut, M., Lee, C., Kazemzadeh, A., Mower, E., Kim, S., Chang, J., Lee, S., Narayanan, S.: Iemocap: interactive emotional dyadic motion capture database. Lang. Resour. Eval. 42, 335–359 (2008) 49. Volkova, E., De La Rosa, S., B¨ ulthoff, H., Mohler, B.: The MPI emotional body expressions database for narrative scenarios. PLoS ONE 9, e113647 (2014)
Experimental Analysis and Verification of a Multi-modal-Biometrics Identity Verification Framework Based on the Dempster-Shafer Theory Alfredo Cuzzocrea1,2(B) , Majid Abbasi Sisara1,3 , and Carmine Gallo1 1 iDEA Lab, University of Calabria, Rende, Italy {alfredo.cuzzocrea,majid.abbasi,carmine.gallo}@unical.it 2 LORIA, University of Lorraine, Nancy, France 3 DIA Department, University of Trieste, Trieste, Italy
Abstract. By extending some previous research contributions, this paper focuses the attention on a state-of-the-art data fusion algorithm for multi-modal biometric identification, which makes use of the Dempster-Shafer (DS) theory for supporting the combination of two different beliefs that can be derived from the algorithm. The specific contributions of this paper are represented by a complete case study that focuses the attention on a reference architecture that supports multi-modal bio-metric identity verification over big data settings, deep overview on state-ofthe-art results, and a comprehensive experimental assessment and analysis of the proposed framework. Keywords: Identity verification · Multi-modal identity verification · Intelligent systems · Dempster-shafer theory
1 Introduction Identity Verification (e.g., [3, 4, 7]) is a state-of-the-art research problem which has gained the attention of numerous communities of researchers in both the academic and industrial sectors. Multi-Modal Identity Verification (e.g., [2, 5, 6, 8]) is a specialized problem where multiple biometrics (e.g., voice, fingerprint, signature, etc.) are applied in order to achieve more robust identification methods. Recently, these issues have become of great interest even as stirred-up by the recent explosion of big data methodologies (e.g., [10–14, 36]), which thus realize emerging trends (e.g., [21–25]). Following some previous research contributions (e.g., [1]), in this paper we present our consolidated data fusion algorithm for multi-modal biometric identification, which combines the fusion of voice and fingerprints. This combination of biometrics is rarely used in verification systems although this biometric pair is simple to use and not too This research has been made in the context of the Excellence Chair in Computer Engineering at LORIA, University of Lorraine, Nancy, France. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 118–129, 2022. https://doi.org/10.1007/978-3-031-08812-4_12
Experimental Analysis and Verification of a Multi-modal-Biometrics Identity
119
invasive. A framework for the combination of several data fusion algorithms is described, while the described method is just a specialization of the general framework. In more details, the Weighted-Sum Fusion algorithm is described. From this algorithm, two beliefs that the identity is verified can be derived. The two beliefs are combined using the Dempster-Shafer (DS) approach to obtain the final decision. The specific contributions of this paper are represented by a complete case study that focuses the attention on a reference architecture that supports multi-modal biometric identity verification over big data settings, deep overview on state-of-the-art results, and a comprehensive experimental assessment and analysis of the proposed framework.
2 Case Study: A Multi-modal Biometric Identity Verification Architecture Over Big Data Settings In this Section, we describe an innovative case study where we show how our proposed framework can be exploited within the context of a multi-modal biometric identity verification architecture over big data settings, with particular regard to the case of big-data-enabled smart city applications (e.g., [15]). Figure 1 shows the described architecture. As well recognized, security is one of the major challenge in big data applications (e.g., [16]), and, specially, in smart city applications, where personal data are very often involved. The proposed identity verification framework is suitable to address this problem, and provide secure access to a plethora of smart city applications via the bimodal combination of voice biometrics and fingerprint biometrics identity verification method implemented by the main algorithm of our framework. To this end, the proposed identity verification framework can be integrated in the reference architecture as shown in Fig. 1, where the architecture is depicted.
Fig. 1. Multi-modal biometric identity verification architecture over big-data-enabled smart city applications.
120
A. Cuzzocrea et al.
Here, the following hierarchical layers can be identified: • Voice and Fingerprint Biometrics Layer: it is the layer where the speech segments and the fingerprints of the architecture users are located. • MMB Identity Verification Layer: it is the layer where our proposed bi-modal identity verification framework is implemented, on top of a Cloud-based infrastructure (e.g., [17]), as to gain into efficiency and scalability over big data. • IoT Layer: it is the layer where the Internet of Things (IoT) (e.g., [18]) devices that play the role of human machine interfaces for the target architecture are located – they are used to receive the speech segments and the fingerprints of users, and locally execute our bi-modal identity verification algorithm as a part of the whole running identity verification process implemented by our framework. • Smart City Layer: it is the layer that identifies the target smart city in terms of collection of hardware and software components, within the reference architecture. • Smart City Applications Layer: it is the layer where next generation smart city applications are located, among which some relevant ones are: (i) smart grid; (ii) smart infrastructure; (iii) smart security; (iv) smart services – it should be noted that, in all such applications, personal (user) data play the major role, so that security issues are of symmetrical relevance for the reference architecture (from here, the significance of our proposed identity verification framework can be derived). Several considerations should be made about the reference architecture shown in Fig. 1. First, when big datasets are considered, performance and scalability play major roles, like highlighted in several studies (e.g., [19]). Indeed, due to the well-known properties of big data repositories (e.g., [20]), classical multi-modal identity verification algorithms cannot be applied like they are, but suitable optimizations must be devised. In this respect, popular big data processing platforms, like Apache Spark and Hadoop, can be useful to this end. Secondly, even heterogeneity of big data repositories (e.g., [20]) is a relevant challenge to be considered when dealing with multi-modal identity verification algorithms over big data, due to the fact that the same identity verification process is based on strongly-heterogeneous sources (e.g., face, hand vein, signature, etc.). The latter is another challenge for emerging big-data-enabled applications (including smart city ones). Finally, the described multi-modal biometric identity verification architecture over big-data-enabled smart city applications has clearly shown the benefits coming from our proposed identity verification framework in emerging big data settings.
3 Supporting Identity Verification via a Multi-modal Biometrics Approach This Section illustrates the general framework of multi-modal biometric approach. We use two kinds of biometrics, i.e. fingerprint and voice. In verification we want to evaluate the probability of the two following events: L1 : the identity is verified (1) L0 : the identity is not verified
Experimental Analysis and Verification of a Multi-modal-Biometrics Identity
121
Fig. 2. Block diagram of the proposed multi-modal biometric identity verification framework.
The Fig. 2 illustrates the proposed framework for the fusion of decisions resulting from multiple data fusion algorithms among the biometric scores. The figure shows that we consider only two types of biometrics and K different data fusion algorithms. The DS theory of evidence [26] combines different measures of evidence, illustrating the uncertainty and lack of knowledge. In this case our hypotheses set is as follow: = {L0 , L1 }
(2)
According to this assumption, there are 2|| possible outcomes: {∅, {L0 }, {L1 }, {L0 , L1 }}
(3)
We define Basic Belief Assignment (BBA) as a function which apply on every subset A of the hypothesis set and assign a value in the range [0, 1]. The bba(.) function define as follows: A⊆ bba(A) = 1 (4) bba(∅) = 0 The belief function, b(.) is defined as follows: bba(A) b(B) =
(5)
A⊆B
It assigns a value in [0, 1] to every nonempty subset B of . The belief function works like a probability function and we can perform it on the outputs of different fusion algorithms (Fig. 2) and convert the scores from the data fusion system into probabilistic value based on the threshold T related to each fusion algorithm. Each data fusion algorithm becomes an expert in DS fusion. The following probability is presented by expert j under this assumption: bbaj ({Ti }) = pj
(6)
122
A. Cuzzocrea et al.
Fig. 3. Block diagram of the proposed bi-modal biometric identity verification framework.
The probability of the subset Ti will be distributed to all the other subsets of , for all the other subsets C of , we obtain the following: bbaj (C) =
1 − pj 2K − 1
(7)
For example, if K is 2 then there are 22 = 4 subsets. Thus, the j expert assigns 1−p bbaj ({Ti }) = pj and bbaj (C) = 3 j , C ⊆ , C = Tj . Two bba1 (.), bba2 (.) can be combined as follows [26]: j,k,Aj Bk =c bba1 (Aj )bba2 (Bk ) (8) bba(C) = bba1 ⊕ bba2 = 1 − j,k,Aj Bk =φ bba1 (Aj )bba2 (Bk ) where Aj and Bk are subset of . Also, we can combine more than two bba(.) in same way. Figure 3 represents the algorithm proposed in this paper, where two biometrics and two data fusion algorithms are used. Voice and fingerprint biometrics recorded from the same person and the system uses Gaussian Mixtures and a list of minutiae respectively for classify person in authorized or impostors group. In order to generate a score, called D, comparing the two representations of authorized persons and the voice and fingerprints of the impostors, it is necessary to compare the two sets of voices and fingerprints to achieve the following property: if the person is allowed to access the resource, the score will be high. If the person is not, the score will be low. In the next section, we will discuss how the scores are determined.
4 An Innovative Weighted-Sum Data Fusion Algorithm for Supporting Multi-modal Biometrics Identity Verification In this Section first we explain how to compute scores for identity verification via voice biometrics and fingerprint biometrics, then we apply weighted-sum data fusion algorithm on these scores for supporting multi-modal biometrics identity verification.
Experimental Analysis and Verification of a Multi-modal-Biometrics Identity
123
For computing the speech-segment score, first, we divide speech in short frames, then each frame is parameterized by Mel Frequency Cepstral Coefficients (e.g., [28]). Speech models described by GMM, such as those in [27] are a mixture of Gaussian distributions. To estimate the GMM parameters we need to have a training phase and we use Maximum Likelihood (ML) technique for this purpose. An iterative procedure can be used instead to obtain ML estimations of the parameters using an Expectation Maximization (EM) algorithm. A-priori initialization of the model parameters before the EM algorithm and selecting the M order of the mixture are two critical aspects of training. The first problem can be solved experimentally and, for the second one K-means algorithm can be used. The GMM model of an authorized person, DAS , and the average GMM model of the S , enable us to compute the score D as follows: non-authorized person, DNA S p x|DAS DS = S p x|DNA
(9)
where x is a segment of speech and the S apex stands clearly for speech. Eq. (9) in logarithmic terms, becomes as follows:
S (10) DS = logDS = logp x|DAS − logp x|DNA In the data fusion algorithms, the latter term is the score of speech. For computing the fingerprint score of the person to identify, as for example described in [28], we also use a variant based on minutiae detection. First, the fingerprint is scanned by sensors and then its image is divided into blocks and determine the direction and frequency of the ridges. Then we create a mask in black and white colors respectively for damaged and valid areas of image. For enhancement the image and to better highlight the edges of the crests, a Gabor filter is applied. After that we use a threshold binarization the image to simplify the next operation. Thinning is the process of reducing all crests within a binarized image to the same thickness as a single pixel. At the and we apply the minutiae extraction and filtering operator to check verification. The score between the two fingerprints is computed as follows: K score(S, V ) = min
k=1 d (k)w(k) K k=1 w(k)
(11)
where S is pattern (from “Stored”) and V (from “to be Verified”). The minimum is computed based on the length of all warping paths, w(k) is a weighting factor, and the denominator is the length of each warping path. d (.) is the Euclidean metrics of k point from S and V . Using the score of authorized (A) and the average score of non-authorized (NA), we have: DF =
score(S, V )A score(S, V )NA
DF = logDF = logscore(S, V )A − logscore(S, V )NA
(12) (13)
124
A. Cuzzocrea et al.
This term is the score used in the data fusion algorithms. Now the normalized quantities DF and DS are ready and the Weighted-Sum Fusion is used to accept or reject identity of a person. We use a weighted average combination, as follows: D = γ DF + (1 − γ )DS
(14)
D is the index which obtained as a result of the combination. A threshold τ is used to accept or reject the declared identity as follows: D ≥ τ → identity could be verified (15) D < τ → identity could be rejected γ and τ are two important parameters in final decision (reject or accept), then we need to calculate them. We use a simple optimization approach that will be described below. During a training phase, we enter a sequence of observations by the authorized person as input to the system: O1F , O2F , . . . . , OFK
(16)
O1S , O2S , . . . . , OSK
(17)
and:
Also there is a sequence of DF and DS values, respectively, as follows: D1F , D2F , . . . , DKF
(18)
D1S , D2S , . . . , DKS
(19)
and:
Next, we calculate the mean and variance of each sequence, as follows: μ(DF ) μ(DS ) and:
σ 2 (DF ) σ 2 (DS )
The mean of D is a good value for the threshold, denoted as follows:
τ = γ μ DF + (1 − γ )μ(DS )
(20)
(21)
(22)
Also, if an authorized person provides fingerprint and vocal observations, D value is high, however its variance shows some variability. The variance of D, however, should be
Experimental Analysis and Verification of a Multi-modal-Biometrics Identity
125
minimized to reduce the number of false positives and false negatives in the verification process. Based on the Eq. (14) the σ 2 (D) is defined as follows: σ 2 (D) = γ 2 σ 2 DF + (1 − γ )2 σ 2 DS + 2γ (1 − γ )Cov(DF , DS )
(23)
since the fingerprint and voice are statistically independent, the last term of the Eq. (23) is zero. By setting to zero the derivative of σ 2 (D) with respect to γ , the value of γ is obtained, as follows: γ =
σ 2 (DS ) σ 2 (DF )
(24)
Therefore, the algorithm developed for biometry is shown in Fig. 4. Algorithm Multi_Modal_Biometrics(DF,DS) Input speech scores DS; fingerprint scores DF Output final decision L import Math; Begin double y; # weighting coefficient double t; # threshold double D; double mS; # mean of DS double mF; # mean of DF double vS; # variance of DS double vF; # variance of DF int L; # final decision # Calculate mean and variance mS = Math.mean(DS); mF = Math.mean(DF); vS = Math.var(DS); vF = Math.var(DF); # Calculate weight coefficient and threshold y = vS/vF; t = y*mF + (1-y)*mS; # Final decision D = y*DF + (1-y)*DS; if D ≥ t then L = 1; else L = 0; return L End; Fig. 4. The multi-modal biometrics algorithm.
5 Experimental Assessment and Analysis In this Section, we provide the experimental assessment and analysis of our proposed identity verification framework.
126
A. Cuzzocrea et al.
Since no public datasets for these biometrics are easily available, we developed our own dataset as described next in this Section. Let us now first describe the speaker verification operations. Some operations were performed with the help of the tools provided by the ALIZELIA RAL Speaker Verification Toolkit [29–31]. For the training and testing phases of the identity verification algorithm, we have developed a dataset made up of vocal samples from 50 students, 15 females and 35 males, average age of 28 years. In particular, the subjects were required to read a series of 30 words 4 times each word. The acquired data is used for the first training phase and for the subsequent testing phase. The audio files were recorded with a slight background noise in order to evaluate the performance of the programs in the presence of a system that has noises (the average signal/noise ratio of the files is 15 dB). The group of people is divided into 20 authorized and 30 unauthorized. For each of the 20 authorized, a model of each word is constructed using two repetitions of each word; the other two repetitions are used to obtain false positives in the test phase. Of the 30 not used, 20 are used to build the unauthorized model and 10 for the false negative test. Ultimately, the dataset we have created is made up of 6, 000 files. Of these, 1, 200 files are used for training and 1, 200 for false positive testing. In addition, 2, 400 files are used to build the unauthorized model and 1, 200 for the false negative test. All 6, 000 files are converted into Cepstral parameters using the bin/sfbcep tool. Voice detection is provided by the bin/EnergyDetector tool. The normalization of the Cepstral parameters is realized with the bin/NormFeat tool while the model of the unauthorized with the bin/TrainWorld tool. The bin/TrainTarget tool is used for creating templates of authorized users. Finally, the tests are performed with the bin/ComputeTest tool. Fingerprint identity verification has been performed using a series of tools developed by us in C++. From speaker and fingerprint verification modules, the respective scores are obtained. For the other biometrics, i.e. face, hand vein and signature, we performed what follows. For every student, face has been photographed 3 times, while hand vein and signature have been collected 5 and 12 times, respectively. Since the latter biometrics are not central in our work, we considered alternative approaches for implementing their identification. In particular, we considered [32] for face identification, [33] for hand vein identification, and [34] for signature identification, respectively. Similarly, to our main approach, even in all this implementation we considered, for each biometrics, a sort of identification score, which is simple modeled as the identification likelihood provided by the current technique adopted to perform the target identification process. For what regards our identity identification framework, the data fusion algorithms, the conversion of respective scores to likelihood and the DS combination rule of the two belief functions are developed by us in C++ to obtain the final result. Instead, for the other biometries, we implemented the methods with we considered among the most suitable that are available in literature (i.e., [32] for face identification, [33] for hand vein identification, and [34] for signature id33entification, respectively). In our experiment, we mostly focused on the accuracy of identification, and we compared our voice and fingerprint pair with the following biometric pairs: (i) face and voice; (ii) face and signature; (iii) face and hand vein; (iv) hand vein and signature. This while ranging the number of tentatives performed by accessing the current identification
Experimental Analysis and Verification of a Multi-modal-Biometrics Identity
127
Fig. 5. Accuracy Analysis: Comparison between the Pairs voice, fingerprint and face, voice (a), and Pairs voice, fingerprint and face, signature (b).
procedure. Figure 5(a) shows the experimental results comparing the voice and fingerprint pair with the face and voice pair, while Fig. 5(b) shows the same experimental pattern for the comparison with the face and signature pair. Similarly, Fig. 6(a) shows the experimental results comparing the voice and fingerprint pair with the face and hand vein pair, while Fig. 6(b) shows the same experimental pattern for the comparison with the hand vein and signature pair.
Fig. 6. Accuracy Analysis: Comparison between the Pairs voice, fingerprint face, handvein (a), and Pairs voice, fingerprint and handvein, signature (b).
and
6 Conclusions and Future Work By extending some previous research contributions, this paper has focused the attention on a state-of-the-art data fusion algorithm for multi-modal biometric identification, which makes use of the DS theory for supporting the combination of two different beliefs that can be derived from the algorithm. Deep analysis of state-of-the-art proposals and comprehensive experimental assessment and analysis have been described in details. Future works mainly concerns with extends the actual contribution towards adaptive metaphors, perhaps inspired by distant-but-conceptually-related research experiences (e.g., [35]).
128
A. Cuzzocrea et al.
Acknowledgements. This research has been partially supported by the French PIA project “Lorraine Université d’Excellence”, reference ANR-15-IDEX-04-LUE.
References 1. Cuzzocrea, A., Mumolo, E.: Dempster-shafer-based fusion of multi-modal biometrics for supporting identity verification effectively and efficiently. In: 2nd IEEE International Conference on Human-Machine Systems, ICHMS 2021, Proceedings, pp. 1–8 (2021) 2. Wang, Y., Tan, T., Jain, A.K.: Combining face and iris biometrics for identity verification. In: Proceedings of International Conference on Audio- and Video-Based Biometric Person Authentication, pp. 805–813 (2003) 3. Ashbourn, J.: Biometrics: Advanced Identity Verification: The Complete Guide. Springer (2014) 4. Kittler, J., Matas, J., Jonsson, K., Ramos Sánchez, M.U.: Combining evidence in personal identity verification systems. Pattern Recognit. Lett. 18(9), 845–852 (1997) 5. Xu, J., Cha, M., Heyman, J.L., Venugopalan, S., Abiantun, R., Savvides, M.: Robust local binary pattern feature sets for periocular biometric identification. In: Proceedings of 2010 Fourth IEEE International Conference on Biometrics: Theory, Applications and Systems (BTAS), , pp. 1–8 (2010) 6. da Costa, D.M.M., dos Santos Passos, H., Peres, S.M., Lima, C.A.M.: A comparative study of feature level fusion strategies for multimodal biometric systems based on face and Iris. In: Proceedings of the Annual Conference on Brazilian Symposium on Information Systems, Information Systems: A Computer Socio-Technical Perspective, SBSI 2015, Goiania, Brazil, May 26–29, 2015, pp. 219–226 (2015) 7. Chiu, C.-C., Chuang, C.-M., Hsu, C.-Y.: A novel personal identity verification approach using a discrete wavelet transform of the ECG signal. In: Proceedings of 2008 International Conference on Multimedia and Ubiquitous Engineering (MUE 2008), pp. 201–206 (2008) 8. Li, Y., et al.: A survey of multimodal fusion for identity verification. Journal of Physics Conference Series, pp. 1–5 (2020) 9. Wang, Y., Wang, Y., Tan, T.: Combining fingerprint and voiceprint biometrics for identity verification: an experimental comparison. In: Zhang, D., Jain, A.K. (eds.) ICBA 2004. LNCS, vol. 3072, pp. 663–670. Springer, Heidelberg (2004). https://doi.org/10.1007/978-3-540-259480_90 10. Cuzzocrea, A., Chakravarthy, S.: Event-based lossy compression for effective and efficient OLAP over data streams. Data Knowl. Eng. 69(7), 678–708 (2010) 11. Bonifati, A., Cuzzocrea, A.: efficient fragmentation of large XML documents. In: Wagner, R., Revell, N., Pernul, G. (eds.) DEXA 2007. LNCS, vol. 4653, pp. 539–550. Springer, Heidelberg (2007). https://doi.org/10.1007/978-3-540-74469-6_53 12. Cuzzocrea, A., Furfaro, F., Saccà, D.: Enabling OLAP in mobile environments via intelligent data cube compression techniques. J. Intell. Inf. Syst. 33(2), 95–143 (2009) 13. Cuzzocrea, A., Darmont, J., Mahboubi, H.: Fragmenting Very Large XML data warehouses via K-means clustering algorithm. Int. J. Bus. Intell. Data Min. 4(3/4), 301–328 (2009) 14. Ceci, M., Cuzzocrea, A., Malerba, D.: Effectively and efficiently supporting roll-up and drilldown OLAP operations over continuous dimensions via hierarchical clustering. J. Intell. Inf. Syst. 44(3), 309–333 (2015) 15. Pramanik, M.I., Lau, R.Y., Demirkan, H., Azad, M.A.K.: Smart health: big data enabled health paradigm within smart cities. Expert Syst. Appl. 87, 370–383 (2017) 16. Lafuente, G.: The big data security challenge. Netw. Secur. 2015(1), 12–14 (2015)
Experimental Analysis and Verification of a Multi-modal-Biometrics Identity
129
17. Kloch, C., Petersen, E.B., Madsen, O.B.: Cloud based infrastructure, the new business possibilities and barriers. Wireless Pers. Commun. 58(1), 17–30 (2011) 18. Li, S., Da Xu, L., Zhao, S.: The internet of things: a survey. Inf. Syst. Front. 17(2), 243–259 (2015) 19. Tolosana, R., Vera-Rodriguez, R., Fierrez, J., Ortega-Garcia, J.: Deepsign: deep on-line signature verification. IEEE Trans. Biom. Behav. Identity Sci. 3(2), 229–239 (2021) 20. Marx, V.: The big challenges of big data. Nature 498(7453), 255–260 (2013) 21. Verlinde, P., Chollet, G., Acheroy, M.: Multi-modal identity verification using expert fusion. Inf. Fusion 1(1), 17–33 (2000) 22. Oh, K., Oh, B., Toh, K., Yau, W., Eng, H.: Combining sclera and periocular features for multi-modal identity verification. Neurocomputing 128, 185–198 (2014) 23. Zhang, S., et al.: CASIA-SURF: a large-scale multi-modal benchmark for face anti-spoofing. IEEE Trans. Biom. Behav. Identity Sci. 2(2), 182–193 (2020) 24. Bai, Q., Xia, W., Yin, F., Yang, Y.: Identity-Guided Face Generation with Multi-Modal Contour Conditions, CoRR, vol. abs/2110.04854 (2021) 25. Spinoulas, L., Mirzaalian, H., Hussein, M.E., Abd Almageed, W.: Multi-modal fingerprint presentation attack detection: evaluation on a new dataset. IEEE Trans. Biom. Behav. Identity Sci. 3(3), 347–364 (2021) 26. Shafer, G.: A mathematical theory of evidence turns 40. Int. J. Approx. Reason. 79, 7–25 (2016) 27. Bimbot, F., et al.: A tutorial on text-independent speaker verification. EURASIP J. Adv. Signal Process. 2004(4), 1–22 (2004). https://doi.org/10.1155/S1110865704310024 28. Molau, S., Pitz, M., Schluter, R., Ney, H.: Computing mel-frequency cepstral coefficients on the power spectrum. In: 2001 IEEE International Conference on Acoustics, Speech, And Signal Processing. Proceedings (cat. No. 01CH37221), vol. 1, pp. 73–76 (2001) 29. Yager, N., Amin, A.: Fingerprint verification based on minutiae features: a review. Pattern Anal. Appl. 7(1), 94–113 (2004) 30. Bellomo, F., Beritelli, F., Sciacca, E.: Robustness of forensic speaker verification systems based on alizelia_ral toolkit. In: Proceedings of IEEE Workshop on Biometric Measurements and Systems for Security and Medical Applications (BIOMS), pp. 1–6 (2014) 31. Bonastre, J.F., Wils, F., Meignier, S.: Alize, a free toolkit for speaker recognition. In: IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP ’05, Philadelphia, Pennsylvania, USA, March 18–23, 2005, pp. 737–740 (2005) 32. Campanella, S., Belin, P.: Integrating face and voice in person perception. Trends Cogn. Sci. 11(12), 535–543 (2007) 33. Kumar, A., Wong, D.C.M., Shen, H.C., Jain, A.K.: Personal verification using palmprint and hand geometry biometric. In: Kittler, J., Nixon, M.S. (eds.) AVBPA 2003. LNCS, vol. 2688, pp. 668–678. Springer, Heidelberg (2003). https://doi.org/10.1007/3-540-44887-X_78 34. Monwar, M.M., Gavrilova, M.: Fes: a system for combining face, ear and signature biometrics using rank level fusion. In: Proceedings of Fifth International Conference on Information Technology: New Generations (ITNG 2008), pp. 922–927 (2008) 35. Cannataro, M., Cuzzocrea, A., Mastroianni, C., Ortale, R., Pugliese, A.: Modeling adaptive hypermedia with an object-oriented approach and XML. In: Proceedings of the Second International Workshop on Web Dynamics, WebDyn@WWW 2002, pp. 35–44 (2002) 36. Bellatreche, L., Cuzzocrea, A., Benkrid, S.: F&A: a methodology for effectively and efficiently designing parallel relational data warehouses on heterogenous database clusters. In: Proceedings of the 12th International Conference on Data Warehousing and Knowledge Discovery (DAWAK 2010), pp. 89–104 (2010)
Breast Ultrasound Image Classification Using EfficientNetV2 and Shallow Neural Network Architectures Hai Thanh Nguyen, Linh Ngoc Le, Trang Minh Vo, Diem Ngoc Thi Pham, and Dien Thanh Tran(B) Can Tho University, Can Tho, Vietnam {nthai.cit,ptndiem,ttdien}@ctu.edu.vn
Abstract. Health is the foundation of life. Along with the development of science and technology, and modern medicine, there are many methods to help diagnose diseases quickly and accurately detect diseases early with ultrasound images are the most popular and effective technique. This study has investigated and explored methods of predicting breast cancer based on ultrasound images classified into three categories: normal, benign, and malignant, with three deep learning techniques such as Fully Connected neural Network (FCN), Shallow Convolutional Neural Networks (CNN), and EfficientNetV2. Such techniques are evaluated on both the original data set and the dataset are applied data augmentation techniques to compare the efficiency of each algorithm and evaluate the influence of increasing data on the training phase. Experimental results show that the recent emerging CNN architecture, EfficientNetV2 provides the highest classification performance when taking the lead in all three metrics of ACC, AUC, and MCC with 0.77, 0.804, and 0.619, respectively. The results also show that data augmentation techniques can improve the efficiency of the classification algorithms on the ultrasound image dataset and are expected to combine the recent emerging CNN architectures to provide efficient disease diagnosis on ultrasound images. Keywords: Ultrasound images · Breast cancer Image classification · Data augmentation
1
· Neural networks ·
Introduction
In today’s modern life, health is essential for us. It is the foundation of life. Having new health can lead to a happy, fulfilling life. Everyone needs to detect diseases early and timely control. In addition, the doctor coordinates the medical examination and the paraclinical examination during the general physical examination. It is clinically based on disease symptoms, sensation, palpation, percussion, and patient history. Clinical learning will perform tests image processing techniques obtained from medical equipment such as X-rays, MRI, and ultrasound of organs in the body to determine the body’s health status, patient health, and disease c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 130–142, 2022. https://doi.org/10.1007/978-3-031-08812-4_13
Breast Ultrasound Image Classification Using EfficientNetV2
131
risk. Today, with the development of modern science and medicine, there are many methods to support rapid disease detection, detect the manifestations on the body, and give timely control to the user disease use. With the development of artificial intelligence applied in image processing, tomography images in medical images, with high accuracy, help doctors easily identify tumors in a short time and without giving away up to the other body of the base. Ultrasound is a non-classified, nondispersive, non-clinical method using sound waves as an easy, fast, and affordable disease control tool, effectively supporting doctors in treating diseases. Diagnostic. In cases where symptoms, diseases, or other disease symbols are proven, the doctor may ask the patient to perform an ultrasound method to see the overall and detailed images of the internal organs. Safety and patient monitoring are widely applied in medicine. For example, doctors can measure the size corresponding to particular organs in the abdominal cavity (liver, spleen, fascia, breast, and breast tumor) and detect abnormal tumors based on ultrasound imaging. In addition, it is possible to determine the structure and size of the working time, the heart valves, and the great vessels from ultrasound imaging. In the production department, ultrasound helps to identify and monitor the growth of the fetus in the womb, detect fetal malformations, etc. Leveraging advancements in deep learning techniques in image classification, we have deployed and evaluated recent learning algorithms to diagnose breast diseases on ultrasound images. The principal contributions of this study include: – We have investigated the efficiency of several classification techniques from traditional neural and convolutional ones (Fully Connected Network, shallow Convolutional Neural Networks) to recent emerging architectures (EfficientNetV2) on ultrasound images dataset to diagnose breast cancer. EfficientNetV2 was recently proposed in [1] with a remarkable performance with 87.3% top-1 accuracy on ImageNet ILSVRC2012 [2], so an evaluation on its efficiency on ultrasound images which is challenging to numerous researchers, is essential. However, as observed from our experiments, EfficientNetV2 outperforms considered shallow neural networks in all considered metrics. – Data augmentation techniques are applied to increase data size in the training phase, including adjusting brightness, zooming in, zooming out, flipping the image, or rotating the image. The results show that such techniques improve classification performance by developing all three ACC, AUC, and MCC measures. The rest of the paper is organized as follows. First, we will discuss some related work in Sect. 2. Then, the following section describes our workflow for breast ultrasound images (Sect. 3). Subsequently, The results are presented in Sect. 4. Finally, in the conclusion part, we will summarize key features and our development plan (Sect. 5).
2
Related Work
In [3], Ilovitsh et al. developed this breakthrough technology during research in the lab of Professor Katherine Ferrara at Stanford University. This technique
132
H. T. Nguyen et al.
uses low-frequency ultrasound (250 kHz) to detonate bubbles that target microscopic tumors. In vivo test, cell destruction reached 80% tumor cells. Microbubbles are microscopic bubbles filled with gas, with a diameter as small as onetenth the size of a blood vessel. Sound waves cause microbubbles to behave like balloons at a particular frequency and pressure: they periodically expand and contract. This process increases the conduction of substances from the blood vessels into the surrounding tissue. Targeted treatments are safe and effective and can destroy most tumors. The authors in [4] presented a method to distinguish between benign and malignant tumors on breast ultrasound using deep learning with neural networks and using deep learning with complex neural networks (CNN) to distinguish between benign and malignant tumor images from ultrasound. The work in [5] provided expert-level prenatal detection of complex congenital heart disease. Congenital heart disease (CHD) is the most common congenital disability. Fetal screening ultrasound provides five heart views that can detect 90% of complex CHD. Using 107,823 images from 1,326 retrospective echocardiograms and screening ultrasounds from 18 to 24 weeks fetuses, a neural network was created to identify recommended cardiac imaging and differentiate between normal hearts and complex hearts. In [6], Yi-Cheng Zhu et al. built a Visual Geometry Group (VGG) - 16T model to classify thyroid nodules (TNs) as benign or malignant. The model is built based on the VGG-16 architecture but adds bulk normalization (BN) layers and removes fully connected layers. The results show that model VGG-16T has high performance, sensitivity, and accuracy in TN classification. In addition, the acting of the models is considered superior to that of experienced doctors. Sudharson et al. in [7] proposed a method to automatically classify B-mode renal ultrasound images with four classes: normal, cyst, stone, and tumor, based on a set of deep neural networks (DNNs) using transfer learning. Variant datasets are fed to models pre-trained for feature extraction and then classed with the help of a vector machine. The final results of the trained DNNs, including ResNet101, ShuffleNet, and MobileNet-v2, were obtained based on the majority voting method. Using a combination of predictions from multiple DNNs, the composite model has shown superiority compared with conventional individual classification models. In [8], Sudipan Saha and Nasrullah Sheikh proposed an ultrasound image classification model, focusing on a headache in the field of machine learning - small data sets. This method uses the Auxiliary Classifier Generative Adversarial Network (ACGAN) combined with the Generative Adversarial Network (GAN) and a traditional classifier and transfer learning. The team tested on a dataset of breast ultrasound images consisting of 250 images and achieved high results. Using ACGAN helps the model learn features superior to traditional classification models. In [9], Weibin Chen and associates. A New Classification Method in Ultrasound Images of Benign and Malignant Thyroid Nodules Based on Transfer Learning and Deep Convolutional Neural Network. The ultrasound imaging diagnostics method for identifying thyroid nodules by Weibin Chen et al. is based on transfer learning and deep convolutional neural network (CNN), classifying thyroid ultrasound images into benign and malignant. The authors
Breast Ultrasound Image Classification Using EfficientNetV2
133
propose to use the transfer learning method with GoogLeNet convolutional neural network trained to extract image features and then perform joint training and secondary transfer learning to improve model accuracy. The GoogLeNet (improved) model tested on the thyroid ultrasound image dataset showed superior results to the networks established with LeNet5, VGG16, and GoogLeNet.
3
Methods
This section presents primary techniques to classify breast ultrasound images with the workflow as detailed in Fig. 1. First, we begin with Data augmentation techniques to increase the size of considered data. Deep learning usually shows poor performance when fetched with only minor data. Therefore, it is necessary to increase the size by several Data augmentation methods to improve the classification tasks’ performance. Following sections, we describe the considered learning algorithm from traditional neural ones, shallow convolutional neural networks, and emerging deep learning architecture, EfficientNetV2 [1].
Fig. 1. The workflow for the proposed method.
3.1
Data Augmentation
Currently, in deep learning, data problems play a significant role. Therefore, there are regions with little data, so it is not easy to train the model to produce good results in prediction. Hence one needs a technique called data augmentation to serve that if you have little data, you can still generate more data based on existing data. It can be seen that data has been playing an essential role in
134
H. T. Nguyen et al.
science and life. In Machine Learning in general and Deep Learning models in particular, it is necessary to have a good enough data set to build a model with high accuracy. This problem has been mentioned in the study “Data augmentation to improve deep learning in image classification problems” by Agnieszka Mikolajczyk and Michal Grochowski [10], To solve the above problem, people often find ways to enhance data by many methods such as “Collect more data”. However, data collection is often costly in time, effort, and money. In this study, we use data augmentation, a technique used to expand a data set by modifying existing data samples or creating newly aggregated data. More specifically, in this ultrasonic image classification problem, we use techniques such as adjusting brightness, zooming in/out, flipping the image, or rotating and translating the image to increase the number of samples in the data set, as illustrated in Fig. 2.
Fig. 2. Examples of images created from data augmentation. The top left is the original image, followed by the converted, flipped, horizontally shifted, rotated, zoomed, and dimmable images from the original image.
3.2
Fully Connected Neural Network (FCN)
A fully connected network consists of layers in which each neuron of one layer is fully connected to every neuron of the other layer. The advantage of fully connected networks is that they are “Structure agnotic”, i.e., no special assumptions are needed with the network’s input. With this property, the fully connected network can be widely applied in different fields of machine learning, for example, classification [10], and image segmentation [11]. This work has deployed a fully connected network with an input size of 192 × 192 and 3 Dense (neural) layers (or fully-connected layer) using ReLU activation and a Dropout of 0.1 while the
Breast Ultrasound Image Classification Using EfficientNetV2
135
output size is 3, representing three classes, including normal, benign, and malignant. However, it is the “Structure agnotic” property that makes fully connected networks perform worse than special networks that are fine-tuned according to the spatial structure of the problem. Therefore, we propose other models to apply and compare in this ultrasound image classification problem. 3.3
Convolutional Neural Networks (CNNs)
Convolutional neural networks (CNNs) introduced by LeCun et al. [12] are a class of deep neural networks that consist of overlapping convolutional layers and use nonlinear activation functions such as ReLU and Tanh to enable the weights in the nodes. It also uses built-in weights and probe classes. This architecture allows CNNs to take advantage of the 2D structure of the input data, which is why CNNS are widely used in problems where the input is an image, especially in image classification [13,14]. We use CNNs network extended from (Fig. 3) with input image size of 192 × 192 × 3 (color images). CNNs contain four convolutional layers (Conv2D) followed by a MaxPooling layer, a Flatten layer, and a fully connected layer. The convolution layers play the role of extracting features from the input image. Our convolutional layers contain 32, 64, 128, 256 filters, with each filter being a matrix of numbers 3 × 3, the Regularizer function used for kernel weight matrix is “L2”. The MaxPooling layer is used to reduce the input data size but still keep the main attributes of the image. In the CNNs network with many convolutional layers creating many Feature Maps, we will have many MaxPooling layers for each Feature Map, followed by convolutional layers. Next is the Dense layer or Fully-connected layer. In this layer, each neuron receives input from all the neurons of the previous layer. Activation in the model is Softmax. Softmax activation is widely used for classification in use optimizer Adam [15], with learning rate = 0.0005, loss function is “categorical cross-entropy” function that calculates loss between actual and predicted labels. The model has a total parameter of about 1.5 million parameters. 3.4
EfficientNetV2
EfficientNetV2 is a new family of complex networks built on top of EfficientNetV1 [16], introduced by Mingxing Tan and Quoc V. Le in [1]. EfficientNetV2 includes a family of classification models with better accuracy, smaller size, and faster speed than previous models. EfficientNetV2 has tested the efficiency of the ImageNet dataset [1], and it achieved top 1 accuracy of 85.7% with size 6.8 times smaller and a training speed three times faster.
136
H. T. Nguyen et al.
Fig. 3. The structure of convolutional neural networks is used in this study.
Fig. 4. The details of EfficientNetV2 architecture.
We use the EfficientNetV2 model pre-trained on ImageNet21k with input as an image set of size 192 × 192 × 3, the loss function is “categorical crossentropy” with a label smoothing of 0.1, using Adam optimizer, has a learning rate of 0.0001. The model has a total of 5.9 million parameters with the structure described in Fig. 4.
Breast Ultrasound Image Classification Using EfficientNetV2
4
137
Results
This section presents our experimental results with the Breast Cancer dataset obtained from the National Center for Biotechnology Information (NCBI) database. The dataset will be processed. The images are resized to fit the input of the models. All models trained in 100 epochs and the model at the epoch with the best performance will be saved to evaluate the model on the following measures: ACC, AUC, and MCC. Detailed experimental results of the models and assessments are shown in the section. In this section, we evaluate the models based on the Breast Cancer dataset [17], obtained from the NCBI (National Center for Biotechnology Information) database. We have changed different params sets including ckpt type = 1k, 21k, 21k-ft1k and hub type = classification, feature-vector. The study continues to select hyperparameters for the classification model by changing two parameters of the EfficientNetV2 model, ckpt type, and hub type. In which, ckpt type (checkpoint type) allows to select the dataset where the model is trained with a set of values including: 1k (ImageNet ILSVRC2012 [20]), 21k (ImageNet21k [20]), 21k -ft1k (combining 2 datasets ImageNet ILSVRC2012 and ImageNet21k). Hub type has two values: classification or vector extraction (feature-vector). From the experimental results (presented in Table 2), data enhancement techniques still improve the performance of the model in different sets of parameters, and the model achieves the best results when using ckpt type 1k. with hub type feature-vector when reaching accuracy 0.79, AUC: 0.678, MCC: 0.641 on the evaluation dataset. The study continues to resize the input image of the model with the aim of evaluating the impact of the input image size on the performance of the machine learning model. Due to configuration limitations of the running environment, the experiment changed the image to 64 × 64, 128 × 128, and 164 × 164 sizes, running on the hyperparameters ckpt type 1k and hub type feature-vector just tested. for the highest accuracy. The results are presented in Table 3. 4.1
Data Description
The dataset collected in 2018 included 780 breast ultrasound images of 600 women aged 25 to 75 years. The average image size is 500 × 500 pixels, PNG format, consisting of 133 normal, 437 benign, and 210 malignant samples. [17], obtained from the NCBI (National Center for Biotechnology Information) database. 4.2
Classification Results on Breast Ultrasound Images of Three considered Algorithms
The dataset is divided into two parts, train, and test with a ratio of 4:1, including ultrasound images taken from the Breast Cancer dataset, images that are uniformly resized to 192 × 192 are used as input. For models (limited image size due to computer configuration issues). The training dataset is cloned into two parts, one part is kept intact, and the other part applies the Data augmentation
138
H. T. Nguyen et al.
Table 1. The classification performance of various methods on breast ultrasound images. Model FCN
Data augmentation Fold No augmentation
ACC
AUC
MCC
Train Test
Train Test
Train Test
1
0.340 0.354 0.459 0.432 0.319 0.173
2
0.315 0.324 0.478 0.444 0.363 0.341
3
0.338 0.342 0.429 0.425 0.352 0.296
4
0.338 0.331 0.452 0.494 0.312 0.302
5
0.320 0.310 0.480 0.419 0.312 0.369
Average 0.329 0.332 0.460 0.443 0.332 0.296 With augmentation 1
0.358 0.371 0.450 0.397 0.358 0.206
2
0.330 0.340 0.454 0.429 0.372 0.310
3
0.341 0.347 0.422 0.451 0.347 0.326
4
0.337 0.346 0.439 0.496 0.364 0.334
5
0.354 0.348 0.514 0.500 0.089 0.000
Average 0.343 0.351 0.456 0.455 0.306 0.235 CNNs
No augmentation
1
0.563 0.558 0.499 0.500 0.048 0.000
2
0.566 0.577 0.497 0.488 0.083 0.167
3
0.561 0.558 0.500 0.500 0.000 0.000
4
0.561 0.564 0.505 0.496 0.051 0.030
5
0.577 0.577 0.504 0.508 0.154 0.123
Average 0.565 0.567 0.501 0.498 0.067 0.064 With augmentation 1
0.561 0.558 0.500 0.500 0.000 0.000
2
0.659 0.635 0.481 0.484 0.365 0.312
3
0.569 0.558 0.509 0.500 0.097 0.000
4
0.559 0.564 0.500 0.519 0.000 0.022
5
0.559 0.564 0.500 0.500 0.000 0.000
Average 0.581 0.576 0.498 0.501 0.092 0.067 EfficientNetV2 No augmentation
1
0.942 0.750 0.972 0.824 0.905 0.585
2
0.941 0.833 0.962 0.837 0.902 0.715
3
0.928 0.705 0.964 0.673 0.881 0.505
4
0.936 0.750 0.970 0.832 0.895 0.580
5
0.883 0.750 0.955 0.823 0.809 0.598
Average 0.926 0.758 0.965 0.798 0.878 0.597 With augmentation 1
0.929 0.763 0.976 0.794 0.882 0.597
2
0.950 0.801 0.978 0.818 0.917 0,662
3
0.928 0.756 0.962 0.721 0.879 0.580
4
0.942 0.795 0.968 0.857 0.906 0.668
5
0.893 0.737 0.945 0.829 0.827 0.589
Average 0.928 0.770 0.966 0.804 0.882 0.619
Breast Ultrasound Image Classification Using EfficientNetV2
139
technique (mentioned in Sect. 3.1) to evaluate the influence of the data augmentation method on accuracy of the model. Because the data set is relatively small, we use Stratified 5-Fold Cross-Validation to evaluate the model objectively and accurately. The results are presented in Table 1 as follows. The table shows the results of a Fully Connected Neural Network, a Shallow Convolutional Neural Network (with the architecture revealed in Fig. 3) and EfficientNetV2 (Fig. 4) architecture trained and evaluated through Stratified 5-Fold Validation 100 epochs with two types of data sets with data augmentation and without applying data augmentation (Table 1). Table 2. The classification performance of various methods on breast ultrasound images Model
Data augmentation ACC Train Test
1k classification 21k classification 21k-ft1k classification 1k feature-vector 21k feature-vector
AUC
MCC
Train Test
Train Test
Without
0,963 0.762 0.921 0,636 0.937 0.584
With
0.964 0.765 0.918 0.642 0.938 0.592
Without
0.904 0.688 0.94
With
0.891 0.687 0.941 0.785 0.829 0.496
0.792 0.845 0.487
Without
0.927 0.75
0.947 0.841 0.876 0.571
With
0.947 0.76
0.965 0.86
Without
0.968 0.764 0.944 0.639 0.945 0.595
With
0.968 0.79
Without
0.957 0.774 0.951 0.69
With
0.959 0.788 0.935 0.692 0.931 0.645
21k-ft1k feature-vector Without With
0.91
0.586
0.941 0.678 0.946 0.641 0.929 0.627
0.939 0.753 0.963 0.781 0.901 0.596 0.955 0.773 0.972 0.785 0.925 0.625
Table 3. Resize image for experiment Model
Data augmentation ACC Train Test
1k feature-vector 64
AUC
MCC
Train Test
Train Test
Without
0.802 0.674 0.742 0.607 0.659 0.424
With
0.793 0.699 0.757 0.601 0.638 0.469
1k feature-vector 128 Without With 1k feature-vector 192 Without With
0.955 0.741 0.977 0.8
0.925 0.558
0.961 0.737 0.972 0.796 0.934 0.547 0.968 0.764 0.944 0.639 0.945 0.595 0.968 0.79
0.941 0.678 0.946 0.641
First, looking at both bar charts, we can easily see: that the accuracy of the models has a significant difference, clearly showing the gradual increase in
140
H. T. Nguyen et al.
Fig. 5. Confusion matrix on each (test) fold of EfficientNetV2.
Fig. 6. The average performance on the training and the test sets in accuracy.
the efficiency of the old to new models. More specifically, the Fully Connected Network structure gives the lowest accuracy with ACC in two cases: the data set does not use Data Augmentation and uses Data Augmentation with results equal to 0.332 and 0.351, respectively, followed by Convolutional Neural Networks reaching ACC equals 0.567, 0.576. Finally, the latest model achieves the highest accuracy - EfficientNetV2 with ACC equal to 0.758, 0.77. This shows that modern and complex structures are gradually improving the accuracy of machine learning models and the problem of ultrasound image classification. Next comes the accuracy improvement with the effect of the Data Augmentation method. In all cases, including training and testing, on all three models, it is not difficult to see that the model’s accuracy using the data set applied Data Augmentation technique is higher than that of the model using the original data set. Although the improvement is not too significant, it is a remarkable success term of this problem’s relatively small data set. Furthermore, it improved accuracy in most cases and proves that the Data Augmentation method has not lost the main features of the image and is effective in building the ultrasound image classification method. EfficientNetV2 model combines Data augmentation technique for the highest testing results with ACC = 0.77, AUC = 0.804 and MCC = 0.619. To ensure the model’s accuracy, let us look at the Confusion matrix of each Fold of this method. The result below is the Confusion matrix of each Fold in the model with the highest results (EfficientNetV2 + Data augmentation) as
Breast Ultrasound Image Classification Using EfficientNetV2
141
exhibited in Fig. 5. The above results show that the prediction model is accurate in all classes. There is no “rote learning” situation. However, in the Malignant line, the cells are quite dark, equivalent to many cases of confusion, especially confusion with the Normal class. A part of the ultrasound image belonging to the Malignant class may have similar extraction characteristics leading to this phenomenon. However, in general, we can see that the model has done a good job of classification, as exhibited in Fig. 6.
5
Conclusion
This study investigated and compared the efficiency of an emerging CNN architecture, EfficientNetV2, for classification tasks on breast ultrasound images to diagnose breast cancer. Our model has achieved pretty high accuracy in many different measures, proving that the model is effective in classification. We hope that our research will help study the classification of breast cancer in particular and other diseases in general as a premise for developing automatic diagnostic systems to improve the quality of medical examination and treatment. And disease treatment. In the future, we will increase the size of the dataset, increase the input image size if there is no hardware limitation, and apply more advanced models to improve the performance of the analysis model breast cancer by ultrasound.
References 1. Tan, M., Le, Q.V.: Efficientnetv2: smaller models and faster training (2021). https://arxiv.org/abs/2104.00298 2. et al., O.R.: ImageNet large scale visual recognition challenge. Int. J. Comput. Vis. (IJCV) 115(3), 211–252 (2015) 3. et al., T.I.: Low-frequency ultrasound-mediated cytokine transfection enhances t cell recruitment at local and distant tumor sites. Proc. Natl. Acad. Sci. 117(23), 12674–12685 (2020). https://doi.org/10.1073/pnas.1914906117 4. Fujioka, T., et al.: Distinction between benign and malignant breast masses at breast ultrasound using deep learning method with convolutional neural network. Japan. J. Radiol. 37(6), 466–472 (2019). https://doi.org/10.1007/s11604019-00831-5 5. Arnaout, R., Curran, L., Zhao, Y., Levine, J.C., Chinn, E., Moon-Grady, A.J.: An ensemble of neural networks provides expert-level prenatal detection of complex congenital heart disease. Nat. Med. 27(5), 882–891 (2021). https://doi.org/ 10.1038/s41591-021-01342-5 6. Zhu, Y.C., Jin, P.F., Bao, J., Jiang, Q., Wang, X.: Thyroid ultrasound image classification using a convolutional neural network. Ann. Transl. Med. 9(20), 1526– 1526 (2021). https://doi.org/10.21037/atm-21-4328 7. Sudharson, S., Kokil, P.: An ensemble of deep neural networks for kidney ultrasound image classification. Comput. Methods Programs Biomed. 197, 105709 (2020). https://doi.org/10.1016/j.cmpb.2020.105709
142
H. T. Nguyen et al.
8. Saha, S., Sheikh, N.: Ultrasound image classification using ACGAN with small training dataset. In: Bhattacharyya, S., Mrˇsi´c, L., Brkljaˇci´c, M., Varghese Kureethara, J., Koeppen, M. (eds.) ISSIP 2020. AISC, vol. 1333, pp. 85–93. Springer, Singapore (2021). https://doi.org/10.1007/978-981-33-6966-5 9 9. Chen, W., et al.: A new classification method in ultrasound images of benign and malignant thyroid nodules based on transfer learning and deep convolutional neural network. In: Complexity 2021, pp. 1–9, September 2021. https://doi.org/10.1155/ 2021/6296811 10. Mikolajczyk, A., Grochowski, M.: Data augmentation for improving deep learning in image classification problem. In: 2018 International Interdisciplinary Ph.D. Workshop (IIPhDW). IEEE, May 2018. https://doi.org/10.1109/iiphdw.2018. 8388338 11. Chen, L.C., Papandreou, G., Kokkinos, I., Murphy, K., Yuille, A.L.: Semantic image segmentation with deep convolutional nets and fully connected CRFs (2014). https://arxiv.org/abs/1412.7062 12. LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. Nature 521(7553), 436–444 (2015). https://doi.org/10.1038/nature14539 13. Ding, C., Li, Y., Xia, Y., Wei, W., Zhang, L., Zhang, Y.: Convolutional neural networks based hyperspectral image classification method with adaptive kernels. Remote Sens. 9(6), 618 (2017). DOIhttps://doi.org/10.3390/rs9060618 14. Sultana, F., Sufian, A., Dutta, P.: Advancements in image classification using convolutional neural network. In: 2018 Fourth International Conference on Research in Computational Intelligence and Communication Networks (ICRCICN). IEEE, November 2018. https://doi.org/10.1109/icrcicn.2018.8718718 15. Kingma, D.P., Ba, J.: Adam: a method for stochastic optimization (2014). https:// arxiv.org/abs/1412.6980 16. Tan, M., Le, Q.V.: EfficientNet: rethinking model scaling for convolutional neural networks (2019). https://arxiv.org/abs/1905.11946 17. Al-Dhabyani, W., Gomaa, M., Khaled, H., Fahmy, A.: Dataset of breast ultrasound images. Data in Brief 28, 104863 (2020). https://doi.org/10.1016/j.dib.2019. 104863
Steganographic Approaches for Carrier Related Information Hiding Marek R. Ogiela(B) and Urszula Ogiela AGH University of Science and Technology, 30 Mickiewicza Ave., 30-059 Kraków, Poland {ogiela,mogiela}@agh.edu.pl
Abstract. In this paper we’ll define new steganographic techniques for information hiding, which depend on the carrier. In traditional steganographic solutions the way of secret hiding is usually not dependent on the content or features of container, in which new information is placed. Most approaches modify parameters in spatial or frequency domains, and not consider local carrier parameters. Such parameters can also determine the way of secret hiding. Such techniques allow to hide particular information in different manner, especially when we consider containers with different graphical features.
1 Introduction Steganographic techniques are widely used in advanced IT systems for secret data hiding and communication [1, 2]. Hidden data can have various forms, and can be transferred in many different containers having the form of images, audio files or communication protocols. One of the most important aspect of digital steganographic techniques is possibilities of hiding several different secret data in one container [3]. Such secrets can be placed in container in fully independent ways or can be linked using blockchain technologies. In this paper we’ll define another interesting problem connected with different storage possibilities of the same secret information, in different ways depending on the local features of container or carrier. This means that the same secret data should be hidden over the whole container depending on the actual content of the container [4, 5]. Having different containers, the same secret data can be hidden in different ways considering local graphical features of source images. In following section will be presented different heuristics, which allow to implement such techniques in real applications [6].
2 Carrier Related Steganography In this section will be described steganographic approaches, which allow to hide secret data in visual containers [7]. Hidden data will be embedded into carrier images in the way, which will be dependent from local pixel features, and nearest neighbor configuration.
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 143–146, 2022. https://doi.org/10.1007/978-3-031-08812-4_14
144
M. R. Ogiela and U. Ogiela
2.1 Pixels MSB Related Method The first heuristic techniques for placing secret data into visual carrier can be dependent on the RGB pixel values. In particular to achieve different secret distribution for different pixels values, we can consider values of the most significant bits (MSB) in each RGB component. Counting only the most significant bits, and creating from it 3-bits values, depending from the obtained results we can place a single bit of secret information in the least significant bits. The whole procedure can work as follows: • For each pixel in container convert the largest RGB bits into 3-bits integers and reduce it mod 4, i.e. Result: = MSB(R)MSB(G)MSB(B) mod 4; • Depending on the obtained results (i.e. values 0, 1, 2, 3), we place the single bit of secret information in the least significant bit (LSB) of the selected pixel component according to the values: If Result = 0 → place secret bit in R component If Result = 1 → place secret bit in G component If Result = 2 → place secret bit in B component If Result = 3 → we omit this point from placing secret information Having a visual container with embedded secret data, it is possible to restore the secret in the same manner i.e. finding the sum of largest RGB bits, and reduced mod 4. Depending on the obtained result we can decode single secret bits from particular RGB components or simply omit particular pixel from consideration. 2.2 Pixels Indexes Method The second steganographic approach of hiding secret information in visual container in the way fully dependent on image features, can be defined in such manner, in which we additionally consider vertical and horizontal indexes (row and column numbers) of carrier pixels. In this method we can use previous approach described in Sect. 2.1, and additionally add pixel indexes, then reduce the obtained result mod 4. Depending on the obtained results it will be possible to hide a single bit of secret information in least significant bits of pixels. The whole procedure can work as follows: For each pixel in container convert the largest RGB bits into 3-bits integers and add to this value the sum of the row (i) and column (j) numbers for particular pixel, then reduce it mod 4, i.e. Result: = (MSB(R)MSB(G)MSB(B) + i + j) mod 4; Depending on the obtained results (i.e. values 0, 1, 2, 3), it is possible to place the single bit of secret information in the least significant bit of the selected pixel component according to the values: If Result = 0 → place secret bit in R component If Result = 1 → place secret bit in G component If Result = 2 → place secret bit in B component If Result = 3 → we omit this point from placing secret information
Steganographic Approaches for Carrier Related Information Hiding
145
In this approach the secret data distribution will be dependent not only from local image features, but also from indexes of container points. It is important extension of the method described in Sect. 2.1. 2.3 Structural Elements Method The last approach for carrier dependent steganography can be based on using specially define structural elements defining pixels neighborhood. For this purpose, we can define structural elements having different sizes (e.g. 3x3, 5x5 mask), which allow to consider direct or indirect neighbor pixels of a given carrier pixel. Then we can calculate some values from this neighborhood (full colors or only particular RGB components) in order to decide whether the carrier pixel will be modified or not by placing in them secret bits of information. This is general approach, in which for secret revealing will be necessary to know how was defined structural element and embedding condition should also be defined.
3 Conclusions In this paper we’ve described possible approaches for hiding secret information in steganographic container in the ways, which are fully dependent on container features. Creating such algorithms, we can define methods, which allow to hide the same information in different ways depending on local features or carrier’s parameters. The same secret will be hidden, and splited over container in different ways for different carriers. Such methods increase the security levels of information hiding, and increase complexity of stegaanalysis approaches compared to the static methods of information hiding. Presented approaches can be extended towards considering additional personal features for secret hiding like biometric patterns, which before secret hiding allow it to share in user dependent manner [8, 9]. Acknowledgments. This work has been supported by the AGH University of Science and Technology research Grant No 16.16.120.773. This work has been supported by the National Science Centre, Poland, under project number DEC-2016/23/B/HS4/00616.
References 1. Pizzolante, R., Carpentieri, B., Castiglione, A., Castiglione, A., Palmieri, F.: Text compression and encryption through smart devices for mobile communication. In: Seventh International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, pp. 672–677 (2013) https://doi.org/10.1109/IMIS.2013.121 2. Koptyra, K., Ogiela, M.R.: Steganography in IoT: information hiding with APDS-9960 proximity and gestures sensor. Sensors 22(7), 2612 (2022). 1–11 https://doi.org/10.3390/s22 072612 3. Ogiela, M.R., Koptyra, K.: Visual pattern embedding in multi-secret image steganography. In: ICIIBMS 2015 - International Conference on Intelligent Informatics and BioMedical Sciences, November 28–30, 2015, Okinawa, Japan, pp. 434–437. IEEE (2015). ISBN 978-1-4799-85623/15
146
M. R. Ogiela and U. Ogiela
4. Ogiela, L., Ogiela, M.R.: Cognitive security paradigm for cloud computing applications. Concurr. Comput.: Pract. Exp. 32(8), e5316 (2020). https://doi.org/10.1002/cpe.5316 5. Ogiela, L.: Transformative computing in advanced data analysis processes in the cloud. Inf. Process. Manag. 57(5), 102260 (2020) 6. Ogiela, M.R., Ogiela, L., Ogiela, U.: Biometric methods for advanced strategic data sharing protocols. In: 2015 9th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing IMIS 2015, pp. 179–183 (2015). https://doi.org/10.1109/IMIS.201 5.29 7. Ogiela, U., Ogiela, L.: Linguistic techniques for cryptographic data sharing algorithms. Concurr. Comput. Pract. Exp. 30(3), e4275 (2018). https://doi.org/10.1002/cpe.4275 8. Ogiela, M.R., Ogiela, U.: Secure information splitting using grammar schemes. New Challenges in Computational Collective Intelligence. SCI, vol. 244, pp. 327–336. Springer, Heidelberg (2009). https://doi.org/10.1007/978-3-642-03958-4_28 9. Ogiela, M.R., Ogiela, U., Ogiela, L.: Secure information sharing using personal biometric characteristics. In: Kim, T.-H., Kang, J.-J., Grosky, W.I., Arslan, T., Pissinou, N. (eds.) FGIT 2012. CCIS, vol. 353, pp. 369–373. Springer, Heidelberg (2012). https://doi.org/10.1007/9783-642-35521-9_54
A Bi-objective Genetic Algorithm for Wireless Sensor Network Optimization 1 Amit Dua1 , Pavel Kr¨ omer2(B) , Zbigniew J. Czech1 , and Tomasz Jastrzab 1
Department of Algorithmics and Software, Silesian University of Technology, ul. Akademicka 16, 44-100 Gliwice, Poland [email protected], {zbigniew.czech,tomasz.jastrzab}@polsl.pl 2 Department of Computer Science, VSB-Technical University of Ostrava, 17. listopadu 15, 70800 Ostrava, Czech Republic [email protected]
Abstract. When designing a wireless sensor network several performance metrics should be considered, e.g., network lifetime, target coverage, sensor energy consumption. As a rule, these metrics are in conflict with each other, which means that by optimizing some of them we worsen the others. Designing the network is therefore a problem of multiobjective optimization. In this work, we propose a bi-objective genetic algorithm that optimizes network lifetime and target coverage. We consider two variants of the algorithm, in which the fitness function comprises only the network lifetime, or where it includes both, the network lifetime and target coverage. This makes it possible to find a trade-off between these two objectives. In-depth experimental studies are carried out for both variants of the algorithm. Keywords: Wireless sensor networks optimization · Genetic algorithm
1
· Coverage problem · Lifetime
Introduction
Wireless sensor networks (WSN) are of growing interest due to the large number of their applications. They can be used in various sectors such as industry, agriculture, etc. A WSN is a set of wireless devices, called sensor nodes or sensors, distributed over a certain area where they work together to accomplish given task. The WSN monitors the area, collects and processes data, and sends them to the base station. In some cases, access to a deployed WSN can be limited or impossible, e.g., in deserts, mountains, or other environments in which sensor batteries cannot be recharged or replaced. Since the availability of sufficient energy is essential for long-term WSN operation, wise planning of sensor activity is an important design issue. Sensors can be deployed in deterministic or random manner. The first approach makes it possible to place sensor nodes at predetermined positions to meet c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 147–159, 2022. https://doi.org/10.1007/978-3-031-08812-4_15
148
A. Dua et al.
the requirements such as coverage or connectivity. Deterministic deployment is suitable for controllable environments like factories, hospitals, office buildings. In hard-to-reach environments, sensor nodes can be dropped from an aircraft that leads to random placement. Generally, random deployment uses more nodes compared to deterministic deployment, and needs more complicated sensor activation schedules. One of a WSN performance metrics is its lifetime understood as the time in which the network is able to carry out the tasks ensuring the required quality of service. In monitoring applications the quality of service can be associated with the target coverage requirement. A method to extend the sensor network life is to organize the sensor nodes into a maximal number of disjoint cover sets that are activated one by one. Only sensors from the current active set are monitoring the targets ensuring the desired coverage rate, while the other sensor nodes are in a low-power sleep state. The sequence of active cover sets corresponds to network lifetime. Another performance metric of a WSN is the number of targets in a cover set that exceeds the assumed minimum coverage. Maximization of both objectives— network lifetime and the number of targets above the minimum—leads to the bi-objective optimization problem. Clearly, these objectives are in conflict with each other. This paper presents a genetic algorithm to solve this optimization problem. The rest of this paper is organized as follows: Sect. 2 introduces the problem under consideration and Sect. 3 gives a short overview of the related work. The proposed approach is detailed in Sect. 4. Experimental evaluation on several WSNs is shown in Sect. 5. Finally, conclusions and future work are described in Sect. 6.
2
Problem Formulation
The monitoring of a target environment by a WSN with randomly deployed sensor nodes is in this work mathematically modelled using the following formulations. Assume a rectangular target environment (field), F , with the size W × H, on which a regular mesh with a step, g, has been stretched in the vertical and horizontal direction. The intersections of the horizontal and vertical lines of the mesh correspond to the so-called targets, z(x, y), that represent the phenomena that are supposed to be tracked by the network. The sensors of a WSN, S = {si = (xi , yi )}, i = {1, 2, . . . , n}, are randomly distributed inside the target environment and, together with F , form an instance of the problem. An example of a problem instance is shown in Fig. 1. All sensors, si ∈ S, are homogeneous with the same battery capacity, b. The operation of the WSN is done in uniform time steps (intervals). In each step, a node remains active (ON) or inactive (OFF). An active node performs the monitoring but depletes energy. An inactive node, in turn, saves energy but does not fulfill its task.
A Bi-objective Genetic Algorithm for Wireless Sensor Network Optimization
149
Fig. 1. A 100 × 100 field with targets (small circles) and sensors (crosses) placement. The large circles denote the tracking range of each sensor.
Let S (si , tj ) ∈ {ON, OFF} denote the state of the sensor, si , in the time step, tj , and S (tj ) = {si : S (si , tj ) = ON} be the set of all sensors active in the time step tj . We say that a sensor, si (xi , yi ), covers a target, z(x, y), if the Euclidean distance between the sensor and the target is less than the tracking range of the sensor, Rsi . Let p(si ) denote the set of targets covered by sensor si . Then the set of all targets covered by a WSN in the time step tj can be defined as: P(tj ) = ∪ni=1 (p(si ) : si ∈ S (tj )). The degree of coverage, c(tj ), in the time step, tj , is equal to the ratio of the number of targets covered by the active sensors, |P(tj )|, and the number of all targets, P : c(tj ) =
|P(tj )| . P
(1)
The minimum degree of coverage, q ∈ (0, 1], represents the smallest number of targets that must be covered by active sensors in every time step tj , i.e., c(tj ) ≥ q. Finally, the problem of WSN lifetime maximization is formulated as follows. Find a disjoint cover set, P = {S (t1 ), S (t2 ), . . . , S (tm )}, where c(tj ) ≥ q for each tj , j ∈ {1, 2, . . . , m}, and S (ti )∩S (tj ) = ∅ for each i, j ∈ {1, 2, . . . , m}, i = j, and the number of time steps, m, is maximized. The sequence of cover sets, S (t1 ), S (t2 ), . . . , S (tm ), is also called the sensor ON and OFF schedule. WSN lifetime maximization formulated in such a way is NP-hard [5].
3
Related Work
Energy efficiency is an important aspect of WSN design. Usually, WSN nodes operate on stringent energy budgets and their operations need to be as energyefficient as possible. In the case of battery-powered WSNs, the source of energy is limited and its re-charging (replacement) might be an issue. A number of solutions has been proposed to conserve battery power. A widely adopted approach is to schedule node activities by placing some of them into a low-power mode so that the main task of the WSN is still maintained. In the case of monitoring, the required level of coverage must be achieved. Such a solution benefits from
150
A. Dua et al.
the use of redundant sensor nodes to increase the likelihood that the coverage requirements are kept [4,25]. The main types of coverage tasks in monitoring WSN applications are area coverage and point coverage. The goal of area coverage is to monitor an area under consideration, while point coverage aims at covering a required set of target points at every moment of network operation. When the number of sensors deployed in the area is larger than the smallest number of sensors needed to cover the area (points), they can be divided into cover sets that independently cover the area. The cover sets can be disjoint or non-disjoint. These sets are enabled (activated) one by one so that when one of them is active, all other nodes are placed in the idle mode to conserve the network’s energy. The objective of network scheduling is then to maximize the number of cover sets since their number directly corresponds to network lifetime. 3.1
Disjoint Cover Sets
The lifetime of a WSN with homogeneous sensor nodes cannot exceed the battery capacity of any single node if all of them are constantly active. If the network contains redundant nodes, which is not uncommon in randomly deployed WSNs, the sensor nodes can be divided into cover sets. If one sensor node is present in only one cover set, we talk about disjoint cover sets. The concept of disjoint cover sets is illustrated in Fig. 2. The figure shows four sensors, S1, S2, S3, S4, and four targets, T1, T2, T3, T4. The goal of the network is to cover all the targets for as long as possible. In the example, the sensors can be divided into two cover sets {S1, S3} and {S2, S4}, both of which satisfy the required coverage of all four targets. Apparently, the network lifetime can be extended by a factor of two if only the sensors in a single cover set are active at a time instead of all sensors of the network. S1
S2 T1
T2
S3 T3
S4 T4
Fig. 2. Disjoint cover sets example
The approach of using disjoint cover sets to tackle the network lifetime optimization problem has been an important area of WSN research. An exact algorithm to the maximum disjoint cover set problem proposed in [6] uses a mixed integer programming algorithm that produces optimal cover sets but requires high computational time to find the solution as it relies on exhaustive search. The cover set generation process in [6] gives priority to nodes having higher coverage which results in early depletion of their energy leading to reduced coverage at a later time. Another method for generating the covers considers giving
A Bi-objective Genetic Algorithm for Wireless Sensor Network Optimization
151
priority to sensor nodes with higher residual battery level [14]. However, it still results in the depletion of their energy at a higher rate and optimal network lifetime might not be achieved. Other methods based on disjoint cover sets are discussed, e.g., in [1–3]. Among the metaheuristic methods for monitoring network lifetime optimization, genetic algorithms (GA) have been used with success [12,13]. Manju et al. [13] introduced a genetic algorithm to enhance the network lifetime. In finding the maximum disjoint cover sets, the sensor nodes having more residual energy levels and also covering the sparsely covered targets were given more priority. A GA-based model to extend the network lifetime in applications when a target needs to be covered by more than one sensor (k-coverage) was proposed in [8]. The model took into account parameters such as targets’ positions, expected consumed energy, and sensors’ tracking range. Another efficient method for disjoint cover sets maximization was based on a distributed GA [23], where a two-level fitness function was evaluated in parallel by a set of processors. Genetic algorithms were also proposed for the optimization of the area coverage in sensor deployment by sensor placement [21,22]. For a WSN with heterogeneous sensors’ tracking range, the problem of maximizing area coverage was solved by a modified improved genetic algorithm which combined the use of Laplace and arithmetic crossover operators [10]. Lion optimization is another fast-converging approach for getting optimal coverage [20]. Complex surveys on the use of multi-objective optimization for WSNs were presented in [9,16]. Among others, they demonstrate that the coverage versus lifetime trade-off is a common issue that needs to be addressed by WSN optimization. In an attempt to fulfill the coverage requirement, the lifetime of the network gets shorter and while fulfilling the lifetime objective, the coverage requirement is not optimized. A Pareto-based bi-objective approach to optimize the sensor placement positions involves the use of two objective functions and a search for an acceptable compromise between the objectives [17]. The Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) decomposes a multiobjective problem into scalar sub-problems and optimizes them simultaneously. Two hybrid algorithms, MOEA/D-I (a hybrid of genetic algorithm and differential evolution) and MOEA/D-II (a combination of MOEA/D-I and discrete particle swarm optimization), have been used to improve the network lifetime while ensuring coverage [24]. In this work, we take another step forward and design a bi-objective genetic algorithm that allows the joint optimization of both, network lifetime (the number of disjoint cover sets) and the average number of targets covered by each cover. The disjoint cover set problem, in this work, is formulated as a bi-objective optimization challenge and a straightforward fitness function allowing a userdefined prioritization of any of the objectives is defined. This way, the problem of lifetime versus coverage trade-off is resolved in a convenient and user-friendly manner.
152
4
A. Dua et al.
Bi-objective GA for WSN Optimization
In this work, we propose a GA for the lifetime optimization of a monitoring WSN satisfying a coverage requirement. The WSN is homogeneous, composed of randomly deployed uniform nodes. The energy costs, battery capacity and sensing range are equal for all sensors. Hence, no sensor is used more often than the others and the task becomes equivalent to the maximum disjoint cover set problem [3,4]. Here, a slightly different monitoring scenario is considered. It assumes no apriori knowledge about the solution space, e.g., sparsely covered targets and/or critical sensors. In contrast, all sensors are considered equally important and the fundamental criterion of correct operation is the number of covered targets. The minimum coverage is a hard constraint—the covers that violate it, are not included in the network lifetime. The number of covered targets that exceeds the required minimum is considered by the fitness function, too, and the problem is effectively cast as a bi-objective optimization task. This bi-objective evaluation strategy allows improving both, network lifetime and the mean rate of coverage (i.e., the number of targets covered), an approach useful especially for WSNs with randomly deployed sensors. The GA adopts the simple yet intuitive solution encoding used, e.g., in [12]. For a WSN with n sensors, the set of disjoint covers is represented by a chromosome, x = (x1 , x2 , . . . , xn ), of length n. Each gene, xi ∈ x, has a discrete value taken from the set {1, 2, . . . , n}. This value—a cover id—represents the index of the cover and can take any value from 1 to n depending upon the number of disjoint covers that can be formed. In other words, it indicates the membership of the corresponding sensor in a particular cover. This allows the use of the encoding under the worst case scenario, i.e., when every sensor forms a separate cover. All sensors with the same cover id are ON at the same time, all others are OFF. The GA is used as a general optimization method to iteratively shuffle the cover ids so that an optimized schedule is formed. For each sensor, the ids of covered targets are recorded which helps the GA to identify the feasible disjoint covers. An example of chromosome representation and also the crossover and mutation operations is shown in Fig. 3. For more information on GAs and their operations, the reader is referred to [7,11,15].
Fig. 3. Chromosome encoding (left), crossover and mutation operations. Sensors S1, S2, S8, and S10 form cover set 1 (orange), S4, S6, and S9 form cover set 2 (light blue), S3, S5, and S7 form cover set 3. Each set covers all targets (a line means that the target is within sensor’s range).
A Bi-objective Genetic Algorithm for Wireless Sensor Network Optimization
153
The proposed bi-objective GA uses network lifetime as the primary objective and the mean number of covered targets as the secondary objective. The former criterion corresponds to the number of sensor covers, while the latter introduces selection pressure towards covers monitoring as many extra targets as possible. The objectives are aggregated into a fitness function using the Fβ variant of the F-score measure [18,19]. The first term of the function is the normalized number of covers, c (2) Cn = , n where c is the number of covers and n is the total number of sensors. The maximum possible number of covers is n (i.e., every sensor constitutes a cover), so Cn ∈ [0, 1]. The second term is the normalized mean number of targets covered by a single set, T , (3) Tn = T where T is the mean number of covered targets and T is the total number of targets. Clearly, Tn ∈ [0, 1]. Both, Cn and Tn are to be maximized. The two terms are aggregated into a single fitness value using the Fβ score, Fβ = (1 + β 2 )
Cn Tn . β 2 Cn + Tn
(4)
The normalization of Cn and Tn guarantees that Fβ ∈ [0, 1]. The Fβ is a suitable aggregation function allowing a seamless expression of preferences towards any objective via the β parameter, showing the relative importance of the second term. It defines the optimization process and shifts the priorities between the objectives.
5
Experiments
A series of computational experiments was performed to assess the ability of the proposed algorithm to find disjoint cover sets. To evaluate the GA a collection of artificial monitoring network configurations was created. Each configuration (problem instance) consisted of a regular mesh of 121 targets and u ∈ {10, 20, 30} randomly deployed sensors. The battery capacity, b, was the same for each sensor. For simplicity, we used b = 1. For each u, 10 problem instances with different random sensor placements were created to emulate random deployment. A few examples of test instances are shown in Fig. 4. Because the sensor placement was randomized, the maximum possible coverage was not the same for all problem instances, even with the same u. Consequently, multiple target coverage levels were always considered. The GA variant used in all experiments was a steady-state GA with generation gap 0.02 (2 chromosomes out of 100 are replaced between the generations), a population of 100 candidate solutions, mutation probability pM = 0.2, crossover probability pC = 0.8, the maximum number of generations 1,000,000, and chromosome encoding described in Sect. 4. The GA was used as a single-objective
154
A. Dua et al.
Fig. 4. Sample problem instances with 10 (first row), 20 (second row), and 30 sensors
meta-heuristic with Cn (cf. Eq. (2)) as the fitness function and as a bi-objective method with Cn and Tn aggregated by the weighted harmonic mean, Fβ (cf. Eq. (4)). All experiments were repeated 32 times and the presented results were averaged over these 32 runs. The general ability of the GA to find disjoint cover sets was studied using the single-objective GA variant. For each problem instance, the GA was used to find sets covering at least 5, 10, 15, . . . targets. The experiments with each problem instance were interrupted when none of the 32 optimization runs found a cover. The results are shown in Table 1. The table shows for each family of problem instances, labeled S10, S20, and S30 (instances with 10, 20, and 30 sensors respectively), average fitness, network lifetime (the number of sensor covers), and mean cover size of initial (random) and optimized solutions. Column k, shows how many runs per problem instance, on average, discovered a valid problem solution (i.e., a solution with non-zero fitness). The last two columns illustrate the difference (ratio) of both values between the (averaged) initial and optimized solutions. The value greater than 1 indicates an improvement, while the value smaller than 1 indicates a deterioration. The table clearly demonstrates both, the ability of the GA to improve network lifetime and the peculiarity of the problem. The hardness of the problem is illustrated by the average properties of the initial solutions: for S10 instances, no random solution was found when the coverage of 25 targets was required and
A Bi-objective Genetic Algorithm for Wireless Sensor Network Optimization
155
Table 1. Single-objective optimization results for the initial (random) and optimized disjoint cover sets. The values are averaged over all runs with non-zero fitness. n – number of sensors, q × T – minimum coverage, Cn – fitness value, c – network lifetime, T – mean cover size, k – average number of solutions with non-zero fitness. n
q × T Initial Cn
10
c
Optimized T
k
Cn
c
32.0 0.49 4.90
Change ratio T
k
c
T
6.39
32.0 1.18 0.96
5
0.42 4.17
6.64
10 10
0.17 1.74
10.95 32.0 0.27 2.70
10.55 32.0 1.55 0.96
10 15
0.10 1.00
15.91 23.0 0.17 1.69
15.44 32.0 1.69 0.97
10 20
0.10 1.00
21.22 1.5
0.10 1.00
20.53 27.1 1.00 0.97
10 25
0
0
0
0.10 1.00
25.45 9.3
20
5
0.39 7.84
6.90
32.0 0.50 10.00 6.50
20 10
0.15 3.03
11.09 32.0 0.29 5.82
10.54 32.0 1.92 0.95
20 15
0.06 1.13
16.09 31.5 0.18 3.56
15.56 32.0 3.15 0.97
20 20
0.05 1.00
21.29 7.3
0.10 1.99
20.69 32.0 1.99 0.97
20 25
0.05 1.00
25.67 0.3
0.05 1.07
25.54 26.6 1.07 1.00
20 30
0
0
0
0
0.05 1.00
30.52 4.1
-
-
20 35
0
0
0
0
0.05 1.00
35.00 0.3
-
-
30
0.37 10.96 7.00
5
0
32.0 0.50 15.00 6.45
-
-
32.0 1.28 0.94
32.0 1.37 0.92
30 10
0.13 3.91
11.10 32.0 0.29 8.70
10.47 32.0 2.22 0.94
30 15
0.04 1.26
16.11 32.0 0.17 4.98
15.57 32.0 3.96 0.97
30 20
0.03 1.00
20.97 11.1 0.09 2.78
20.80 32.0 2.78 0.99
30 25
0.03 1.00
25.75 0.3
0.04 1.35
25.49 29.9 1.35 0.99
30 30
0
0
0
0
0.03 1.00
30.41 7.1
-
-
30 35
0
0
0
0
0.03 1.00
35.00 0.4
-
-
only a single random cover set was identified for 15 and 20 targets. For S20 and S30 instances, random cover sets failed to satisfy the coverage of 30 and more targets and, on average, only a single cover set providing the coverage of 20 and 25 targets was found. The results were largely improved by the use of the GA. The algorithm was able to find solutions with up to 3.96 times longer network lifetime (S30 with target coverage 15). Moreover, it has found cover sets satisfying the coverage requirement for 5 problem configurations for which all initial random solutions failed. The experiment clearly shows the usefulness of the GA for network lifetime optimization. However, it also confirms that when the sensors are placed randomly, they are able to cover only a small percentage (up to 28.92%) of the targets. Finally, it also illustrates the inverse relationship between the network lifetime and the mean cover size. For all problem configurations, the mean cover size decreased through the optimization. The bi-objective problem formulation problem was explored in a series of experiments with one instance of the problem with 30 randomly placed sensors (S30R0). The problem was solved for coverage levels 10, 15, and 20, selected based on the previous experiment. For each coverage level, 10 different values of the β parameter were considered: β ∈ (0.2, 0.4, . . . , 2.0) and the ability of the
156
A. Dua et al.
Table 2. Bi-objective optimization results for one instance with 30 sensors (S30R0) with different levels of target coverage and different values of β. q × T – minimum coverage, Fβ – fitness value, c – network lifetime, T – mean cover size. β
q × T = 10 Fβ c T
q × T = 15 Fβ c T
q × T = 20 Fβ c T
0.2 0.33 1.00 61.00 0.33 1.00 61.00 0.33 1.00 61.00 0.4 0.22 2.03 43.00 0.22 2.00 43.38 0.22 2.00 43.39 0.6 0.18 3.13 30.97 0.18 3.06 31.42 0.18 2.50 37.63 0.8 0.17 4.34 22.80 0.17 4.00 24.48 0.15 2.75 34.72 1.0 0.16 5.09 19.29 0.16 5.00 19.60 0.13 2.69 35.49 1.2 0.17 6.06 16.19 0.16 5.22 18.89 0.12 2.62 36.16 1.4 0.17 7.00 14.00 0.17 5.28 18.68 0.12 2.78 34.55 1.6 0.18 8.00 12.25 0.17 5.25 18.78 0.11 2.78 34.53 1.8 0.19 8.91 11.02 0.17 5.19 18.99 0.11 2.94 32.79 2.0 0.20 9.00 10.89 0.17 5.31 18.58 0.10 2.72 35.35
algorithm to find disjoint cover sets was investigated. The GA was repeated 32 times for each pair of target coverage level and value of β, and the average results were analyzed. The results of the experiment are summarized in Table 2. The table shows for each target coverage and value of β the average fitness (F-score) of the best-found solution and the average values of both objectives. The table clearly shows that weighted harmonic mean can be used to drive the optimization of both target objectives. Moreover, the results demonstrate that the parameter β can shift the priorities between the objectives. The lowest value of β gives preference to the mean cover size and the optimization results in a single cover with the maximum size. The same solution is in this case found for all target coverage levels. On the other hand, the largest value of β gives preference to cover count and results in solutions with mean cover size just above the required target coverage and with the largest cover count. This corresponds most closely to the single-objective formulation of the problem. The value of β between the two extremes, 0.2 and 2.0, allows fine-tuning of the optimization towards solutions with the desired ratio of network lifetime and mean cover size. This is shown in Table 2 and Fig. 5. The figure shows in each plot the change in average network lifetime and mean cover size between initial and optimized solutions. The plots very well illustrate the inverse relationship between network lifetime and mean cover size. They also show that in each case, certain values of β lead to problem solutions with balanced trade-off between the optimized factors. This confirms the usefulness of the Fβ -based fitness function for the customization of the optimization process.
A Bi-objective Genetic Algorithm for Wireless Sensor Network Optimization
157
Fig. 5. Changes of avg. network lifetime and avg. mean cover size for solutions of S30R0 evolved by bi-objective GA with different values of β, for the coverage requirements of 10 (top left), 15 (top right) and 20 targets. Green arrows mean that both objectives improved, red arrows otherwise.
6
Conclusions
In this work, a bi-objective GA was proposed and evaluated. The investigated method observes two objectives—the number of cover sets satisfying the coverage requirement (network lifetime), and the mean number of targets covered by each cover (cover size). Both objectives are combined by a flexible aggregation function, the Fβ score, that corresponds to the weighted harmonic mean of both terms. An extensive experimental evaluation showed that the proposed method can discover ON and OFF schedules that increase the network lifetime. The experiments also showed that the aggregation function parameter, β, can be effectively used to modify the properties of optimized schedules towards one of the objectives. This allows a seamless adjustment of the trade-off between the two criteria. The work will continue in two main directions. First, other metaheuristic optimization methods such as differential evolution and particle swarm optimization will be used for the evolution of ON and OFF schedules. Second, optimization of the schedules for WSNs with heterogeneous sensor nodes will be investigated. Acknowledgements. The authors would like to thank the following computing centres where the computation of the project was performed: Academic Computer Center in Gda´ nsk (TASK), and Wroclaw Centre for Networking and Supercomputing (WCSS). This work was also supported by the Ministry of Education, Youth and Sports of the
158
A. Dua et al.
Czech Republic in the project “Metaheuristics Framework for Multi-objective Combinatorial Optimization Problems (META MO-COP)”, reg. no. LTAIN19176, and in part by the SGS grants no. SP2022/11 and SP2022/77, VSB-TU Ostrava. Czech Republic.
References 1. Abdulhalim, M.F., Attea, B.A.: Multi-layer genetic algorithm for maximum disjoint reliable set covers problem in wireless sensor networks. Wirel. Pers. Commun. 80(1), 203–227 (2015) 2. Ahn, N., Park, S.: A new mathematical formulation and a heuristic for the maximum disjoint set covers problem to improve the lifetime of the wireless sensor network. Ad Hoc Sens. Wirel. Netw. 13(3–4), 209–225 (2011) ¨ 3. Attea, B.A., Khalil, E.A., Ozdemir, S., Yildiz, O.: A multi-objective disjoint set covers for reliable lifetime maximization of wireless sensor networks. Wirel. Pers. Commun. 81(2), 819–838 (2015) 4. Cardei, M., Du, D.: Improving wireless sensor network lifetime through power aware organization. Wirel. Netw. 11(3), 333–340 (2005) 5. Cardei, M., Thai, M., Li, Y., Wu, W.: Energy-efficient target coverage in wireless sensor networks. In: Proceedings of the 24th Annual Joint Conference of the IEEE Computer and Communications Societies, vol. 3, pp. 1976–1984 (2005) 6. Cardei, M., Wu, J.: Energy-efficient coverage problems in wireless ad-hoc sensor networks. Comput. Commun. 29(4), 413–420 (2006) 7. Das, A.K., Das, S., Ghosh, A.: Ensemble feature selection using bi-objective genetic algorithm. Knowl. Based Syst. 123, 116–127 (2017) 8. Elhoseny, M., Tharwat, A., Farouk, A., Hassanien, A.E.: K-coverage model based on genetic algorithm to extend WSN lifetime. IEEE Sens. Lett. 1(4), 1–4 (2017) 9. Fei, Z., Li, B., Yang, S., Xing, C., Chen, H., Hanzo, L.: A survey of multi-objective optimization in wireless sensor networks: metrics, algorithms and open problems. IEEE Comm. Surv. Tutor. 19 (2016) 10. Hanh, N.T., Binh, H.T.T., Hoai, N.X., Palaniswami, M.S.: An efficient genetic algorithm for maximizing area coverage in wireless sensor networks. Inf. Sci. 488, 58–75 (2019) 11. Holland, J.H.: Adaptation in Natural and Artificial Systems. University of Michigan Press, Ann Arbor (1975) 12. Lai, C.C., Ting, C.K., Ko, R.S.: An effective genetic algorithm to improve wireless sensor network lifetime for large-scale surveillance applications. In: 2007 IEEE Congress on Evolutionary Computation, pp. 3531–3538 (2007) 13. Manju, Chand, S., Kumar, B.: Genetic algorithm-based meta-heuristic for target coverage problem. IET Wirel. Sens. Syst. 8(4), 170–175 (2017) 14. Mini, S., Udgata, S., Sabat, S.: A heuristic to maximize network lifetime for target coverage problem in wireless sensor networks. Ad Hoc Sens. Wirel. Netw. 13(3–4), 251–269 (2011) 15. Mitchell, M.: An Introduction to Genetic Algorithms. MIT Press, Cambridge (1996) 16. Moshref, M., Al-Sayyed, R., Al-Sharaeh, S.: Multi-objective optimization algorithms for wireless sensor networks: a comprehensive survey. J. Theor. Appl. Inf. Technol. 98, 2839–2871 (2020) 17. Nong, S.X., Yang, D.H., Yi, T.H.: Pareto-based bi-objective optimization method of sensor placement in structural health monitoring. Buildings 11(11) (2021)
A Bi-objective Genetic Algorithm for Wireless Sensor Network Optimization
159
18. van Rijsbergen, C.J.: Information Retrieval, 2nd edn. Butterworths, London (1979) 19. Sammut, C., Webb, G.I.: Encyclopedia of Machine Learning, 1st edn. Springer, Boston (2011). https://doi.org/10.1007/978-0-387-30164-8 20. Singh, A., Sharma, S., Singh, J.: Nature-inspired algorithms for wireless sensor networks: a comprehensive survey. Comput. Sci. Rev. 39, 100,342 (2021) 21. Tarnaris, K., Preka, I., Kandris, D., Alexandridis, A.: Coverage and k-coverage optimization in wireless sensor networks using computational intelligence methods: a comparative study. Electronics 9(4) (2020) 22. Tossa, F., Abdou, W., Ezin, E.C., Gouton, P.: Improving coverage area in sensor deployment using genetic algorithm. In: Krzhizhanovskaya, V.V., et al. (eds.) ICCS 2020. LNCS, vol. 12141, pp. 398–408. Springer, Cham (2020). https://doi.org/10. 1007/978-3-030-50426-7 30 23. Wang, Z.J., Zhan, Z.H., Zhang, J.: Solving the energy efficient coverage problem in wireless sensor networks: a distributed genetic algorithm approach with hierarchical fitness evaluation. Energies 11(12) (2018) 24. Xu, Y., Ding, O., Qu, R., Li, K.: Hybrid multi-objective evolutionary algorithms based on decomposition for wireless sensor network coverage optimization. Appl. Soft Comput. 68, 268–282 (2018) ´ Dumitrescu, E.: Nodes self-scheduling approach for 25. Zairi, S., Zouari, B., Niel, E., maximising wireless sensor network lifetime based on remaining energy. IET Wirel. Sens. Syst. 2(1), 52–62 (2012)
104 Fruits Classification Using Transfer Learning and DenseNet201 Fine-Tuning Khanh Vo Hong(B) , Tin Tang Minh, Hoa Le Duc, Nam Truong Nhat, and Huong Luong Hoang Information Technology Department, FPT University, Can Tho, Vietnam {khanhvh,tintmce130438,hoaldce140469,namtnce140370}@fpt.edu.vn
Abstract. In the real life, when you see an apple, eggplant, or beet, how do you know the name of them in another language? In this article, we propose a new approach to identify vegetables and fruits by using deep learning and transfer learning based on DenseNet201 model. The result of this research is a model that help people to identify 104 types of vegetables and tubers. This model is trained on the original dataset and the processed dataset, the highest accuracy at fine-tuning is 94%. Keywords: DenseNet-201 · Fruit recognition Transfer learning · Image classification
1
· Deep learning ·
Introduction
The fruits and vegetables are very useful for our health, in the daily meals of families in Vietnam we rarely lack the face of green foods. They help people maintain a balanced condition, not be obese and other diseases such as diabetes, nearsightedness, etc. Red foods such as tomatoes, carrots, and beetroot provide plenty of vitamins. Dark green vegetables provide a lot of minerals and fiber, especially very low calories that help increase immunity and anti-aging. Identifying vegetables, tubers, and fruits based on color, shape, size has a lot of research papers. One of the applications of recognition research is learning vocabulary through objects, used to teach cooking robots to distinguish foods. The ability to identify high accuracy and the time to return results within an acceptable time are requirements for research. The challenge of current identification research is the lack of data for less common vegetables, tubers, and fruits, high accuracy for input image without a background but in reality, it is wrong to recognize. This research is trying to solve the problem of incorrect identification in reality and enhance accuracy. The research was divided into a structure of five sections. Section one is about the introduction of the paper. Section two explores the similarities between our research and previous research. Section three includes some materials and proposed methods to improve accuracy, solve the problem outlined above. Section four is about the results of research, evaluation, and comparison of the proposed model with other models. The final section is about the method and final result. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 160–170, 2022. https://doi.org/10.1007/978-3-031-08812-4_16
104 Fruits Classification Using Transfer Learning
2
161
Related Work
In recent years, there are many research to solve the fruit recognition problem. Researchers are constantly improving their accuracy in innovative ways. In this paper, they propose an efficient fruit and vegetables classification system using image saliency to draw the object regions and convolutional neural network and VGG model is chosen to train for fruit and vegetable classification [1]. There are 26 types of fruits and vegetables in their dataset. Classification from live video streams of [2] using image segmentation method, feature extraction mainly based on color and classification with Gaussian Bayes classifier. In this article, the author researched the similarity of three fruits include apple, pear and peach, thereby training the model to recognize and classify them [3]. The identification of fruits is also applied in retail stores to reduce the time for employees at checkout, in this article [4] the author identifies fruits in normal environment and covered by a layer of plastic bags. The fruit classification method uses six layers of convolutional neural networks as shown here [5]. They designed a six-layer CNN consisting of convolution layers, pooling layers and fully connected layers, after all, they achieve an accuracy of 91,44%. In addition, in agriculture, the researchers developed an independent system capable of classifying apples, oranges, pears and lemons presented in this paper [6]. The results after experimentation show that the Top 5 accuracy on the dataset used is 90% and the Top 1 accuracy is 85% which targets the accuracy limitation of previous attempts. The common difficulty that these researchers have is that it is difficult to distinguish fruits from each other because of similarities in shape, size, and color.
3 3.1
Materials and Proposed Methods CNN Architecture
Convolutional neural networks have one input layer, one output layer, and many hidden layers. These hidden layers are intertwined and fully-connected. The more complex the convolutions, the more time and computation are required during training. CNN includes many types of layers such as convolutional layers, pooling layers, and fully-connected layers. The CNN architecture is formed by the stacked arrangement of those layers. This is a simple CNN architecture used in MNIST classification as shown in Fig. 1 [7]. 3.2
DenseNet Architecture
DenseNet is a type of complex neural network that uses dense connections between layers, through dense blocks, where they directly connect all layers together. To maintain transition properties, each layer receives additional inputs from all previous layers and passes its own feature maps to all subsequent layers [9].
162
K. V. Hong et al.
Fig. 1. An simple CNN architecture
Fig. 2. Architecture of DenseNet
Each convolutional layer except the first one, receives the output of the previous convolutional layer and generates an output feature map, which is then passed to the next convolutional layer. For L classes, there is L directly connected to the middle of each layer and its next layer [9]. The DenseNet architecture is all about modifying the standard CNN architecture as as shown in Fig. 2. In the DenseNet architecture, each layer is connected to another layer, so it is called Densely Connected Convolutional NetWork. For L layers, there are L(L + 1)/2 direct connections. For each layer, the feature maps of all previous layers are used as input and its feature map is used as input for each subsequent layer. DenseNet associates every layer with every other layer. This is the main idea that is incredibly powerful. The input to a layer inside DenseNet is a concatenation with feature maps from previous layers [9]. 3.3
Transfer Learning
Transfer learning is a method by which recognition models take advantage of the power of previous pre-train models to increase accuracy, reduce training time, and require less data through knowledge transfer. With these advantages, the transfer learning method is gradually becoming popular for real problems. The figure below is a simulation for using the transfer learning method to re-utilize the knowledge, solve and optimize the recognition problems.
104 Fruits Classification Using Transfer Learning
163
Fig. 3. Simulation of transfer learning
3.4
Data Gathering
Fruits-262 [10] dataset (version 7) is a large data set and open benchmark on images of fruits, vegetables, and tubers and was published by Mihai Minut on Kaggle. This dataset contains a total number of images of 225,640 images of fruits, vegetables, and tubers including 262 classes with original sizes and many sets with resized. Based on this, the above dataset has selected 104 classes of fruits used in this research to objectively evaluate and verify the performance of our proposed model, creating a premise for future researchers. Our dataset has a total of 138,006 images, of which 94 classes are over 1000 images, 10 classes are under 1000 images, but the smallest number is 660 images per class and the highest is 1600 images per class. 3.5
Data Preprocessing
Some of the original images are quite large, up to tens of MB, together with the relatively large number will make storage become more memory intensive, especially making training time-consuming. Therefore, reduce the image quality to 95% but still ensure the quality for training. In addition, each type of pre-trained model has different input sizes, so we have resized it to best fit each model. Data augmentation will increase the number of images significantly through the methods of zooming, flipping, rotating, etc. 3.6
Split Dataset
This research divided the dataset into three sets: training set, validation set, and testing set. The training set is used to train the model, the validation set is used to verify the model during learning, and the testing set is used to verify the model after training has been completed.
164
K. V. Hong et al.
Fig. 4. Split dataset method
3.7
Proposed Methods - Recognition Model
This research uses the method of training and comparing different models. Train models with original data and compare with each other, train models with processed data, and continue to compare, thereby choosing the model with the highest accuracy. To identify 104 different types of fruits, vegetables and tubers, in this research we propose to use the DenseNet201 network model. Information about the overview model architecture and parameters of the DenseNet201 model will be presented in Fig. 6.
Fig. 5. Method of training and comparing different models
Fig. 6. DenseNet201 network model
In order to obtain the best results during training, this research used methods such as transfer learning and correction for the training model implementation, illustrated in Fig. 7. At transfer learning phase, we took advantage of the previously trained DenseNet201 model. This model has been trained on the ImageNet dataset with over 1 million images and 1000 classes such as keyboard, mouse, pencil, animals, etc. With this network, the input size is 224 × 224, in order to apply to the identification of fruits, vegetables, and tubers, the DenseNet201 model will be removed from the last layer and replaced by a softmax classifier function for 104 classes corresponding to 104 types of fruits, vegetables, and tubers.
104 Fruits Classification Using Transfer Learning
165
At this fine-tuning stage, the hyperparameters of the model are fine-tuned with the aim of achieving the highest accuracy. The parameters used in this process are as follows: (a) Epoch: to find the appropriate threshold we tested from 50–400; (b) Batch sizes tested are: 8, 16, 32, 64, 128, 256; (c) Hidden layer tested 1024–64; (d) The tested learning rate is: 1e−4, 1e−7, 1e−9; (e) Input size: 224 × 224 × 3, 512 × 512 × 3; (f) Optimizer: Adam; (g) Learning rate scheduler: Learning rate * 0.2. In addition, during the training and validation process, we also use some image enhancement methods with the goal of increasing the number of images in the input data set for the model. The values are set as follows: Random Flip (“horizontal”), Random Rotation (0.2), Random Height (0.2), Random Width (0.2), Random Zoom (0.2).
Fig. 7. Transfer learning model with DenseNet201 network for vegetable, tuber, and fruit recognition.
3.8
Training Model
To train the model according to the proposed method, we used two environments The first is Google Colab with Python version 3.7.12, Tesla K80 GPU, Intel(R) Xeon(R) CPU @ 2.30 GHz, 13 GB RAM, Storage 78 GB. Second environment is Kaggle with Python 3.7.12, GPU Tesla P100-PCIE-16 GB, Intel(R) Xeon(R) CPU @ 2.00 GHz, 16 GB RAM, Storage 73 GB.
4
Result
In the evaluation of the model, our research divided the data with a ratio of 7015-15. We evaluate the performance of the models based on evaluation criteria such as accuracy(acc), test accuracy(test acc), loss, validation (val), F1 score to analyze and evaluate the models DenseNet201, MobileNet, and EfficientNetB0.
166
4.1
K. V. Hong et al.
Experiment 1: Identification of the Original Data Set
In the CNN architecture model trained with a learning rate of 1e−4, using softmax activation in the full-connected layer along with hyperparameters: batch size is 32, hidden units are 1920, hidden layer is the number of layers of DenseNet201 plus 5 custom layers, the maximum number of training times is 350 epochs. In it, the values of transfer learning and fine-tuning have been mentioned in the previous section. Table 1 shows the best experimental results on the DenseNet201 model. Table 1. Comparison of the results of experiment 1 Stage
Model
Original data Acc Loss Val
Test acc F1
Transfer learning DenseNet201 0.80 0.78 0.80 0.78 MobileNet 0.70 1.15 0.72 0.73 EfficientNetB0 0.80 0.75 0.70 0.66
0.80 0.73 0.70
Fine-tuning
0.90 0.83 0.90
DenseNet201 0.92 0.23 0.90 0.89 MobileNet 0.91 0.61 0.82 0.83 EfficientNetB0 0.92 0.57 0.90 0.87
Fig. 8. Comparison of the results of experiment 1
104 Fruits Classification Using Transfer Learning
167
Fig. 9. Confusion matrix on original data, view confusion matrix on github.
4.2
Experiment 2: Identification of the Processed Data Set
In the CNN architecture model trained with a learning rate of 1e−4, using softmax activation in the full-connected layer along with hyperparameters: batch size is 32, hidden units are 1920, hidden layer is the number of layers of DenseNet201 and five custom layers, the maximum number of training times is 350 epochs. In it, the values of transfer learning and fine-tuning have been mentioned in the previous section. Table 2 shows the best experimental results on the DenseNet201 model. Table 2. Comparison of the results of experiment 2 Stage
Model
Transfer learning DenseNet201 MobileNet EfficientNetB0 Fine-tuning DenseNet201 MobileNet EfficientNetB0
Original data Acc Loss Val
Test acc F1
0.78 0.70 0.77 0.94 0.93 0.93
0.79 0.73 0.67 0.94 0.84 0.87
0.77 1.14 0.79 0.21 0.2 0.21
0.78 0.73 0.71 0.93 0.83 0.88
0.79 0.73 0.67 0.95 0.84 0.88
168
K. V. Hong et al.
Fig. 10. Comparison of the results of experiment 2
Fig. 11. Confusion matrix on processed data, view confusion matrix on github.
4.3
Evaluation of Test Results
Through the results of experiment 1 and experiment 2, we can see the model improvement when using original data and processed data based on transfer learning and fine-tuning methods. The best results obtained at the fine-tuning stage of the two experiments were compared with the following criteria: evaluate with accuracy criteria, DenseNet201 in experiment 1 (92%) was lower than DenseNet201 in experiment
104 Fruits Classification Using Transfer Learning
169
2 (94%); with loss criterion, DenseNet201 in experiment 1 (0.23) is higher than DenseNet201 in experiment 2 (0.21); with validation criteria, DenseNet201 in experiment 1 (0.88) is lower than DenseNet201 in experiment 2 (0.93); with test accuracy criteria, DenseNet201 in experiment 1 (89%) is lower than DenseNet201 in experiment 2 (94%); with the F1 score criterion, DenseNet201 in experiment 1 (90%) was lower than DenseNet201 in experiment 2 (95%). In experiment 2, the accuracy and test accuracy results are higher than in experiment 1 because we have performed a variety of image preprocessing, applying data augmentation techniques with the purpose of increasing the number of images, and highlighting the features in each image. Besides, we also change different hyperparameters to get the model with the best results. Besides, the results from experiment 2 bring relatively better results than the research presented in [10] with top 5 predict accuracy is 58%, research [11] with accuracy of 85.8% of F1 score, research [12] with accuracy is 88.75%, research [13] with accuracy is 92%, research [14] with accuracy is 90.7% compared to our research is 94% for accuracy and F1 scores are 95%.
5
Conclusion
This paper explores the fine-tuning technique on the DenseNet201 architecture along with image preprocessing and image augmentation techniques. The recognition rate has shown a lot of improvement during experiment two. Among all the cases, the model achieves the highest training accuracy of 94% and the highest test accuracy of 94%.
References 1. Zeng, G.: Fruit and vegetables classification system using image saliency and convolutional neural network. In: 2017 IEEE 3rd Information Technology and Mechatronics Engineering Conference (ITOEC), pp. 613–617 (2017). https://doi.org/10. 1109/ITOEC.2017.8122370. 2. Danev, L.: Fruit and vegetable classification from live video, p. 63 (2017) 3. Wu, L., Zhang, H., Chen, R., Yi, J.: Fruit classification using convolutional neural network via adjust parameter and data enhancement. In: 2020 12th International Conference on Advanced Computational Intelligence (ICACI), pp. 294–301, August 2020. https://doi.org/10.1109/ICACI49185.2020.9177518. 4. Pawel, K.: A vision-based method utilizing deep convolutional neural networks for fruit variety classification in uncertainty conditions of retail sales. Appl. Sci. 9(19), 3971 (2019). https://doi.org/10.3390/app9193971 5. Lu, S., Lu, Z., Aok, S., Graham, L.: Fruit classification based on six layer convolutional neural network. In: 2018 IEEE 23rd International Conference on Digital Signal Processing (DSP), pp. 1–5, October 2018. https://doi.org/10.1109/ICDSP. 2018.8631562. 6. Pande, A., Munot, M., Sreeemathy, R., Bakare, R.V.: An efficient approach to fruit classification and grading using deep convolutional neural network. In: 2019 IEEE 5th International Conference for Convergence in Technology (I2CT), pp. 1–7, March 2019. https://doi.org/10.1109/I2CT45611.2019.9033957.
170
K. V. Hong et al.
7. O’Shea, K., Nash, R.: An Introduction to Convolutional Neural Networks. In: 2017 IEEE 3rd Information Technology and Mechatronics Engineering Conference (ITOEC), p. 12 (2017) 8. Darwish, M.: Fruit Classification Using Convolutional Neural Network, p. 67 (2020) 9. Huang, G., Liu, Z., van der Maaten, L., Weinberger, K.Q.: Densely Connected Convolutional Networks ArXiv160806993 Cs, January 2018. arXiv:1608.06993, p. 9. Accessed 22 Feb 2022 10. Minut, M.-D., Iftene, A.: Creating a dataset and models based on convolutional neural networks to improve fruit classification. In: 2021, 23rd International Symposium on Symbolic and Numeric Algorithms for Scientific Computing (SYNASC), Timisoara, Romania, pp. 155–162, December 2021. https://doi.org/ 10.1109/SYNASC54541.2021.00035. 11. Bargoti, S., Underwood, J.: Image segmentation for fruit detection and yield estimation in apple orchards. ArXiv161008120 Cs. arXiv:1610.08120, p. 27, October 2016 12. Bird, J.J., Barnes, C.M., Manso, L.J., Ek´ art, A., Faria, D.R.: Fruit quality and defect image classification with conditional GAN data augmentation. ArXiv210405647 Cs Eess. arXiv:2104.05647, p. 16, April 2021 13. Fan, S., et al.: On line detection of defective apples using computer vision system combined with deep learning methods. J. Food Eng. 286 (2020). https://doi.org/ 10.1016/j.jfoodeng.2020.110102 14. Liu, L., Li, Z., Lan, Y., Shi, Y., Cui, Y.: Design of a tomato classifier based on machine vision 2019. PLOS One 14(7), 16 (2019). https://doi.org/10.1371/journal. pone.0219803.
Transfer Learning with Fine-Tuning on MobileNet and GRAD-CAM for Bones Abnormalities Diagnosis Huong Hoang Luong1 , Lan Thu Thi Le1 , Hai Thanh Nguyen2(B) , Vinh Quoc Hua1 , Khang Vu Nguyen1 , Thinh Nguyen Phuc Bach1 , Tu Ngoc Anh Nguyen1 , and Hien Tran Quang Nguyen1 1
FPT University, Can Tho, Viet Nam 2 Can Tho University, Can Tho, Vietnam [email protected]
Abstract. Osteoarthritis is a common medical condition. Unfortunately, despite the support of X-ray imaging technology in diagnosis, the accuracy of diagnostic results still depends on human factors. Furthermore, when errors do occur, they are often detected late, leading to a waste of time, money, and even disability for the patient. This study has deployed and evaluated transfer learning techniques in abnormal and normal bone images classification on X-ray images collected from the dataset of MUsculoskeletal RAdiographs (MURA) with 17,367 images and then leveraged techniques for results explanations of learning algorithms such as Gradient-weighted Class Activation Mapping (GRADCAM) to provide visual highlighted interesting areas in the images which can be signals for anomalies in bones. The classification performance using MobileNet with techniques of hyper-parameters fine-tuning can reach an accuracy of 0.84 in abnormal and normal bone classification tasks on the wrist, humerus, and elbow. The work is expected to provide visual support for doctors in diagnosing and identifying bone anomalies on X-ray images based on leveraging advancements from Artificial Intelligence techniques for medical imaging analysis. Keywords: Transfer learning X-ray images
1
· Bone · Medical imaging analysis ·
Introduction
According to the World Health Organization (WHO) [1], the period from 2020 to 2030 is considered as “Decade of Osteoarthritis”. In addition, according to the statistics of the Vietnam Musculoskeletal Association in1 , Vietnam is one of the countries with the highest rate of people suffering from osteoarthritis in the 1
https://tuoitre.vn/song-tu-chu-hon-nho-cham-soc-xuong-khop-dung-cach20211010205952585.htm, accessed on 20 March 2022.
c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 171–179, 2022. https://doi.org/10.1007/978-3-031-08812-4_17
172
H. H. Luong et al.
world with over 30% of people over 35 years old and 60% of people over 65 years old suffering from this disease. The misdiagnosis of pathology after an X-ray is often due to human factors such as the omission or misjudgment of radiographs by experts due to high workload, work pressure, or inexperience in radiology in some special situations. This misdiagnosis often leads to severe consequences when skipping the golden time leads to delayed treatment, increasing costs and time of treatment [2]. In some cases, this can lead to disabling consequences for the patient. Nowadays, the problem of digitizing medical examination and treatment information, test results, and diagnostic imaging for management has become popular. In order to effectively exploit this amount of data in medical examination and treatment, there are many applications based on artificial intelligence (AI) that are researched and applied to assist doctors in medical examination and treatment, decrease errors caused by human factors, increase accuracy, reduce cost and time of diagnosis. This study applies a transfer learning model to identify bone and joint problems from X-ray images. First, the input image will be preprocessed and transfer learning from the pre-trained MobileNet model to identify bone diseases. Next is the calibration phase to find the suitable hyperparameters that give the best results for the model. In addition, we use Gradient-weighted Class Activation Mapping (GRAD-CAM) [3] to highlight the location of the diagnosed bone disease to help check the accuracy of this autodiagnostic suggestion. The article consists of 5 sections. Section 1 is an introduction and problem definition. Related studies will be presented in Sect. 2. Section 3 is the implementation methodology. The next (Sect. 4) is the experiment. And the last is the conclusion in Sect. 5.
2
Related Work
There have been several studies in the field of identifying bone diseases through X-ray images. Most of these studies used the MURA image dataset, which is one of the most prominent bone datasets freely available, i.e., MURA: Large Dataset for Abnormality Detection in Musculoskeletal Radiographs [4]. The work in [4] applied a 169-class baseline DenseNet model to detect and localize anomalies. This model achieved an AUROC of 0.929, with an operating point of 0.815 sensitivity and 0.887 specificities. The authors in [5] had deployed a Convolutional Neural Networks (CNN) architecture based on two criteria: good performance of the architecture with fast model training and the possibility of eliminating error propagation during the process of the learning sessions. This study gave good results with the Kappa score index that were achieved by different solutions belonging to the wrist (0.942), hand (0.862), and shoulder (0.735). The study in [6] focused on the detection of shoulder fractures on radiographs. This study develops two blended learning models, EL1 (Ensemble learning-1) and EL2 (Ensemble learning-2). They are developed using pre-trained models
Transfer Learning with Fine-Tuning on MobileNet and GRAD-CAM
173
(ResNet, ResNeXt, DenseNet, VGG, Inception, MobileNet, and Spinal FC) with the best performance, and test accuracy is 0.8455, 0.8472, kappa of Cohen is 0.6907, 0.6942. The area associated with the fault layer under the receiver operating characteristic (ROC) curve (AUC) is 0.8862 and 0.8695. As a total result of 28 different classifiers, the highest test accuracy and Cohen’s kappa value were obtained in the EL2 model, and the highest AUC value was obtained in the EL1 model. In another work in [7], the authors applied a 169-layer convolutional neural network based on the architecture of DenseNet training on a dataset of 40,561 images from musculoskeletal 14,863 studies. The AUC of the model was 0.929 and used a Layer Activation Map (CAM) to visualize broken bones in multiple body parts. The study in [8] detected osteoporosis in rejected bones using radioactive machines and deep learning techniques. A learning model based on ResNet50 and XGBoost Classifier was used to predict gout or osteoporosis. In addition, the team used the RADTorch library to preprocess the X-ray images to identify defects quickly. Reliability is 98% and 91% with ResNet50 and the combined ResNet50 and XGBoost Classifier model, respectively. Research group in [9] identified abnormalities in musculoskeletal radiographs. The DenseNet-201 and InceptionResNetV2 models used deep transfer learning to optimize training on limited data, detected abnormalities in chest radiographs with 95% accuracy CI of 83–92% and higher sensitivity greater than that 0.9. However, in the case of finger radiographs, they are not satisfactory. The work [10] deployed GMDH-type neural networks in reading highresolution X-ray images to support the early diagnosis of knee arthritis. As a result, diagnostic accuracy reached 85% for the Lateral images and 77.5% for the Medial images. The team in [11] built a Decision Support System to help doctors detect broken bones from X-ray images. This system is trained using ResNet18 model and results in 85.99% accuracy for training data and 85.00% for testing. The work in [7] an integrated neural network (CNN) achieved results comparable to those of humans in fracture classification. The system is precisely designed to assist physicians, significantly save time, and limit the number of misdiagnoses. Accuracy achieved is 82.7%, 89.4% and 90.5 for VGG19, InceptionV3 and ResNet50. The authors in [12] introduced a new classification network. First, use the Faster Region with Convolutions Neural Network (Faster R-CNN) to detect 20 different bony regions in the X-ray images and then recognize whether each bone region is fractured using CrackNet. The results achieved 90.11% accuracy and 90.14% F-measure.
174
3
H. H. Luong et al.
Methods
Fig. 1. MobileNet for anomalies identification in bones.
Transfer Learning with Fine-Tuning on MobileNet and GRAD-CAM
175
We propose a combination of steps such as image preprocessing and model retraining from existing deep learning models (MobileNet) through the transfer learning method as exhibited in Fig. 1. We fine-tune the hyperparameters to fit the data being surveyed during the training process. For example, after determining bone disease through image recognition, we use trained models combined with GRAD-CAM to help mark areas that may be abnormal signs on the bone to support the diagnosis of bone disease. 3.1
Image Preprocessing
To increase the number of images used for the training, we applied the following solutions: Flip the image horizontally, rotate the image 15◦ , and limit random contrast from –0.2 to 0.2. Random gamma from 80 to 120, brightness increase from –0.2 to 0.2. The input size will be fine-tuned to 224 × 224 when included in the model. 3.2
Bone Disease Identification Based on Transfer Learning Approach
This study proposes using the MobileNet network model, a compact architecture with a few parameters that still achieves high image classification performance to identify bone problems. The overall architecture and the parameters of the MobileNet model are presented as shown below. For the training process to have the best effect, this study also used transfer learning and fine-tuning in the model training process. Transfer Learning Techniques. During training, we reuse the pre-trained MobileNet model to train on the dataset we collected. This model is pre-trained on the ImageNet dataset consisting of 1.4 million images and 1000 classes. Hyperparameter Tunning. During the model calibration phase, we continue to calibrate the hyperparameters to help the model achieve reasonably high accuracy on the survey data. The hyperparameters used for the calibration process are as follows: (a) Epoch: tested from 10–120 to find a suitable epoch threshold; (b) Batch size: tested from 16–512; (c) Learning rate tested: 0.001–0.000000001.
4 4.1
Experiments Data Description
The surveyed data set consists of WRIST, HUMERUS, and ELBOW are taken from the MURA V1.1 dataset showing abnormal conditions and no abnormality detected on these three sets. MURA [4] is one of the most extensive public X-ray image datasets. The number of data for these three parts includes
176
H. H. Luong et al.
17,367. First, experts and radiologists have labeled all photos. Then, each part is divided into two subclasses, including positive and negative, showing whether the image has a bone abnormality (positive) or not (negative). Abnormalities in the bones include fractures, splints in bone, etc. Of these, 4282 images show WRIST abnormalities (POSITIVE WRIST), and 6129 images show WRIST but without found abnormalities (NEGATIVE WRIST). For the HUMERUS, there were 739 images showing anomalies on HUMERUS (POSITIVE HUMERUS), including 821 images with no anomalies detected on HUMERUS (NEGATIVE HUMERUS), ELBOW images with 2236 anomalies on ELBOW (POSITIVE ELBOW) and 3160 negative images on ELBOW (NEGATIVE ELBOW). 4.2
Metrics for Comparison
This study divides the data set into two subsets for training and evaluation to evaluate the model. Accuracy (accuracy - acc) and F1 measures averaged over five runs to compare performance measures between MobileNet, VGG16, VGG19, and NASNetMobile architectures. 4.3
Anomalies Identification in Bones on Three Parts Including WRIST, HUMERUS, and ELBOW
Table 1. Comparison of convolutional neural networks for abnormal classifications in bone. Model
Train Acc Loss
MobileNet 0.8324 Transfer Learning NASNetMobile 0.1634 with default VGG19 0.2585 hyper-parameters VGG16 0.2585
Fine-Tuning
MobileNet NASNetMobile VGG19 VGG16
0.8374 0.2585 0.2585 0.2585
Test Acc F1
0.4703 1.8487 1.7906 1.7914
0.83 0.17 0.26 0.26
0.83 0.17 0.26 0.26
0.5528 1.792 1.7911 1.7911
0.84 0.26 0.26 0.26
0.84 0.26 0.26 0.26
This scenario evaluates transfer learning with three famous convolutional neural networks, including VGG16 [13], VGG19, and NASNetMobile [14] on anomalies identification on: WRIST, HUMERUS, and ELBOW. All training stages are performed on the same set of parameters: input image size = 224 × 224, epoch = 20, batch-size = 64. In which the learning process is transfer, correction, and values. Experimental values were presented in the previous section. Table 1 presents the comparison results between the methods. This result shows that
Transfer Learning with Fine-Tuning on MobileNet and GRAD-CAM
177
Fig. 2. Confusion matrix of MobileNet on the test set.
the MobileNet model can improve its accuracy after preprocessing the data and refining the model. In particular, we notice that the results on the training set and the test set are nearly equivalent. As exhibited in Fig. 2, the results are rather promising. The distinction between the normal and abnormal states of the wrist is higher than the others. In order to increase the reliability of the accuracy of automatic diagnostic results, we use GRAD-CAM (as illustrated in Fig. 3) to colorize the places that need to be noted on the image to help readers of the X-ray images quickly avoid capturing the focus on the X-ray image. For example, as observed from Fig. 3), the red area covered abnormalities in the bones as described in [4] that indicated that they could be fractures, splints in bone, etc.
178
H. H. Luong et al.
Fig. 3. X-ray image of elbow with bone problems before and after applying GRADCAM for identifying abnormal areas in the bone image.
4.4
Results Comparison with State-of-the-art and Discussion
From the above experimental results, we can see that transfer learning with MobileNet gives the best results compared to deep convolutional neural networks, which is suitable for the problem of diagnosing bone and joint problems. In addition, we also compare the results from the improved MobileNet model with the original MobileNet model. The results obtained from the proposed model give positive results with an accuracy of 0.84 compared to the original MobileNet model of 0.77. In addition, the results from this study are also entirely satisfactory compared with the studies presented in [9] with the highest accuracy on the part of 0.77 and measure F1 is 0.78 and in [15] with accuracy, and F1 measure is 0.62 and 0.44, respectively.
5
Conclusion
The experimental results also show that fine-tuning the hyperparameters on MobileNet to match the survey data has given better results than using the default set of hyperparameters. Furthermore, among the algorithms surveyed, MobileNet gives the best results. After the training, we used GRAD-CAM to highlight the location of interest (areas marked by “abnormalities” in the bone) in the image visually through color instead of black and white X-ray images. In the illustrated color range, the reds are the areas of the image that need attention so that X-ray readers can focus on the image areas containing important information after receiving the system’s automatic diagnosis results. It can
Transfer Learning with Fine-Tuning on MobileNet and GRAD-CAM
179
help provide a final diagnosis to the patient with high accuracy and saves the diagnostician’s imaging time. Regarding education, the proposal in this study provides a tool for medical students to practice reading X-ray images in diagnosing bone and joint diseases, improving their experience in X-ray imaging, avoiding mistakes due to lack of practical experience when going to work later, and improving the quality of medical examination and treatment for patients.
References 1. Antony, B., Singh, A.: Imaging and biochemical markers for osteoarthritis. Diagnostics 11(7), 1205 (2021). https://doi.org/10.3390/diagnostics11071205 2. Hallas, P., Ellingsen, T.: Errors in fracture diagnoses in the emergency department– characteristics of patients and diurnal variation. BMC Emerg. Med. 6(1) (2006). https://doi.org/10.1186/1471-227x-6-4 3. Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., Batra, D.: GradCAM: visual explanations from deep networks via gradient-based localization. Int. J. Comput. Vis. 128(2), 336–359 (2019). https://doi.org/10.1007%2Fs11263-01901228-7 4. Rajpurkar, P., et al.: Mura: large dataset for abnormality detection in musculoskeletal radiographs (2017). https://arxiv.org/abs/1712.06957 5. Solovyova, A., Solovyov, I.: X-ray bone abnormalities detection using mura dataset (2020). https://arxiv.org/abs/2008.03356 6. Uysal, F., Hardala¸c, F., Peker, O., Tolunay, T., Tokg¨ oz, N.: Classification of shoulder x-ray images with deep learning ensemble models. Appl. Sci. 11(6), 2723 (2021). https://doi.org/10.3390/app11062723 7. Tanzi, L., Vezzetti, E., Moreno, R., Moos, S.: X-ray bone fracture classification using deep learning: a baseline for designing a reliable approach. Appl. Sci. 10(4), 1507 (2020). https://doi.org/10.3390/app10041507 8. Nandi, R., Mulimani, M.: Detection of COVID-19 from x-rays using hybrid deep learning models (April 2021). https://doi.org/10.21203/rs.3.rs-468236/v1 9. Chada, G.: Machine learning models for abnormality detection in musculoskeletal radiographs. Reports 2(4), 26 (2019). https://doi.org/10.3390/reports2040026 10. Jakaite, L., Schetinin, V., Hlad˚ uvka, J., Minaev, S., Ambia, A., Krzanowski, W.: Deep learning for early detection of pathological changes in x-ray bone microstructures: case of osteoarthritis. Sci. Rep. 11(1) (2021). https://doi.org/10.1038/ s41598-021-81786-4 11. Hardala¸c, F., et al.: Fracture detection in wrist x-ray images using deep learningbased object detection models. Sensors 22(3), 1285 (2022). https://doi.org/10.3390 %2Fs22031285 12. Ma, Y., Luo, Y.: Bone fracture detection through the two-stage system of cracksensitive convolutional neural network. Inform. Med. Unlocked 22, 100452 (2021). https://doi.org/10.1016/j.imu.2020.100452 13. Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition (2014). https://arxiv.org/abs/1409.1556 14. Zoph, B., Vasudevan, V., Shlens, J., Le, Q.V.: Learning transferable architectures for scalable image recognition (2017). https://arxiv.org/abs/1707.07012 15. Mall, P.K., Singh, P.K., Yadav, D.: GLCM based feature extraction and medical x-ray image classification using machine learning techniques. In: 2019 IEEE Conference on Information and Communication Technology, pp. 1–6 (2019)
A Fuzzy-Based System for Handover in 5G Wireless Networks Considering Network Slicing Constraints Phudit Ampririt1(B) , Ermioni Qafzezi1 , Kevin Bylykbashi2 , Makoto Ikeda2 , Keita Matsuo2 , and Leonard Barolli2 1
2
Graduate School of Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan {bd21201,bd20101}@bene.fit.ac.jp Department of Information and Communication Engineering, Fukuoka Institute of Technology, 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan [email protected], [email protected], {kt-matsuo,barolli}@fit.ac.jp
Abstract. Handover in 5G Wireless Networks introduces new and complex challenges, because a user does not handover to different base stations or access technologies but also different slices. The constraints on Network Slicing (NS) should be considered when making a handover decision for satisfying user requirements. In this paper, we propose a Fuzzy-based system for Handover considering three parameters: Slice Delay (SD), Slice Bandwidth (SB) and Slice Stability (SS). From simulation results, we conclude that the considered parameters have different effects on the Handover Decision (HD). When SD is increased, the HD parameter is increased but when SB and SS are increasing, the HD parameter is decreased.
1
Introduction
In 5G wireless networks, the massive growth of users device with unpredictable traffic patterns will create large data volume on the Internet, causing congestion and Quality of Service (QoS) deterioration [1]. Also, Handover process is the one critical component for mobility management and can affect the overall network performance [2]. Many Handover (HO) scenarios and different HO rates may occur, which bring problems on ensuring stable and reliable connections [3]. For dealing with these problems, the 5G Wireless Networks will provide increased performance in terms of throughput, latency, reliability and mobility in order to fullfil the QoS requirements in many application scenarios. The 5G is developing by considering three main different usage scenarios which have been identified as enhanced mobile broadband (eMBB), ultra-reliable & low latency communications (URLLC) and massive type communication (mMTC). The eMBB is related to human-essential and has greater accessibility to multimedia content and services by enhancing seamless Quality of Experience (QoE). c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 180–189, 2022. https://doi.org/10.1007/978-3-031-08812-4_18
A Fuzzy-Based System for Handover
181
The URLLC can efficiently reduce the latency and enhance reliability. The mMTC can accommodate a large number of connected devices while maintaining a long battery life [4–6]. Recently, several research projects are attempting to develop systems that are suited for the 5G era. One of them is the Software-Defined Network (SDN) [7]. Also, introducing Fuzzy Logic (FL) to SDN controllers enhance the QoS. In addition, the mobile handover method with SDN is used to reduce changeover processing delays [8–10]. This paper presents a Fuzzy-based system for Handover in 5G Wireless Networks considering three parameters: Slice Delay (SD), Slice Bandwidth (SB), Slice Stability (SS). The rest of the paper is organized as follows. In Sect. 2 is presented an overview of SDN. In Sect. 3, we present 5G Network Slicing. In Sect. 4, we describe the proposed Fuzzy-based system and its implementation. In Sect. 5, we discuss the simulation results. Finally, conclusions and future work are presented in Sect. 6.
2
Software-Defined Networks (SDNs)
The SDN is a new networking paradigm that decouples the data plane from control plane in the network. By SDN is easy to manage and provide network software based services from a centralised control plane. The SDN control plane is managed by SDN controller or cooperating group of SDN controllers. The SDN structure is shown in Fig. 1 [11,12]. • Application Layer builds an abstracted view of the network by collecting information from the controller for decision-making purposes. The types of applications are related to network configuration and management, network monitoring, network troubleshooting, network policies and security. • Northbound Interfaces allow communication between the control layer and the application layer and can provide a lot of possibilities for networking programming. Based on the needs of the application, it will pass commands and information to the control layer and make the controller creates the best possible software network with suitable qualities of service and acceptable security. • Control Layer receives instructions or requirements from the Application Layer. It contains the controllers that control the data plane and forward the different types of rules and policies to the infrastructure layer through the Southbound Interfaces. • Southbound Interfaces allow connection and interaction between the control plane and the data plane. The southbound interface is defined as protocols that allow the controller to create policies for the forwarding plane. • Infrastructure Layer receives orders from SDN controller and sends data among them. This layer represents the forwarding devices on the network such as routers, switches and load balancers.
182
P. Ampririt et al.
The SDN can manage network systems while enabling new services. In congestion traffic situation, the SDN can control and adapt resources appropriately throughout the control plane. Mobility management is easier and quicker in forwarding across different wireless technologies (e.g. 5G, 4G, Wifi and Wimax). Also, the handover procedure is simple and the delay can be decreased.
Fig. 1. Structure of SDN.
3
5G Network Slicing
The Network Slicing (NS) is a technology that divides a single virtualized infrastructure into multiple virtual end-to-end networks which can be called as “Slices” that is configured into virtualized function follow the demand of application to respond to the user’s requests. Each slice is logically independent and doesn’t have any effect on other virtual logical networks [13–15]. A network with NS compared with the traditional networks can provide better performance and can be flexible for different service requirements and number of users. Also, because the slices don’t affect each other, the slice reliability and security can be improved [16]. The 5G NS concept is developed by the Next Generation Mobile Networks (NGMN) as shown in Fig. 2. The NS process is divided into three main layers [17, 18]. • The Service Instance Layer represents a service (end-user service or business services) which is provided by application provider or mobile network operator.
A Fuzzy-Based System for Handover
183
Fig. 2. The NGMN NS concept.
• The Network Slice Instance Layer is a set of network functions and resources which provide the network slice instance to accommodate the required network characteristics (ultra-low-latency, ultra-reliability) by the service instances. • The Resource Layer comprises of physical resources and logical resources for the slice deployment.
4
Proposed Fuzzy-Based System
In this work, we use FL to implement the proposed system. In Fig. 3, we show the overview of our proposed system. Each evolve Base Station (eBS) will receive controlling order from the SDN controller and they can communicate and send data with User Equipment (UE). Also, each eBS can cover many slices for different applications. On the other hand, the SDN controller will collect all the data about network traffic status and control eBS to manage inter-eBS handover and inter-slice handover by using the proposed Fuzzy-based system. The SDN controller will be a communication bridge between eBS and the 5G core network. The proposed system is called Fuzzy-based Handover System (FBHS) in 5G Wireless Networks. The structure of FBHS is shown in Fig. 4. For the implementation of our system, we consider three input parameters: Slice Delay (SD), Slice Bandwidth (SB), Slice Stability (SS) and the output parameter is Handover Decision (HD). Slice Delay (SD): The slice with high delay causes high queueing and link delay. Therefore, the Handover is needed to fulfill the QoS. Slice Bandwidth (SB): Slice Bandwidth is the available bandwidth of a slice. When SB is higher, the Handover possibility will be lower. Slice Stability (SS): The slice with high stability can provide consistent communication service and exhibit a stable performance. If the SS is low, the user will consider the handover to another slice with higher stability.
184
P. Ampririt et al.
Fig. 3. Proposed system overview.
Fig. 4. Proposed system structure.
Handover Decision (HD): The HD parameter determines whether or not to perform the handover procedure. The membership functions are shown in Fig. 5. We use triangular and trapezoidal membership functions because they are more suitable for real-time operations [19–22]. We show parameters and their term sets in Table 1. The Fuzzy Rule Base (FRB) is shown in Table 2 and has 27 rules. The control rules have the form: IF “condition” THEN “control action”. For example, for Rule 1: “IF SD is L, SB is Lo and SS is Lw THEN HD is HD5”.
A Fuzzy-Based System for Handover Table 1. Parameter and their term sets. Parameters
Term set
Slice Delay (SD)
Low (L), Medium (M), High (H)
Slice Bandwidth (SB)
Low (Lo), Medium (Me), High (Hi)
Slice Stability (SS)
Low (Lw), Medium (Mi), High (Hg)
Handover Decision (HD)
HD1, HD2, HD3, HD4, HD5, HD6, HD7 Table 2. FRB.
Rule SD SB SS HD 1
L
Lo Lw HD5
2
L
Lo Mi HD4
3
L
Lo Hg HD3
4
L
Me Lw HD4
5
L
Me Mi HD3
6
L
Me Hg HD2
7
L
Hi
Lw HD3
8
L
Hi
Mi HD2
9
L
Hi
Hg HD1
10
M
Lo Lw HD6
11
M
Lo Mi HD5
12
M
Lo Hg HD4
13
M
Me Lw HD5
14
M
Me Mi HD4
15
M
Me Hg HD3
16
M
Hi
Lw HD4
17
M
Hi
Mi HD3
18
M
Hi
Hg HD2
19
H
Lo Lw HD7
20
H
Lo Mi HD6
21
H
Lo Hg HD5
22
H
Me Lw HD6
23
H
Me Mi HD5
24
H
Me Hg HD4
25
H
Hi
Lw HD5
26
H
Hi
Mi HD4
27
H
Hi
Hg HD3
185
186
5
P. Ampririt et al.
Simulation Results
In this section, we present the simulation result of our proposed system. The simulation results are shown in Fig. 6, Fig. 7 and Fig. 8. They show the relation of HD with SS for different SB values considering SD as a constant parameter. In Fig. 6, we consider the SD value 0.1 ms. When SS is increased, we see that HD is decreased. For SD 0.1 ms and SB 50%, when SS is increased from 20% to 50% and 50% to 80%, the HD is decreased by 11% and 10%, respectively. That means the present slice is more stable and the handover possibility to the other slices is low.
(a) Slice Delay
(b) Slice Bandwidth
(c) Slice Stability
(d) Handover Decision
Fig. 5. Membership functions.
We compare Fig. 6 with Fig. 7 to see how SD has affected HD. We change the SD value from 0.1 ms to 0.5 ms. The HD is increasing by 11% when the SB value is 50% and the SS is 50%. This is because the present slice delay is higher. Thus, the handover to another slice is needed.
A Fuzzy-Based System for Handover
187
SD=0.1 ms 1
SB=10% SB=50% SB=90%
0.9 0.8
HD [unit]
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0
10
20
30
40
50 60 SS [%]
70
80
90
100
Fig. 6. Simulation results for SD = 0.1 ms. SD=0.5ms 1
SB=10% SB=50% SB=90%
0.9 0.8
HD [unit]
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0
10
20
30
40
50 60 SS [%]
70
80
90
100
Fig. 7. Simulation results for SD = 0.5 ms.
We increase the value of SD to 0.9 ms in Fig. 8. Comparing the results with Fig. 6 and Fig. 7, we can see that the HD values have increased significantly. For SB value 10%, all HD values are higher than 0.5. Thus, the mobile device will make a handover to another slice.
188
P. Ampririt et al. SD=0.9 ms 1 0.9 0.8
HD [unit]
0.7 0.6 0.5 0.4 0.3 0.2 SB=10% SB=50% SB=90%
0.1 0 0
10
20
30
40
50 60 SS [%]
70
80
90
100
Fig. 8. Simulation results for SD = 0.9 ms.
6
Conclusions and Future Work
In this paper, we proposed and implemented a Fuzzy-based system for Handover in 5G Wireless Networks. We considered three parameters: SD, SB and SS to decide the HD value. We evaluated the proposed system by simulations. From the simulation results, we found that three parameters have different effects on the HD. When SD is increasing, the HD parameter is increased but when SB and SB are increasing, the HD parameter is decreased. In the future work, we will consider different parameters and perform extensive simulations to evaluate the proposed system.
References 1. Navarro-Ortiz, J., Romero-Diaz, P., Sendra, S., Ameigeiras, P., Ramos-Munoz, J.J., Lopez-Soler, J.M.: A survey on 5G usage scenarios and traffic models. IEEE Commun. Surv. Tutor. 22(2), 905–929 (2020) 2. Sun, Y., et al.: Efficient handover mechanism for radio access network slicing by exploiting distributed learning. IEEE Trans. Netw. Serv. Manag. 17(4), 2620–2633 (2020) 3. Saad, W.K., Shayea, I., Hamza, B.J., Mohamad, H., Daradkeh, Y.I., Jabbar, W.A.:Handover parameters optimisation techniques in 5G networks. Sensors 21(15), 5202, 22 (2021). https://doi.org/10.3390/s21155202 4. Akpakwu, G.A., Silva, B.J., Hancke, G.P., Abu-Mahfouz, A.M.: A survey on 5G networks for the internet of things: communication technologies and challenges. IEEE Access 6, 3619–3647 (2018) 5. Palmieri, F.: A reliability and latency-aware routing framework for 5g transport infrastructures. Comput. Netw. 179(9), October 2020. https://doi.org/10.1016/j. comnet.2020.107365. Article 107365 6. Kamil, I.A., Ogundoyin, S.O.: Lightweight privacy-preserving power injection and communication over vehicular networks and 5g smart grid slice with provable security. Internet Things 8(100116), 100–116 (2019)
A Fuzzy-Based System for Handover
189
7. Hossain, E., Hasan, M.: 5G cellular: key enabling technologies and research challenges. IEEE Instrum. Meas. Mag. 18(3), 11–21 (2015) 8. Yao, D., Su, X., Liu, B., Zeng, J.: A mobile handover mechanism based on fuzzy logic and MPTCP protocol under SDN architecture*. In: 18th International Symposium on Communications and Information Technologies (ISCIT-2018), pp. 141– 146, September 2018 9. Lee, J., Yoo, Y.: Handover cell selection using user mobility information in a 5G SDN-based network. In: 2017 Ninth International Conference on Ubiquitous and Future Networks (ICUFN-2017), pp. 697–702, July 2017 10. Moravejosharieh, A., Ahmadi, K., Ahmad, S.: A fuzzy logic approach to increase quality of service in software defined networking. In: 2018 International Conference on Advances in Computing,Communication Control and Networking (ICACCCN2018), pp. 68–73, October 2018 11. Li, L.E., Mao, Z.M., Rexford, J.: Toward software-defined cellular networks. In: 2012 European Workshop on Software Defined Networking, pp. 7–12, October 2012 12. Mousa, M., Bahaa-Eldin, A.M., Sobh, M.: Software defined networking concepts and challenges. In: 2016 11th International Conference on Computer Engineering & Systems (ICCES-2016), pp. 79–90. IEEE (2016) 13. An, N., Kim, Y., Park, J., Kwon, D.H., Lim, H.: Slice management for quality of service differentiation in wireless network slicing. Sensors 19, 2745 (2019) 14. Jiang, M., Condoluci, M., Mahmoodi, T.: Network slicing management & prioritization in 5G mobile systems. In: European Wireless 2016; 22th European Wireless Conference, pp. 1–6. VDE (2016) 15. Chen, J., et al.: Realizing dynamic network slice resource management based on SDN networks. In:2019 International Conference on Intelligent Computing and its Emerging Applications (ICEA), pp. 120–125 (2019) 16. Li, X., et al.: Network slicing for 5G: challenges and opportunities. IEEE Internet Comput. 21(5), 20–27 (2017) 17. Afolabi, I., Taleb, T., Samdanis, K., Ksentini, A., Flinck, H.: Network slicing and softwarization: a survey on principles, enabling technologies, and solutions. IEEE Commun. Surv. Tutor. 20(3), 2429–2453 (2018) 18. Alliance, N.: Description of network slicing concept. NGMN 5G P 1, 7 (2016). https://ngmn.org/wp-content/uploads/160113 NGMN Network Slicing v1 0.pdf 19. Norp, T.: 5G requirements and key performance indicators. J. ICT Stand. 6(1), 15–30 (2018) 20. Parvez, I., Rahmati, A., Guvenc, I., Sarwat, A.I., Dai, H.: A survey on low latency towards 5G: ran, core network and caching solutions. IEEE Commun. Surv. Tutor. 20(4), 3098–3130 (2018) 21. Kim, Y., Park, J., Kwon, D.H., Lim, H.: Buffer management of virtualized network slices for quality-of-service satisfaction. In: 2018 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN-2018), pp. 1–4 (2018) 22. Barolli, L., Koyama, A., Yamada, T., Yokoyama, S.: An integrated CAC and routing strategy for high-speed large-scale networks using cooperative agents. IPSJ J. 42(2), 222–233 (2001)
A Focused Beam Routing Protocol Considering Node Direction for Underwater Optical Wireless Communication in Delay Tolerant Networks Keita Matsuo1(B) , Elis Kulla2 , and Leonard Barolli1 1
Department of Information and Communication Engineering, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan {kt-matsuo,barolli}@fit.ac.jp 2 Department of System Management, Fukuoka Institute of Technology (FIT), 3-30-1 Wajiro-Higashi, Higashi-Ku, Fukuoka 811-0295, Japan
Abstract. Recently, underwater communication has been developing in many ways, such as wired communication, underwater acoustic communication (UAC), underwater radio wave wireless communication (URWC), underwater optical wireless communication (UOWC). The main issue in underwater communication is the communication interruption because signals are affected by various factors in underwater environment. Consequently, communication links are unstable and real time communication is almost impossible. Therefore, we considered combining both underwater communication and delay tolerant network technologies. In this paper, we present the Focused Beam Routing considering node Direction (FBRD) protocol for UOWC, we use ONE simulator to evaluate the performance regarding delivery probability. The results show better results for FBR angles around 30◦ , while if we use angles over 30◦ , the delivery probability decreases, because of the short communication ranges. In addition, the proposed FBRD protocol not only achieved high delivery provability than FBR but also has lower hop counts.
1
Introduction
Recently, different technologies are being use to enable communication in underwater environment, such as wired communication (WC), underwater acoustic communication (UAC), underwater radio wave wireless communication (URWC), underwater optical wireless communication (UOWC). So far, underwater communication remains realized until nowadays via communication cables due to the limited development of underwater wireless communications [4], and the high cost of hydrophones and other equipment. UAC is a popular technique which uses sound signals in water and enables transmission of signal which emitted from sensors, robots or submarines to longer c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 190–199, 2022. https://doi.org/10.1007/978-3-031-08812-4_19
A Focused Beam Routing Protocol Considering Node Direction
191
distances. However, the bandwidth is narrow and transmission speed is relatively low. Moreover, the speed of sound is affected by temperature, depth and salinity of underwater environment. These factors produce variations in speed of sound in underwater environment [2]. In URWC, radio waves are used for communication. The radio wave employed for terrestrial wireless communication are also utilized for underwater communication, it achieves high data rate for short communication range and suffers from Doppler effect [4]. There are some former research for absorption losses in water. According to [7], the absorption losses are about 9 dB to 19 dB in freshwater at 2.4 GHz, the absorption losses are about 19 dB to 24 dB in river water at 2.4 GHz, and absorption losses are about 25 dB to 30 dB in seawater at 2.4 GHz. If we use URWC in the sea environment, the losses are bigger than other kind of water. For this reason, it is very difficult to use high frequency radio waves in underwater environment. For instance, if we use radio waves in water environment, the transmission distance would be few centimeters. UOWC uses visible light for communication, which allows longer communication distances in the water. According to [8], the communication distance in the water by using visible light is over 100 m at 20 Mbps. Recently, laser diodes (LDs) and light emitting diode (LEDs) have developed quickly, therefore many lighting devices are shifting from conventional light to diodes as a LEDs light. The main features of LEDs are high energy efficiency and quick response which is sufficient to achieve high speed communication by using UOWC, it is possible to implement broadband underwater network. As we described above, underwater communication links are unstable because they are affected by various factors in underwater environment. Therefore, we considered delay tolerant network (DTN) as a suitable technology for underwater environment. In Table 1 is shown comparison of underwater communications. Table 1. Comparison of underwater communications. Comunication method Communication distance Transmission speed WC
Depend on cable length
High
UAC
Long
Low
URWC
Short
High
UOWC
Middle
High
These communication ways have characteristics strengths and weaknesses. In order to use high speed wireless network in under water environment, we need to choose both URWC or UOWC. Especially, in URWC that can extendet the communication distance by decreasing the radio frequency, but in this case the bandwidth would become narrower, causing lower communication speed. So, we have focused on UOWC by using DTNs technologies to make good communication environment in underwater.
192
K. Matsuo et al.
In this paper, we propose and evaluate Focused Beam Routing considering node Direction (FBRD) protocol for underwater optical wireless network. The evaluation results show that the FBRD protocol was able to carry the data to the surface station and showed better performance than focused beam routing (FBR). The rest of this paper is structured as follows. In Sect. 2, we introduce the related work. In Sect. 3, we present the proposed FBRD protocol with DTNs for UOWC. In Sect. 4, we describe implementation of FBR and FBRD in The ONE Simulator. In Sect. 5, we show the simulation results. Finally, conclusions and future work are given in Sect. 6.
2
Related Work
So far, many studies of UOWC have published in several years. The UOWC reduced absorption window for blue-green light. Due to its higher bandwidth, underwater optical wireless communications can support higher data rates at low latency levels compared to acoustic and RF counterparts [9]. Moreover some commercial UOWC modems were released. One of the research teams started developing UOWC, their try was achieved communication with over 100 m range at 20 Mbps in bi-directional with prototype UOWC modem [8]. The research of UOWC is not new that was started from 1960s to make high speed communication techniques in the lake or sea. However, it has accelerated after invented of blue LEDs. One of the interest research regions is compared LEDs colors that compared to the open ocean where blue LEDs perform well for data communications, in coastal and harbor environments optical transmission becomes worse and the color of lowest attenuation shifts to green. Another problem concerns the“green-yellow gap” of LEDs, as the quantum efficiency of current commercially available green LEDs is poor [11]. The direction of UOWS signals is narrow. If transmitter makes the extremely narrow range visible light signals as laser’s light by using LEDs, it will achieve long distance communication, while it makes the wide range visible light signal, the receiver becomes easier to find the signals instead of long communication distance. Delay tolerant networks (DTNs) are intermittently connected communications that would be helping from repeatedly and long time disconnection because of various reasons such as out of communication range, short of battery, and so on. The DTNs include wireless sensor networks using scheduled intermittent connectivity, mobile ad hoc networks, satellite networks with periodic connectivity, village networks, wildlife tracking networks, and pocket switched networks, etc. Due to the broad application prospect, delay tolerant networks attract much attention [14]. The DTNs have been utilized in various communication paradigms such as vehicular delay tolerant networks (VDTNs) [1,3,10], and DTNs system in a communication network on a railway line [12,13].
A Focused Beam Routing Protocol Considering Node Direction
193
Fig. 1. Image of UOWC by using DTNs.
FBR is a well-known routing protocol for under water sensor network (UWSN), which considers as forward candidates, only active neighbors which location is inside a region specified by the direction from the forwarding node to the destination by considering an arbitrary angle and the communication distance [6].
3
Proposed FBR Considering Node Direction Protocol on DTNs for UOWC
In this section, we propose FBRD protocol. First, we would like to describe about FBR protocol with DTNs for UOWC. In Fig. 1 is shown image of UOWC by using DTNs, autonomous underwater vehicles (AUVs) can communicate using DTNs’ store-carry-and-forward paradigm. Surface station could receive data from AUVs then the station send the data to the surface sink or monitoring center in ground. Conventional marine research using submarine could not exchange data with ground unless WC was used. Therefore, it was necessary to wait for the submarine to return in order to get the data. However, using the method that we propose, it will be possible to send the data from the underwater to the sea surface. Also, it can send large size data such as photos, sounds and videos because of UOWN links. Moreover, by using UOWC various protocols and real time applications can be implemented. If the communication are become intermittently connected between nodes (AUVs, submarine, surface station), they can send the data by DTNs with FBR. If the emitting visible lights of UOWN (signals) is narrow, the communication distance will be longer, whereas when the emitting visible lights has wide range, the distance will be short. We show the image of FBR protocol for UOWC in Fig. 2. When we use FBR with 1◦ , it will transmit the signals to longer distance, while if we use 30◦ , the distance will be short. However, it can transmit the signals to more nodes. Also, if it uses 360◦ the sender node emits the signal to
194
K. Matsuo et al.
Fig. 2. Image of FBR protocol for UOWC.
Fig. 3. Image of FBRD protocol.
omnidirectional like epidemic, though sender node sends many receiver nodes the signals, the transmit distance will be minimum. From here, we would like to explain the FBRD protocol which is shown in Fig. 3. FBRD considers moving direction of receiver node. When a receiver node is in the range of communication and it is moving downwards, the sender node would not forward messages to this node. As shown in Fig. 3, the receiver nodes 1, 5, 6 are inside of communication range. They can get packet signal from sender. However, node 1 and 6 are moving downwards, so they will not be forwarded message from the sender node. Only node 5 will receive the message from the sender node, because it is moving upwards.
A Focused Beam Routing Protocol Considering Node Direction
4
195
Implementation of FBR and FBRD in the ONE Simulator
The Opportunistic Networking Environment (ONE) simulator specifically designed for evaluating DTNs routing and application protocols. It allows users to create scenarios based upon different synthetic movement models and realworld traces and offers a framework for implementing routing and application protocols [5]. Thus, we implemented FBR and FBRD protocol in The ONE Simulator. The following assumptions are made, in order to simplify the implementation: • The environment is considered with only 2 dimensions: the width on the horizontal axis and the depth in the vertical axis. • Every node knows self location and destination. • Every participating node is mobile and moves based on the Random Waypoint mobility model, as implemented in The ONE Simulator. • There is only one surface station, and all transmissions are directed towards this surface station, which is located in the middle top of the simplified 2D environment.
5
Simulations
In order to investigate for FBRD protocol on UOWC, we used The ONE Simulator. We implemented FBRD protocol in the simulator. We assumed the communication ranges of UOWC as shown in Table 2. In Table 3 is shown simulation parameters’ settings. The FBR angle is always directed towards surface station. Table 2. FBR angles and communication ranges. FBR angles [θ/2] Communication ranges [m] 1 15 30 60 90 180
195 142 100 50 25 3
For different angles and different communication range, we show the simulation results for delivery probability in Fig. 4. Also, we show the box plot (see Fig. 5) of delivery probability to confirm the best angle for FBRD protocol. From these results, if we use the FBR angles between 0.5 to around 30◦ , we would achieve higher delivery probability. While, if we use angles above 35◦ the delivery probability decreases noticeably. It is obvious from boxplot graph of
196
K. Matsuo et al. Table 3. Simulation parameters’ settings. Parameters
Value
Transmit speed 20 [Mbps] 800–1200 [kB] Data size 8–12 [s] Ivent interval Number of surface stations 1 Static (middle of top) 16 Noumber of nodes Random waypoint Movement model 10 [h] Simulation time 400 × 400 [m] Simulation area 300 MBytes Buffer size
the delivery probability (Fig. 5), that after 30◦ the value of delivery probability decreases drastically. We also investigated the relation of delivery probability and hop count at 30◦ , for FBR and FBRD protocol. The comparison results are shown in Fig. 6
Fig. 4. Delivery probability at each angle of FBRD protocol for UOWC.
From these results, we can conclude that the FBRD protocol is better than FBR protocol in term of delivery probability with low hop counts. This means the sender node might decrease consumption of electric power, and this will increase their battery life time.
A Focused Beam Routing Protocol Considering Node Direction
197
Fig. 5. Box plot of delivery probability for FBRD protocol.
Fig. 6. Compared delivery probability and hopcounts between FBR and FBRD.
6
Conclusions and Future Work
In this paper, we introduced some of underwater communication technologies and compared them. Also, we explained UOWC as a potential technology in the future of underwater communications. We implemented FBR and FBRD protocol in The ONE Simulator. In the simulation, we evaluated FBRD protocol for different angles, then we compared the performance between FBR and FBRD protocol for UOWC. The results showed the FBRD protocol is better than FBR protocol in terms of delivery probability with low hop counts. This means the sender node consumes less electric power and will increase their battery life time.
198
K. Matsuo et al.
We found the best FBR angle is around 30◦ , while over 30◦ the values drastically decrease, because of the short communication ranges. In addition, the proposed FBRD protocol not only achieved higher delivery provability than FBR. but also to keep lower hop counts. In the future, we would like to add other protocols and functions to the simulation system and carry out extensive simulations to evaluate various underwater environments.
References 1. Ahmed, S.H., Kang, H., Kim, D.: Vehicular delay tolerant network (VDTN): routing perspectives. In: 2015 12th Annual IEEE Consumer Communications and Networking Conference (CCNC), pp. 898–903. IEEE (2015) 2. Awan, K.M., Shah, P.A., Iqbal, K., Gillani, S., Ahmad, W., Nam, Y.: Underwater wireless sensor networks: a review of recent issues and challenges. Wirel. Commun. Mob. Comput. 2019, 20 (2019). https://doi.org/10.1155/2019/6470359. Article ID 6470359 3. Azuma, M., Uchimura, S., Tada, Y., Ikeda, M., Barolli, L.: A hybrid message delivery method for vehicular DTN considering impact of shuttle buses and roadside units. In: Barolli, L., Woungang, I., Enokido, T. (eds.) AINA 2021. LNNS, vol. 227, pp. 211–218. Springer, Cham (2021). https://doi.org/10.1007/978-3-03075078-7 22 4. Jouhari, M., Ibrahimi, K., Tembine, H., Ben-Othman, J.: Underwater wireless sensor networks: a survey on enabling technologies, localization protocols, and internet of underwater things. IEEE Access 7, 96879–96899 (2019) 5. Ker¨ anen, A., Ott, J., K¨ arkk¨ ainen, T.: The one simulator for DTN protocol evaluation. In: Proceedings of the 2nd International Conference on Simulation tools and Techniques, pp. 1–10 (2009) 6. Kulla, E., Katayama, K., Matsuo, K., Barolli, L.: Enhanced focused beam routing in underwater wireless sensor networks. In: Barolli, L., Natwichai, J., Enokido, T. (eds.) EIDWT 2021. LNDECT, vol. 65, pp. 1–9. Springer, Cham (2021). https:// doi.org/10.1007/978-3-030-70639-5 1 7. Qureshi, U.M., et al.: RF path and absorption loss estimation for underwater wireless sensor networks in different water environments. Sensors 16(6), 890 (2016) 8. Sawa, T., Nishimura, N., Tojo, K., Ito, S.: Practical performance and prospect of underwater optical wireless communication: -results of optical characteristic measurement at visible light band under water and communication tests with the prototype modem in the sea-. IEICE Trans. Fundam. Electron. Commun. Comput. Sci. 102(1), 156–167 (2019) 9. Schirripa Spagnolo, G., Cozzella, L., Leccese, F.: Underwater optical wireless communications: overview. Sensors 20(8), 2261 (2020) 10. Spaho, E., Barolli, L., Kolici, V., Lala, A.: Performance evaluation of different routing protocols in a vehicular delay tolerant network. In: 2015 10th International Conference on Broadband and Wireless Computing, Communication and Applications (BWCCA), pp. 157–162. IEEE (2015) 11. Sticklus, J., Hoeher, P.A., R¨ ottgers, R.: Optical underwater communication: the potential of using converted green LEDs in coastal waters. IEEE J. Oceanic Eng. 44(2), 535–547 (2018)
A Focused Beam Routing Protocol Considering Node Direction
199
12. Tikhonov, E., Schneps-Schneppe, D., Namiot, D.: Delay tolerant network potential in a railway network. In: 2020 26th Conference of Open Innovations Association (FRUCT), pp. 438–448. IEEE (2020) 13. Tikhonov, E., Schneps-Schneppe, D., Namiot, D.: Delay tolerant network protocols for an expanding network on a railway. In: 2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies (3ICT), pp. 1–6. IEEE (2020) 14. Xiao, M., Huang, L.: Delay-tolerant network routing algorithm. J. Comput. Res. Dev. 46(7), 1065 (2009)
Taming Multi-node Accelerated Analytics: An Experience in Porting MATLAB to Scale with Python Paolo Viviani1(B) , Giacomo Vitali1 , Davide Lengani2 , Alberto Scionti1 , Chiara Vercellino1 , and Olivier Terzo1 1 LINKS Foundation, Turin, Italy {paolo.viviani,giacomo.vitali,alberto.scionti,chiara.vercellino, olivier.terzo}@linksfoundation.com 2 DIME - University of Genova, CINI HPC-KTT Laboratory, Genoa, Italy [email protected]
Abstract. High Performance Data Analytics (HPDA) at scale is a multifaceted problem that involves distributed computing resources, their location with respect to the data placement, fast networks, optimized I/O, hardware accelerators and a fairly complex software stack that draws both from the legacy of HPC and the cloud world. This complexity does not cope well with the needs of domain experts who desire to focus on their algorithms without having to understand all the features of the underlying infrastructure. Among these domain experts, engineers often rely on MATLAB to quickly model their complex numerical computations with a simple, textbook-like syntax and an effective Integrated Development Environment (IDE). On the other end, MATLAB was not designed with large-scale, out-of-core computations in mind, despite the introduction of some parallel computing tools (e.g., distributed arrays and spmd loops). In an ideal world, a domain expert should only focus on its application logic, while runtime/parallel computing experts should provide tools that act behind the scenes to efficiently distribute the computations on available resources and to provide optimal performance. Conversely, in real life it often happens that the domain expert prototypes its code with MATLAB on a small scale, then HPDA/HPC experts will leverage proper tools to deploy it at scale; this causes a significant effort overhead, as development needs to be performed twice, possibly with multiple iterations. The rise of Python as the language of choice for a huge number of data scientists, along with its open ecosystem, led to the development of many tools that tried to achieve a convergence between the goals of domain experts and the need for performance and scalability. Sometimes building upon existing frameworks (e.g. PySpark) or starting from scratch in pure Python (e.g. Dask), these tools bear promise to allow engineers to write their code with a reasonably textbook-like syntax, while retaining the capability to scale among multiple nodes (and accelerators) with minimal intervention to the application code. This paper discusses the process of porting an engineering application written in MATLAB (parallelized with existing toolboxes) to Python using Dask. An indication of the scalability results is also given, including a 20x speedup with respect to the MATLAB code using the same setup. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 200–210, 2022. https://doi.org/10.1007/978-3-031-08812-4_20
Accelerated Analytics: From MATLAB to Python
1
201
Introduction
As the availability of high-fidelity engineering simulations becomes more common thanks to the low access barrier to High-Performance Computing (HPC) resources, the need to extract engineering knowledge and value from the significant amount of data they produce becomes critical. On the other hand, the sheer amount of data itself introduces significant technical challenges to perform the kinds of analyses, due to both the need to handle datasets that possibly do not fit the memory of the given workstation, and the need to provide results in a short time to inform product development decisions. These challenges typically fall within the High Performance Data Analytics (HPDA) topic, that involves distributed computing resources, their allocation with respect to the data placement, fast networks, optimized I/O, hardware accelerators and a fairly stratified software stack that draws both from the legacy of HPC and the cloud world. This complexity does not cope well with the needs of domain experts who desire to focus on their algorithms without having to understand all the features of the underlying infrastructure; in fact, engineers typically like to quickly model their numerical computations using MATLAB, leveraging the simple, textbook-like syntax and the powerful Integrated Development Environment (IDE). On the other end, MATLAB was not designed with large-scale, out-of-core computations in mind, and despite the introduction of powerful parallel computing tools (e.g., distributed arrays), it is necessary to heavily mix the application logic with parallel computing concepts (i.e., when using the spmd construct). Ideally, it would be desirable instead to decouple the concerns between domain experts and runtime/parallel computing experts, with the former focused on algorithms and application logic, with the latter providing suitable tools to efficiently distribute the computations on available resources and to provide optimal performance in a transparent way. Conversely, what often happens is that the domain expert models its application at a small scale, then the code is ported at large scale by an HPC expert, doubling the development effort. In this context, it is possible to look at the rise of Python in the data science field as a virtuous example: its low access barrier and open ecosystem led to the development of many tools that aimed to achieve the much-needed convergence between the goals of domain experts and the need for performance and scalability. These tools build upon existing frameworks (e.g., PySpark [17]), or implement task parallelism facilities from scratch in pure Python (e.g., Dask [3,11]), and provide a syntax that is reasonably close to the habits of domain experts, while trying to hide most of the complexity related to parallelism.
202
P. Viviani et al.
The goal of this work is to present a successful experience in leveraging Python scientific computing ecosystem to port a MATLAB application to an HPC infrastructure, reporting initial results and collecting useful lessons learned, in order to provide guidance for domain experts seeking the best performance for their numerical applications. This paper is structured as follows: Sect. 2 presents the target application and its mathematical components, their theoretical background and computational complexities, and the current scenario of software tools that are suitable to handle this problem in the context of the broader ecosystem, discussing some technical features and the reasons for their use. Section 3 introduces the current approach to MATLAB HPDA and introduces the proposed implementation, discussing the advantages over the baseline. In Sect. 4, are presented the outcomes and quantitative results and, finally, in Sect. 5, are outlined the main lessons learned.
2
Background
The motivation for this work is related to a high-fidelity Large Eddy CFD Simulation (LES) [13] for aerospace applications that is performed iteratively in the context of a design space exploration: each run of this simulation produces a significant amount of data (several terabytes, depending on how frequently the status of the flow field is sampled) that should be processed in order to extract relevant knowledge and to drive the following iterations of the design exploration. This paper will discuss specifically how the analytics applied to the flow field data can be ported to a proper HPC infrastructure, as they are a good representative for a wide class of analytics problems. It should be noted that, while Sect. 4 will report some scalability results achieved for this task on real data, this represents only a subset of the whole analysis of the CFD data. Details of the operations involved in each step are given below. 2.1
Analytics Pipeline
Figure 1 represents the specific instance of the analytics problem at hand. It is possible to identify three main steps of the process: data ingestion, data reshuffling and an intensive floating-point computation (the matrix factorization. The execution of the three steps is also influenced (and sometimes dominated) by I/O operations on the underlying file system. Below are described in detail the three main steps of the analytics pipeline. 1. Data ingestion: this step involves reading a large number (450 in the specific instance of the problem, but potentially multiple thousands for production setups) of CSV files. The rows in the CSV files correspond to the mesh sites coming from a spatial discretisation of the instance’s domain (∼21 M rows, totalling 3 GB for each CSV file). From each file, the three components of the flow velocity are extracted and stored separately.
Accelerated Analytics: From MATLAB to Python Data ingestion
Data reshuffling
203
Matrix factorization
u x 450 Snapshots of flow field (x 450)
u Velocities (u,v,w)
X*
v x 450
v
SVD
~21M rows
w x 450
w
3x21M 450
Parallel file system
Fig. 1. Abstract representation of the data analytics problem considered in this work. [u, v, w] represent the three components of the velocity of the flow field. Velocities are sampled periodically by the CFD simulation and dumped into CSV files (snapshots).
2. Data reshuffling: The velocity components extracted from the CSV files are stacked horizontally into 3 tall-and-skinny matrices containing each all the snapshots of one component of the velocity. The three matrices are then stored to disk, then they are stacked vertically to create a single tall-andskinny matrix. 3. Matrix factorization: a Singular Value Decomposition (SVD) is computed on this matrix, in order to obtain relevant information about the simulation data. The singular values and the singular vectors are stored on disk. This kind of pipeline captures different interesting aspects concerning HPDA problems; it is indeed worthy to note that the data ingestion and reshuffling is not as trivial as reading several CSVs and appending them as additional rows, depending on the specific implementation, this may involve transposing data structures from colum-major to row-major ordering and vice-versa. The SVD is also a good benchmark for distributed matrix computations beyond the typical complexity O(n) algorithms that are mostly used in the context of large-scale, distributed analytics [16]. The particular instance of SVD considered here is the so-called tall-and-skinny case, where the input matrix with M rows and N columns has M N . Its complexity, that for the general case of the SVD is O M 2 N + N 3 , here is dominated by the O M 2 term [5]. From the perspective of computational challenges it should be noted that keeping in memory the full matrix of ∼63 M × 450 double precision floating point values needs more than 220 Gb of memory, and this represents a small instance of the problem. This goes beyond what is feasible on a typical workstation, and also beyond most conventional HPC nodes, meaning that an out-of-core approach is necessary to handle this problem (either spilling data on disk, or going distributed). Also leveraging big memory HPC nodes can be feasible, but
204
P. Viviani et al.
this approach is still constrained by the size of available memory, which can become an issue when considering larger instances of the same problem. 2.2
Available Tools
The rationale of this work is to show how domain experts can leverage Python’s open ecosystem of data science tools, to accelerate their engineering analytics pipelines on HPC infrastructures without having to focus too much on parallel computing aspects. In order to achieve this goal, investigation has been carried out to identify the suitable tools that could satisfy the following requirements: 1. Simple, textbook-like syntax for linear algebra and mathematics in general. 2. Capability to scale out computations without modifying the application logic. 3. Optional: capability to exploit GPUs for parts of the computation. NumPy [6] and Pandas [8,15] emerged in the last decade as de facto standards for array and data manipulation in Python; their simple APIs, abundant documentation and large user base satisfies the first requirement and provide a starting point to convert MATLAB code to Python. On the other hand, while both rely on fast numerical libraries like BLAS [4] and LAPACK [1] to perform numerical computations, none of them is natively able to distribute the computation and the data among multiple nodes. Hence, further investigation has been necessary to identify the right tool to extend NumPy and Pandas capabilities beyond the shared-memory domain. 2.2.1
Distributed Computing with Python
The catalogue of tools to distribute computations with Python is quite large and spans the whole spectrum from low-level, message-passing tools (e.g., mpi4py [2]), to a host of task-based parallelism frameworks (e.g., Dask [3], PySpark [17], Ray [10], MARS [7], PyCOMPS [14]), up to extremely domain-specific libraries to parallelise specific tasks like distributed neural network training (e.g., ad-hoc facilities provided by Tensorflow and Pytorch, as well as Horovod [12]). This work is not meant to be a comprehensive review of distributed computing tools for Python, hence a formal architectural analysis of the available tool is out of the scope of this paper. On the other hand, it is useful to report the results of the scouting performed in order to port the analytics pipeline described in Sect. 3.1. From this perspective, the critical assumption was to be able to simply write the application logic with NumPy/Pandas, then the second requirement dictates that out-of-core scalability should be achieved with minimal modifications with respect to that implementation, ruling out all the low-level approaches that would require the application developer to both know the details of the analytics algorithm and to be familiar with concepts of distributed computing. However, among the tools listed above, some exist that advertise exactly the kind of seamless distributed NumPy/Pandas functionality needed for this use-case.
Accelerated Analytics: From MATLAB to Python
205
Dask, PySpark and MARS all provide some flavour of this concept built on top of graph task-based parallelism. Among them, the combination of PySpark and MLlib [9] has already been tested in a previous work with a similar rationale [16] with promising results, however the complexity of the cluster setup and the features proposed by the other candidates driven the choice away from this option. In fact, Dask and MARS advertise direct interoperability of data structures (NDarrays and DataFrames) with NumPy and Pandas, along with simple cluster set-up and native python implementation. Between the two, Dask has been preferred due to the maturity of the project and the large amount of documentation available. While, the MARS project is definitely worthy of consideration, especially given the interesting architectural choices (i.e., an agent-based model instead of the master-worker scheduler used by Dask) that may provide a performance edge, the goal of this work is not necessarily ultimate performance, but rather the best tradeoff of performance and overall usability of the tools.
3 3.1
Proposed Methodology Baseline Implementation
Prior to this work, the pipeline was implemented with a MATLAB code making heavy use of the spmd construct to distribute the computation among workers living on multiple computational nodes. Figure 2 reports the process as implemented with MATLAB: the CSV files are read by different workers in an spmd loop, then data is manually reshuffled in order to distribute rows among workers instead of columns. It is worth noting that the formulation with the eigenvectors of M T M is equivalent to the one described in the previous paragraph, using Columns are distributed among nodes M1T * M1
M1
u
M2 M2T * M2
Multiple workers
v M3
Sum
MTM
Eigenvectors
X
M3T * M3
....
... .....
w
Mn
MnT * Mn
Rows are distributed among nodes
Parallel file system
Fig. 2. MATLAB implementation of the analytics pipeline. Data is distributed among workers manually using MATLAB primitives.
206
P. Viviani et al.
SVD. This implementation uses the HDF51 to store the intermediate matrices and the eigenvectors. 3.2
Dask Implementation
Since the original MATLAB code implementation of the application logic was heavily interleaved with code related to the parallel implementation (i.e., explicit segmentation of the data, handling of remainders when the data size is not multiple of the number of workers, etc.), the first step has been to re-write the abstract logic presented in Fig. 1 with NumPy and Pandas. This has been accomplished with minimal effort and represented the starting point for the Dask implementation, as well as a benchmark for the development effort to port sequential code. Before diving into the details of the proposed approach, it is worth to review the architecture of the Dask runtime. 3.2.1
Dask Overview
Dask emerged as a tool to implement distributed task graphs with Python and, as such, maintained its task-based parallelism nature also when higher-level concepts like distributed NDarrays were included. Its conceptual architecture is depicted in Fig. 3: the user code runs as the client, which defines the task graph to be executed, and sends it to the scheduler, which in turn assigns individual tasks to the workers. Dask workers can also communicate with each other if the algorithm requires so, possibly leveraging high-performance MPI implementations and networking (e.g., Infiniband). The processes for the client, the schedule and each worker can be started individually via command line interface or by means of facilities provided by the Dask distributed package that can leverage cluster management tools (i.e., ssh, Kubernetes, SLURM, Docker, PBS and also plain mpirun). The rendezvous between all the components, if launched manually can be achieved by pointing to a JSON file created by the scheduler.
DASK Client
DASK Scheduler
Cluster DASK Worker
DASK Worker
DASK Worker
DASK Worker
Distributed in-memory data structure
Fig. 3. Dask architecture.
1
https://www.hdfgroup.org/solutions/hdf5/.
DASK Worker
Accelerated Analytics: From MATLAB to Python
207
If the application logic can be described only using the native Dask data structures and their related primitives (e.g., dask.array, dask.dataframe), then the task graph is defined by Dask itself and it is completely hidden behind the scenes. Otherwise, the user can either access facilities for lazy evaluation of functions, built on the dask.delayed concept, or he can directly manipulate the task graph. 3.2.2
Dask Cluster
This work involved running the HPDA pipeline on the Galileo100 HPC cluster provided by CINECA2 , composed by 528 computing nodes, each with 2 Intel CascadeLake 8260 CPUs with 24 cores each, and 384 GB of memory. Some of the nodes were also equipped with 3TB Intel Optane storage configured as additional system memory. The cluster is managed by the SLURM job scheduler3 .
1 2 3 4 5
#!/bin/bash #SBATCH --nodes=8 #SBATCH --ntasks-per-node=4 #SBATCH --cpus-per-task=12 #SBATCH
6 7 8
9
dask-scheduler --scheduler-file /path/to/scheduler/file & srun dask-mpi --interface ib0 --nthreads 12 --memory-limit --scheduler-file /path/to/scheduler/file --no-scheduler -worker-class distributed.Worker & python client_logic.py /path/to/scheduler/file
Listing 1. Batch script to launch Dask cluster on a SLURM-managed HPC facility.
To run Dask distributed on HPC facilities it is possible either to specify the options of the available job scheduler from the client code using Dask-jobqueue4 . Otherwise, it is possible to launch the jobs using SLURM job scripts; the latter approach has been followed in this case, as it allows the re-use of existing job script configurations and gives more control on the deployment of the components. However, the configuration described in listing 1 deserves some clarification. The scheduler process requires orders of magnitude less resources than the workers, hence if we allow Dask-distributed to launch the scheduler along with the workers, it will count against the number of jobs defined by SLURM and it will cause uneven load balancing; in this case, the scheduler is launched as a standalone process and the option --no-scheduler is specified to prevent Dask to automatically spawn a scheduler alongside the workers. Using dask-mpi to 2 3 4
https://www.hpc.cineca.it/hardware/galileo100. https://slurm.schedmd.com. https://jobqueue.dask.org/en/latest/.
208
P. Viviani et al.
launch the workers ensures that they will use a high-performance MPI library to communicate with each other, moreover, it is possible to force them to use the Infiniband network interface (--interface ib0). The memory limit for each worker should be defined taking into account the --ntasks-per-node=4 option value by dividing the physical node memory by it. The scheduler file is needed to facilitate the rendezvous between the workers, the client and the scheduler: it is important to include its path in the client logic as a parameter, to ensure it can connect to the scheduler. 3.2.3
Implementation Approach
The process in Fig. 1 requires ingesting a certain number of CSV files, which Dask is equipped to do out of the box (i.e., using dd.read csv(‘/path/*.csv’)), however, its API is only meant for appending homogenous rows to a single dataframe. In this case, the task involves extracting columns and appending them in order. This required a dedicated data ingestion process, that has been implemented as a @dask.delayed function to leverage parallelism when reading files from a parallel filesystem. The delayed function extracts the relevant columns into a NumPy array that is then processed by the main as a dask.array.from delayed object where the three velocities are split and aggregated in three separate dask.array. These arrays are then stored to the filesystem as Parquet files5 , allowing for parallel writing of the serialized data structure. The three matrices are then stacked vertically, and dask.array.linalg.svd is applied to compute singular values/vectors. All these operations differ only slightly from the plain NumPy/Pandas counterparts: in fact, only the delayed-based data ingestion differs from the original implementation as a concept, while the Pandas code is almost identical.
4
Results
All the tests have been performed on the HPC cluster presented in Sect. 3.2.2. Table 1 reports the results achieved with the presented implementation of the HPDA process presented in Sect. 2.1. It is worth mentioning that the timing of the MATLAB implementation has been provided by the end-user as an order of magnitude comparison, but it has been tested on the same cluster as the rest of the tests performed. The NumPy/Pandas results refer to the fully sequential rewriting discussed at the beginning of Sect. 3 and the tests have been executed on the Optane-equipped, large-memory nodes of the same Galileo100 cluster. From the numbers in Table 1, it is clear how the advantage of the Dask implementation vs. the MATLAB is not limited to the ecosystem or to the ease of use, but it is also related to the significant improvement of performance. This improvement can be related to several factors, with efficient parallel I/O one of the most impactful (i.e., faster read of CSV files, extremely fast write speed of Parquet file format w.r.t. the HDF5, which must be written sequentially). 5
https://parquet.apache.org.
Accelerated Analytics: From MATLAB to Python
209
Table 1. Time required to complete HPDA process. - not performed or not applicable; *order of magnitude, reported for comparison. Nodes MATLAB (s) NumPy/Pandas (s) Dask (s) 1 2 4 8 16
– 1 h 20 m* – – –
3 h 09 m – – – –
– 23 m 08 m 06 m 03 m
30 s 44 s 02 s 44 s
Finally, Fig. 4 presents the scalability results achieved by the Dask implementation: the results are quite close to the ideal trend up to 16 nodes. Moreover, it is interesting to see how 4 nodes provide superlinear speedup with respect to 2 nodes: it is the author’s opinion that the cause is related to the larger total amount of memory of 4 nodes (with constant problem size) that allows some Dask operation to avoid spilling to disk. Dask scalability
12
Scalability
10 8 6 4 2 2
4
6
8
10
12
14
16
Nodes
Fig. 4. Scalability of the Dask implementation. Grey line represents ideal trend.
5
Conclusion and Future Work
The approach and results presented so far provide a successful case-study in support of the idea of using Python tools to perform engineering data analytics on HPC infrastructures. In particular, Dask satisfied all the requirements mentioned in Sect. 2.2, proving to be capable of delivering most of what its developers advertise: the considered NumPy/Pandas application could be ported to a distributed cluster with small (but still some) modifications to the code. As a bonus, the presented implementation outperformed significantly the original MATLAB code, while sporting good scalability up to the largest test set-up (16 nodes). In future work, it is expected that two main lines of investigation will be built on top of this work: first, other steps of the CFD data analytics pipeline will be implemented with Dask in order to provide results that are more significant for the end users. Moreover, the feasibility and the efficiency of accelerating some parts of the pipeline on GPU will be investigated.
210
P. Viviani et al.
Acknowledgements. This work has been supported by the EuroHPC-02-2019 ACROSS project, grant agreement No. 955648 and by PRACE, which awarded access to the Fenix Infrastructure resources at CINECA, partially funded from the European Union’s Horizon 2020 research and innovation programme through the ICEI project under the grant agreement No. 800858.
References 1. Anderson, E., et al.: LAPACK Users’ Guide, 3rd edn. Society for Industrial and Applied Mathematics, Philadelphia, PA (1999) 2. Dalc´ın, L., Paz, R., Storti, M.: MPI for python. J. Parallel Distrib. Comput. 65(9), 1108–1115 (2005). https://doi.org/10.1016/j.jpdc.2005.03.010 3. Dask Development Team: Dask: Library for dynamic task scheduling (2016). https://dask.org 4. Dongarra, J., Du Croz, J., Hammarling, S., Hanson, R.J.: An extended set of FORTRAN basic linear algebra subprograms. ACM Trans. Math. Softw. 14(1), 1–17 (1988). https://doi.org/10.1145/42288.42291 5. Golub, G.H., Van Loan, C.F.: Matrix Computations, 3rd edn. Johns Hopkins University Press, Baltimore, MD, USA (1996) 6. Harris, C.R., et al.: Array programming with NumPy. Nature 585(7825), 357–362 (2020). https://doi.org/10.1038/s41586-020-2649-2 7. MARS development team: MARS. https://docs.pymars.org/en/latest/ 8. McKinney, W.: Data structures for statistical computing in python. In: Van Der Walt,S., Millman, J. (eds.) Proceedings of the 9th Python in Science Conference, pp. 56–61 (2010). https://doi.org/10.25080/Majora-92bf1922-00a 9. Meng, X., et al.: MLlib: machine learning in apache spark. J. Mach. Learn. Res. 17(1), 1235–1241 (2016) 10. Moritz, P., et al.: Ray: a distributed framework for emerging AI applications. In: Proceedings of the 13th USENIX Conference on Operating Systems Design and Implementation, OSDI 2018, pp. 561–577. USENIX Association, USA (2018) 11. Rocklin, M.: Dask: parallel computation with blocked algorithms and task scheduling. In: Huff, K., Bergstra, J. (eds.) Proceedings of the 14th Python in Science Conference, pp. 130 – 136 (2015) 12. Sergeev, A., Balso, M.D.: Horovod: fast and easy distributed deep learning in TensorFlow. arXiv preprint arXiv:1802.05799 (2018) 13. Smagorinsky, J.: General circulation experiments with the primitive equations: I. the basic experiment. Monthly Weather Rev. 91(3), 99–164 (1963). https://doi. org/10.1175/1520-0493(1963)0910099:GCEWTP2.3.CO;2 14. Tejedor, E., et al.: PyCOMPSs: parallel computational workflows in python. Int. J. High Perform. Comput. Appl. 31(1), 66–82 (2017). https://doi.org/10.1177/ 1094342015594678 15. The pandas development team: pandas-dev/pandas: Pandas (2020). https://doi. org/10.5281/zenodo.3509134 16. Viviani, P., Drocco, M., Aldinucci, M.: Scaling dense linear algebra on multicore and beyond: a survey. In: Proceedings of the 26th Euromicro International Conference on Parallel Distributed and Network-based Processing (PDP). IEEE, Cambridge, United Kingdom (2018) 17. Zaharia, M., et al.: Apache spark: a unified engine for big data processing. Commun. ACM 59(11), 56–65 (2016). https://doi.org/10.1145/2934664
An Adaptive Resource Allocation Protocol for Dynamic Environments Mojtaba Malek-Akhlagh(B) and Jens H. Weber Department of Computer Science, University of Victoria, Victoria, Canada {mojmalek,jens}@uvic.ca Abstract. Resource allocation in a dynamic environment deals with the problem of how participants coordinate in allocating a set of resources among each other, while the environment dynamically changes over time. In this paper, we investigate a variant of this problem in which participants perform tasks which require multiple items of different resource types, while the tasks and resources arrive over time. For example, a network of hospitals may need to coordinate in allocating resources such as blood products, while resource supply and demand change over time. We model this problem as a multi-agent coordination problem, and propose a decentralized, self-organized approach, in which all decisions are made by individual agents and there is no central control. We present an adaptive negotiation-based protocol, which enables an agent to have concurrent negotiations for different resources types, and to combine resource contributions from multiple provider agents. We evaluate our proposed protocol by developing simulation models. Our simulation results illustrate its scalability and efficiency.
1
Introduction
Resource allocation problems consist of a set of participants and a set of resources. The participants require resources in order to gain utility. An optimal solution to a resource allocation problem is an allocation assigning a subset of resources to each participant such that the total utility gained by all participant is maximized. This problem becomes more challenging when the participants act in a dynamic environment, where the set of resources or participants changes over time. As the environment evolves, a mechanism which finds the best possible allocation needs to adapt to the changes by adjusting the solution, i.e. reallocating the resources. In this study, we investigate a variant of the dynamic resource allocation problems, in which the participants require multiple items of different resource types in order to perform their assigned tasks, while the resources and tasks arrive over time. An instance of such problems can be found in the allocation of required resources to hospitals distributed in a region, where the dynamism comes from sudden changes in demand such as an increase in the number of patients appearing, and hence a shortage of required resources such as blood products, which are scarce and perishable. Another example is in the distribution c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 211–222, 2022. https://doi.org/10.1007/978-3-031-08812-4_21
212
M. Malek-Akhlagh and J. H. Weber
of scarce and perishable vaccines among distributed health authorities when the supply and demand changes over time. We model the distributed resource allocation problem as a multi-agent coordination problem. Distributed intelligent systems or multi-agent systems [14] are increasingly employed in practical solutions, in which automated coordination among the participants is crucial in achieving the system objectives. We represent each participant with an autonomous software agent, which is responsible for allocation of required resources. Multi-agent systems can incorporate self-organisation mechanisms [15], which may lead to a common improvement. Self-organisation is a localized and decentralized operation, which allows a distributed system to be more scalable and highly adaptive to dynamic requirements [5]. Serugendo et al. [11] defined self-organization and strong self-organizing systems as follows. Definition 1. (Self-Organization): “ is the mechanism or the process enabling a system to change its organization without explicit external command during its execution time [11].” Definition 2. (Strong Self-Organizing Systems): “are those systems where there is no explicit central control either internal or external [11].” In this paper, we present a negotiation-based protocol, which allows a multiagent system in a dynamic environment to adjust its resource allocation during its execution time. The proposed protocol allows an agent in need of resources to have concurrent negotiations for different resources types, and to receive contributions with different quantities from multiple provider agents. Moreover, the protocol allows the multi-agent system to be strongly self-organized without relying on a central entity either internal or external. Hence, the system avoids a single point of failure in dynamic environments with possibility of agent failure. We evaluate the scalability and efficiency of the proposed protocol by performing simulation experiments. We compare its performance with a centralized approach, and with a basic Contract Net-based protocol [13] for different number of agents, resource types, and resource quantities. The remainder of the paper is organized as follows. Section 2 reviews the related work. Section 3 formally defines the dynamic resource allocation problem. Section 4 presents the proposed protocol specifications. Section 5 presents our simulation models and experiment results. Finally, Sect. 6 gives concluding remarks and outlines future directions.
2
Related Work
The Contract Net Protocol (CNP) [13] presents an auction-based approach for decentralized task allocation problem in a multi-agent system. As described by the Foundation for Intelligent Physical Agents (FIPA) [7], CNP allows a manager and participants to interact through a negotiation process. Initially, the manager
An Adaptive Resource Allocation Protocol for Dynamic Environments
213
initiates the process by sending a Call For Proposals (CFP) for a task specification. The participants who receive the CFP, evaluate it and decide whether to send a proposal or a refuse message. The CFP includes a deadline by which the manager starts evaluating all the received proposals and then awards a contract to the participant with the most appropriate proposal, and rejects all the other proposals. The participant who received the contract starts working on the task and eventually sends a completion message or a failure message to the manager. Many researchers have proposed extensions to the original CNP [2]. Fatima and Wooldridge [6] developed a self-organizing multi-agent system called TRACE, which allows a collection of organizations to cooperate in dynamic environments. TRACE includes a task allocation protocol (TAP) and a resource allocation protocol (RAP). TAP is a Contract Net-based protocol, which enables the individual agents in an organization to allocate tasks among themselves based on their capabilities and local schedules. In a dynamic environment, each organization arbitrarily receives requests for different tasks, so that one organization could have shortage of resources, while another one could have extra resources. RAP allows the organizations to reallocate resources between themselves in accordance with their demands. RAP uses ideas from computational market systems, by calculating the total supply and demand for resources in the system and determining the equilibrium price. Polajnar et al. [10] proposed the Mutual Assistance Protocol (MAP), which enables an agent team to incorporate helpful behavior in performing actions (Action MAP) and providing resources (Resource MAP). The interaction sequence in MAP is similar to the one in the Contract Net protocol. The deliberation on helpful behavior is based on the interest of the team, and is jointly determined by two agents through a negotiation process. A team is given a task, with each agent addressing a subtask. An agent performs actions in order to accomplish its subtask. Individual agents have different abilities with respect to performing actions, hence they can help each other when need arises. Each individual agent estimates the team impact of helpful behavior to its own individual plan. Further extensions of Action MAP allowed both requester-initiated and helper-initiated versions [9], and combining the two approaches in one bidirectional protocol [8]. De Weerdt et al. [3,4] studied a variant of the task allocation problem, where the agents are connected in a social network. In their model, the agents are only allowed to interact with their direct neighbors in the network. Each agent is given a set of tasks to perform, and has a set of resources of different types. Each task requires some resources, and has a fixed benefit. Each agent is only allowed to use resources provided by its neighbors. The problem is to find out which tasks to perform, and which resources of which neighbors to use in order to maximize the total benefit for all agents. They proved that the problem is NP-hard for an arbitrary graph, and proposed a greedy distributed protocol based on the Contract Net protocol, in which an agent in charge of a task acts like a manager and its neighbors act like participants.
214
M. Malek-Akhlagh and J. H. Weber
In this study, we extend the Contract Net protocol for handling concurrent negotiations, in which an agent in need of resources is allowed to select multiple providers and combine their resource contributions. Our work also extends the existing protocols in TRACE [6] and MAP [10] as we study the resource allocation problem when tasks require multiple types of resources. It also contrasts with the work by De Weerdt et al. [3,4] as we investigate the resource allocation problem in dynamic environments, where both sets of tasks and resources may change over time.
3
Problem Definition
We consider a multi-agent system comprising of n agents, A = {a1 , . . . , an }, n > 1. Each agent a ∈ A receives sets of tasks over time defined by a task assignment function: A × T ime → P (T ), where P (T ) denotes the power set of task domain, T = {t1 , . . . , tm }, m > 0. Each task t ∈ T requires resources defined by a requirement function: T × R → N, where R = {r1 , . . . , rl }, l > 0 is set of resource types. Each task has a utility indicating its importance; and has a deadline, which indicates the latest time at which the task could be carried out. Each agent a ∈ A also procures resources over time defined by a resource procurement function: A × R × T ime → N. Each resource item has a lifetime, which indicates a duration in which the item can be used. The environment in which the agents are operating is considered to be dynamic in two ways: the tasks may arrive at different rates over time; also, the resources may be procured at different rates over time. We model the progression of time by the natural numbers, N, and assume that all agents have access to a global clock. The knowledge regarding tasks and resources is local, and each agent decides on which tasks to perform given its available resources at each time step. The remaining tasks and resources are carried over to the next time step. The agents are able to communicate, share information, and reallocate resources among each other. Each agent gains utilities by performing tasks over time. We define social welfare of the multi-agent system as sum of utilities gained by all agents, t t util a , where utila is sum of utilities gained by agent a up to time step a∈A t. Now, given a set of agents A, and sets of all tasks and resources assigned to them at time step t, the problem is to find out which tasks of which agents to perform, and which resources of which agents to use for these tasks, in order to maximize the social welfare. The complexity of finding the optimal solution for this problem is NP-hard [12], which comes from the exponential number of subsets of the set of tasks assigned to the agents.
4
Proposed Protocol
We propose a decentralized adaptive negotiation-based protocol for the resource allocation problem in dynamic environments. The protocol has the selforganizing property as all the decisions are made by individual agents and there
An Adaptive Resource Allocation Protocol for Dynamic Environments
215
is no central controller either internal or external. It allows concurrent negotiations in which an agent in need of resources is enabled to select multiple providers and combine their resource contributions. A high-level description of the protocol is as follows. At each time step, when an agent ai has a set of tasks to do, it checks if it can locally allocate the required resources to fully accomplish them, otherwise it considers creating a request based on the missing quantity for each resource type and sends them to other agents. An agent aj who receives a request decides whether it can fully or partially provide the requested quantity. In that case, it considers sending an offer to ai . Since multiple agents can offer to a request, ai may receive multiple offers. In such case, it confirms a combination of them which maximizes the social welfare. In the following, we present the specification of the three protocol phases: requesting, offering, and confirming. 4.1
Deliberate on Requesting
Agent ai evaluates its current set of tasks using a greedy approach. It sorts the tasks based on a heuristic called task efficiency, defined as the ratio between task utility and its total number of required resources. Then it finds the tasks that are blocked because of lack of resources, and creates a request based on missing quantity for each resource type. A request includes a utility function: U : N −→ N computed as utility gains for partial quantities in {q : 0 < q ≤ Q}, where Q is the missing quantity. It indicates benefit of partial contributions, and allows the other agents to offer a partial amount, q. Algorithm 1 formalizes the requesting process of ai . A request message sent by ai has the following format: Request(r, Q, U ) where Q indicates quantity requested for resource type r, and U : N −→ N is the request utility function.
Algorithm 1: Requesting process Input: current set of tasks assigned to ai : Tai , set of resource types: R, requirement function: req Output: requests sent by ai : requestSet foreach r in R do requiredQuantity ← total required quantity of r in Tai availableQuantity ← available quantity of r if availableQuantity < requiredQuantity then Q ← requiredQuantity - availableQuantity U ← compute request utility function as utility gains for partial quantities in {q : 0 < q ≤ Q} requestSet ← create a request end end
216
4.2
M. Malek-Akhlagh and J. H. Weber
Deliberate on Offering
An agent aj may receive multiple requests from different requester agents for a resource type r. In a greedy approach, while there are available resources, aj considers offering to the request with highest efficiency. The request efficiency is defined as the ratio between its utility and requested quantity. The offerer aj is allowed to offer partially, and determines the amount of resources it can provide, q, up to the requested amount, Q. It deliberates about offering to each request by computing the cost of offering, defined as a function: C : N −→ N. The offer cost function is computed as utility losses for partial quantities in {qp : 0 < qp ≤ q}. It indicates cost of partial contributions by the offerer. The offer cost for a quantity, q, is computed as the loss of utility at the current time step. If such cost is zero; i.e. the requested resources are not needed right now, aj estimates an expected cost for future based on its own history of tasks defined as follows. Definition 3. (Expected Utility of Resources): Expected utility of quantity q of resource type r for agent a is a function, exp : N × R × A → N, defined as: exp(q, r, Ta ) = q(|Tr |/|Ta |)[( t∈Tr utilt )/( t∈Tr qrt )], where Ta is set of all tasks performed so far by a, Tr is a subset of Ta which required r, utilt is the gained utility of each task t ∈ Tr , and qrt is quantity of resource r used in performing t. Note that the expected utility exp is only computed for the resource items which have a lifetime greater than one time step. The offerer aj compares its offering cost of quantity q: C(q), with utility of the request for the same quantity: U (q). Eventually, aj sends an offer if C(q) < U (q). The offering process is formalized in Algorithm 2. An offer message sent by aj has the following format: Algorithm 2: Offering process Input: requests received by aj : requestSet, resource type: r Output: offer sent by aj : offer availableQuantity ← available quantity of r while (availableQuantity > 0 and requestSet is not empty) do selectedRequest ← request with highest efficiency in requestSet Q ← requested quantity in selectedRequest if availableQuantity > Q then q←Q end else q ← availableQuantity end C ← compute offer cost function as utility losses for partial quantities in {qp : 0 < qp ≤ q} if C(q) < U(q) then offer ← create an offer update availableQuantity end remove selectedRequest from requestSet end
An Adaptive Resource Allocation Protocol for Dynamic Environments
217
Offer (r, q, C, items) where q indicates maximum quantity offered for resource type r, C : N → N is the offer cost function, and items : N → N shares information regarding lifetimes of offered resource items by mapping item id to its lifetime. 4.3
Deliberate on Confirming
The requester, ai , may receive multiple offers from multiple providers for a resource type r. In that case, ai selects the combination of offers that maximizes the difference between utility of the request and total cost of all the selected offers. Note that ai is allowed to take partial amounts of offered resources in multiple offers up to the requested amount, Q. Formally, suppose there are n offers. ai selects m ≤ n offers, with a quantity dk ≤ qk from each selected offer k ∈ {1, ..., m} in order to: maximize Ui (
m
k=1
subject to
m
k=1
dk ) −
m k=1
Ck (dk ) (1)
dk ≤ Q
Furthermore, if ai has sent multiple requests for different required resource types for a set of tasks Tai , it waits for offers to all the requests and then confirms the offers only if the utility gain of confirming is greater than the total cost of all selected offers; i.e. there is still benefit in performing at least a subset of Tai . For instance, if there is no offer received for one requested resource type and it is not possible to perform any task in Tai , then ai rejects all offers received for all the other requested resource types. We formalize the offer confirmation process in Algorithm 3. A confirmation message sent by ai for each selected offer has the following format: Confirm(r, d) where d ≤ q indicates partial amount confirmed for resource type r. A reject message has the same format with d = 0. When an offerer agent receives a confirmation, it allocates the confirmed resources to the requester. Proposition 1. For n agents and r resource types, the message complexity of the protocol at each time step is O(rn2 ). Proof. In the worst case, at each time step, each agent lacks all resource types, then there are rn(n−1) request messages, rn(n−1) offer messages, and rn(n−1) confirmation messages. Thus, the total number of sent messages is 3rn(n − 1).
218
M. Malek-Akhlagh and J. H. Weber
Algorithm 3: Confirming process Input: requests sent by ai : requestSet, offers received by ai : offerSet Output: confirmations sent by ai : confSet, rejections sent by ai : rejectSet selectedQuantitiesForAllRequests ← ∅ foreach request in requestSet do if offerSet includes offer(s) for request then selectedQuantities ← select a quantity from each offer using Eq. (1) selectedQuantitiesForAllRequests ← selectedQuantities end end if selectedQuantitiesForAllRequests is not empty then utility ← utility gain of confirming selectedQuantitiesForAllRequests totalCost ← total cost of confirming selectedQuantitiesForAllRequests if utility > totalCost then foreach request in requestSet do confSet ← confirm selectedQuantities for request rejectSet ← reject remaining offers for request in offerSet end end else rejectSet ← reject all offers in offerSet end end
5
Evaluation
In order to evaluate our proposed protocol, we have developed an abstract simulator in Java using the JADE framework [1]. In our simulation model, time proceeds in synchronous rounds for all agents. In every round, each agent: i) receives tasks and resources, ii) negotiates resources using a protocol, iii) performs tasks. Remaining tasks and resources are carried over to the next round; and the knowledge regarding tasks and resources is local. We have developed a centralized agent as a basis for evaluating the scalability and efficiency of our decentralized protocol. The centralized agent receives the same tasks and resources for all agents in each round in order to solve the same resource allocation problem. We have used the same greedy method in selecting tasks based on the task efficiency heuristic. In our experiments, we measure the percentage ratio between the social welfare result computed by the decentralized protocol and the centralized agent: 100 ∗ socialWelfaredec /socialWelfarecen We simulate a dynamic environment by generating random parameters as follows. In every round, each agent receives a number of tasks randomly chosen from the set {1, 2, 4, 8}. There are four different types of resources: {A, B, C, D}. Each task requires a quantity randomly chosen from {0, 1, 2, 4, 8} for each resource type. Task deadline is 20 rounds, and task utility is randomly chosen from {1, 2, 4, 8, 16, 32, 64}. When an agent performs a task by allocating its required resources, it gains the task utility. We run our experiments for 1000 rounds.
An Adaptive Resource Allocation Protocol for Dynamic Environments
219
In our first experiment, we evaluate the scalability of our decentralized protocol by increasing the number of agents. The resource procurement rate is as follows. Each agent receives resources with a quantity randomly chosen from {0, 1, 2, 4, 8} for each resource type in every round. Each resource item lifetime is three rounds. We vary the number of agents from 4 to 64 in our experiment runs. Table 1 presents the results, and shows that our decentralized protocol scales up as the number of agent grows. Table 1. Scalability of proposed protocol considering number of agents Number of agents
4
8 16 32 48 64
Percentage ratio on social welfare 86 78 78 77 77 76
In order to further evaluate the efficiency of our proposed adaptive protocol, we compare it with a basic protocol, which is similar to the standard Contract Net protocol. In the basic protocol, a requester agent does not combine multiple offers. It is only allowed to confirm one offer, the one with highest quantity, and all the other offers are rejected. Moreover, when the requester has sent multiple requests for different required resource types, it does not wait for offers to all the requests, and does not check if the utility gain of confirming the selected offers is greater than their total cost. In our second experiment, we study the impact of shortage of resources. There are eight agents, while the resource procurement rate is different among them. Four of them are considered as agents with shortage of resources. In every round, each of them receives resources with a quantity randomly chosen from {0, 1, 2} for each resource type. The other four agents are considered as the ones with more available resources. In every round, they receive a fixed quantity for each resource type, varying from 2 to 10 in our experiment runs. The other simulation parameters are the same as in the first experiment. Figure 1 illustrates simulation results for comparing the proposed adaptive protocol with the basic protocol. We can observe that the adaptive protocol outperforms the basic protocol in all the resource quantity range. Also, the minimum percentage ratio of 70% shows that our decentralized, self-organized approach is still effective, even when the resources are scarce. In our third experiment, we study the impact of increasing number of resource types. There are eight agents, four of them receive resources with a quantity randomly chosen from {0, 1, 2}, and the other four agents receive a fixed quantity of four items for each resource type in every round. We vary the number of resource types from 2 to 10 in our experiment runs. The other simulation parameters are the same as before. Figure 2 illustrates the simulation results. It can be observed that the adaptive protocol outperforms the basic protocol for all the numbers of resource types; and the adaptive protocol declines relatively less than the basic protocol when the number of resource types increases. Moreover, the minimum percentage ratio of 60% shows that our decentralized, self-organized approach is still effective, even when the number of resource types is high.
220
M. Malek-Akhlagh and J. H. Weber
Fig. 1. Efficiency of proposed protocol for different resource quantities
Fig. 2. Efficiency of proposed protocol when number of resource types increases
6
Conclusions and Ongoing Work
In this paper, we have presented a decentralized adaptive protocol, which enables individual agents to deliberate on receiving and providing resources by engaging in a negotiation mechanism. The protocol allows the agents to coordinate and combine their resource contributions when incoming tasks require multiple items and different types of resources in a dynamic environment. It allows a multiagent system to be self-organized, in which all the decisions are made through local interactions among individual agents, without a central controller either internal or external. Hence, the system can avoid the single point of failure and is robust to failures in communication. Its adaptive behavior allows it to be efficient in dynamic environments where the set of tasks and resources may
An Adaptive Resource Allocation Protocol for Dynamic Environments
221
constantly change over time. Our simulation results illustrate that the proposed protocol is scalable when number of agents increases; and outperforms the basic Contract Net-based protocol for different values of resource types and quantities. We are extending our protocol design to address the dynamic resource allocation problem when the proximity and hierarchy among agents are determined by a social network. Our future work may allow agents to cascade their requests, and combine contributions from multiple offers along the network. Furthermore, it would be interesting to target one or more specific real-world problems such as blood donation, vaccination, or organ transplant; and investigate whether our proposed protocol is effective in real-world applications.
References 1. Bellifemine, F.L., Caire, G. and Greenwood, D.: Developing Multi-agent Systems with JADE, vol. 7. Wiley, Hoboken (2007) 2. Bozdag, E.: A survey of extensions to the contract net protocol. Technical report, CiteSeerX-Scientific Literature Digital Library and Search Engine (2008) 3. De Weerdt, M., Zhang, Y., Klos, T.: Distributed task allocation in social networks. In: Proceedings of the 6th International Joint Conference on Autonomous Agents and Multiagent Systems, pp. 1–8 (2007) 4. De Weerdt, M.M., Zhang, Y., Klos, T.: Multiagent task allocation in social networks. Auton. Agents Multi-Agent Syst. 25(1), 46–86 (2012) 5. Elmenreich, W., D’Souza, R., Bettstetter, C., De Meer, H.: A survey of models and design methods for self-organizing networked systems. In: Spyropoulos, T., Hummel, K.A. (eds.) Self-Organizing Systems. IWSOS 2009. LNCS, vol. 5918, pp. 37–49. Springer, Berlin, Heidelberg (2009). https://doi.org/10.1007/978-3-64210865-5 4 6. Fatima, S.S., Wooldridge, M.: Adaptive task and resource allocation in multi-agent systems. In: Proceedings of the Fifth International Conference on Autonomous Agents, pp. 537–544 (2001) 7. FIPA: FIPA Contract Net Interaction Protocol Specification. FIPA (2001) 8. Malek Akhlagh, M., Polajnar, J.: Distributed deliberation on direct help in agent teamwork. In: Proceedings of the 12th European Conference on Multi-Agent Systems (EUMAS 2014), Prague, Czech Republic, December 2014 9. Nalbandyan, N., Polajnar, J., Mumbaiwala, D., Polajnar, D., Alemi, O.: Requester vs. helper-initiated protocols for mutual assistance in agent teamwork. In: Proceedings of the 2013 IEEE International Conference on Systems, Man, and Cybernetics (SMC 2013), pp. 2741–2746, Manchester, UK, October 2013 10. Polajnar, J., Nalbandyan, N., Alemi, O., Polajnar, D.: An interaction protocol for mutual assistance in agent teamwork. In: Proceedings of the 6th International Conference on Complex, Interactive, and Software-Intensive Systems (CISIS 2012), pp. 6–11, Palermo, Italy, July 2012 11. Di Marzo, G., Serugendo, M.-P.G., Karageorgos, A.: Self-organization in multiagent systems. Knowl. Eng. Rev. 20(2), 165–189 (2005) 12. Shehory, O., Kraus, S.: Methods for task allocation via agent coalition formation. Artif. Intell. 101(1–2), 165–200 (1998) 13. Smith, R.G.: The contract net protocol: high-level communication and control in a distributed problem solver. IEEE Trans. Comput. 29, 1104–1113 (1980)
222
M. Malek-Akhlagh and J. H. Weber
14. Wooldridge, M.: An Introduction to Multiagent Systems. Wiley, Hoboken (2009) 15. Ye, D., Zhang, M., Vasilakos, A.V.: A survey of self-organization mechanisms in multiagent systems. IEEE Trans. Syst. Man Cybern. Syst. 47(3), 441–461 (2016)
FPGA Implementation of an Object Recognition System with Low Power Consumption Using a YOLOv3-tiny-based CNN Yasutoshi Araki(B) , Masatomo Matsuda, Taito Manabe, Yoichi Ishizuka, and Yuichiro Shibata Nagasaki University, Bunkyo-machi 852-8521, Japan {yaraki,matsuda}@pca.cis.nagasaki-u.ac.jp, {tmanabe,isy2}@nagasaki-u.ac.jp, [email protected]
Abstract. Although the use of FPGAs for embedded-oriented CNN accelerators has been spreading owing to a high degree of parallelism and low power consumption, it is still difficult to achieve high performance with FPGAs under strict power constraint. In this paper, we propose a YOLOv3-tiny-based new network model for object detection which uses depthwise separable convolution, and evaluate the effectiveness of the proposal in terms of processing speed, detection accuracy, and power consumption. As a result, we reduced 30% of total latency and 20% of total power. The results showed that depthwise separable convolution is effective not only for improving performance but also for reducing power consumption.
1
Introduction
In recent years, in a variety of industrial fields, machine automation is gaining momentum. One example is the use of AI systems to automate traffic guidance work, where a shortage of workers is a serious problem. Such systems are expected to reduce the workload and alleviate labor shortages. However, supply of electricity is strictly limited in places such as roads and construction sites. In addition, the portability of systems at the site must also be considered. Therefore, a system with high portability, power efficiency and real-time performance is required. For such embedded systems, FPGAs (Field Programmable Gate Arrays) are attractive due to their good balance between low power consumption and high performance. For automatic traffic guidance, a vehicle detection function is indispensable. The traffic guidance system we are currently developing is limited in the power. It can supply only 4 W to the entire FPGA board. While there are various methods of vehicle detection, including ones based on image processing and a millimeter-wave radar, we will focus on vehicle detection based on image processing in this work. CNN (Convolutional Neural Network) is commonly used in the c The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 223–233, 2022. https://doi.org/10.1007/978-3-031-08812-4_22
224
Y. Araki et al.
field of image-based object detection. YOLOv3, a CNN-based object detection model has been reported to perform well in terms of both accuracy and speed [1]. YOLOv3-tiny is a lightweight model which is less accurate than YOLOv3, but achieves a frame rate of about 6.3 times higher. Therefore, the model is suitable for embedded systems [2]. In this paper, we propose a new network model based on YOLOv3-tiny for vehicle detection and implemented the inference process on Zynq, an FPGA SoC for embedded systems. We investigate the effectiveness of the proposed approach in terms of processing performance and power consumption. The paper is organized as follows. Section 2 describes related work, Sect. 3 describes the proposed network architecture, Sect. 4 describes FPGA design and implementation of inference processing and Sect. 5 describes Evaluation and discussion of the designed system. Finally, we conclude this paper in Sect. 6.
2
Related Work
We introduce some related work on FPGA implementation of YOLO in this section. In reference [3], an FPGA architecture for the YOLOv3-tiny model is proposed. The proposed design can be implemented in a low-end FPGA with relatively small amount of resources. Power consumption is about 3.36 W. However, the processing speed is only 1.88 FPS because it is implemented in a low-end FPGA while maintaining as much accuracy as possible. Therefore, as a vehicle detection system monitoring traffic at a construction site, it lacks real-time performance. Yap et al. implemented the YOLOv2 model on a Cyclone V FPGA using OpenCL. In this research, redundancy is reduced by integrating Convolution layer and Batch Normalization layer and performance is further improved by using 16-bit fixed-point numbers [4]. Duy et al. [5] accelerated the inference of YOLOv2 models on FPGA. All parameters are stored in the FPGA’s internal memory, with weight values expressed in 1 bit and activation values in 3 to 6 bits. Since the output of the intermediate layer is also held in FPGA internal memory, access to external memory is minimized, resulting in reduced power consumption and increased throughput. However, the on-chip power consumption is 18.29 W, which is not suitable considering the strict power constraints of this project. The FPGA chip is a relatively large Virtex-7 XC7VX485T, which is not suitable for this project in terms of power consumption and portability. Ning et al. implemented the YOLOv2 model on a Zynq XC7Z035 FPGA. They evaluated detection accuracy and power [6]. In this research, layer concatenation and quantization to 8-bit fixed-point numbers make the network suitable for hardware. PS-DDR and PL-DDR memories are used to store parameters and intermediate layer outputs. The architecture achieves high throughput by processing all layers in parallel and it is suitable in situations where power consumption is limited. However, the on-chip power is 5.96 W, which exceeds the available power in this project.
Object Recognition System by Using FPGA with Low Power Consumption
3
225
Proposed Network Architecture
In this work, we improved the network model of YOLOv3-tiny by replacing convolutions with depthwise separable convolutions to reduce computation cost. 3.1
Depthwise Separable Convolution
Depthwise separable convolution is a method proposed in [7] that splits an usual convolution (Fig. 1a) into spatial direction (depthwise convolution, Fig. 1b) and channel direction (pointwise convolution, Fig. 1c). This method can approximate the convolution process and significantly reduce parameters. If the number of input channels is N , the kernel size is K, and the number of output channels (number of filters) is M , the number of parameters for normal convolution can be expressed as in Eq. 1 and depthwise separable convolution as in Eq. 2. The ratio of these two is expressed as in Eq. 3. The larger the number of output channels, the greater the effect. We introduce depthwise separable convolution to YOLOv3-tiny. Also, we reduced the number of classification classes to one (Car class) in order to simplify the network structure. M K2 N
(1)
(M + K )N 1 1 + 2 M K
(2)
2
3.2
(3)
Proposed Network: YOLOv3-Tiny-Improved
We call the proposed network YOLOv3-tiny-improved. It is common to use normal convolutions instead of depthwise separable convolutions in the first layer [7– 9]. However, as shown in Table 1, the AP (Average Precision) is almost the same no matter whether normal convolutions are used in the first layer (YOLOv3-tinyimproved ) or not (YOLOv3-tiny-improved). In software, the impact of making the first layer a depthwise separable convolution is a small reduction in computation and a small decrease in AP, which is of little benefit. On the other hand, in the FPGA implementation in this research, the use of depthwise separable convolution for all the layers helps to reduce hardware utilization since it eliminate the need to implement an extra module for normal convolutions. Therefore, all Table 1. Accuracy evaluation on the test dataset Model
AP (%) Model size (MB)
YOLOv3-tiny (1 class)
56.5
33.0
YOLOv3-tiny-improved
53.7
5.4
YOLOv3-tiny-improved 54.0
5.4
226
Y. Araki et al.
Fig. 1. (a) Normal convolution, (b) depthwise convolution, (c) pointwise convolution
Fig. 2. Network architecture of YOLOv3-tiny-improved
the 3 × 3 convolution layers in YOLOv3-tiny-improved are composed of depthwise separable convolution, including the first layer. Network structure is shown in Fig. 2. 3.3
Network Learning
We trained the following three models using PyTorch: YOLOv3-tiny, YOLOv3tiny-improved, and YOLOv3-tiny-improved . The last one is a model in which the first layer is composed of a normal convolution layer. As aforementioned, the number of the classification classes was set to 1. For training, we used images for object detection from Google Open Images Datasets [10]. The dataset consists of 31420 vehicle images with all images labeled as Car, Truck, Bus, or Motorcycle. 21994 images were used for train, 6284 images were used for validation and 3142 images were used for test. The learning rate and learning algorithm were determined based on [11]. Table 1 shows the accuracy of the three trained models when evaluated on the test dataset. While the model size of YOLOv3-tiny-improved is about 85% smaller than that of YOLOv3-tiny, the decline in AP was limited to about 2.8 points. 3.4
Model Quantization
CNN operations are generally performed using floating-point numbers, but considering FPGA implementation, it is desirable to use fixed-point operations, which require less hardware than floating-point operations.
Object Recognition System by Using FPGA with Low Power Consumption
227
According to many studies on CNN quantization, quantizing floating-point numbers to 16-bit fixed-point numbers with an 8-bit decimal part causes almost no loss of accuracy [3,4,12]. To verify whether YOLOv3-tiny-improved could maintain the same accuracy, when quantized with 16-bit fixed-point numbers. The conversion from floating-point to fixed-point numbers was performed using shifting and rounding as shown in Eq. 4. xq and xf are the values after and before quantization respectively, and round() is the function to round the value to integer. (4) xq = round(xf × 28 ) The evaluation was performed using Python, which does not natively support fixed-point arithmetic, so the fixed-point arithmetic was simulated using floating-point arithmetic. Table 2 compares the detection accuracy between 32bit floating-point and 16-bit fixed-point numbers. Table 2 shows that the detection accuracy with 16-bit fixed-point numbers is only 1.3 points lower in AP than with 32-bit floating-point numbers. This indicates that quantization to 16-bit fixed-point numbers is also valid in YOLOv3tiny-improved. Therefore, we decided to use 16-bit fixed-point numbers for FPGA implementation. Table 2. Comparison of detection accuracy by quantization Model
4
Operational precision AP (%) Model size (MB)
YOLOv3-tiny-improved Float (32 bit)
53.7
5.4
YOLOv3-tiny-improved Fixed (16 bit)
52.4
2.7
FPGA Design and Implementation of Inference Processing
As an implementation platform, we used an FPGA SoC, which contains a processing system (PS) and a logic section (PL) on a single chip. The implementation is based on the implementation in the literature [3], which has been reported to have low power consumption. Considering the amount of hardware resources, one hardware module is repeatedly utilized by changing the configuration parameters for each layer. As shown in Fig. 3, for each layer of the inference process, the input images, weight parameters, and intermediate layer outputs are transferred from DDR memory to the accelerator using DMA. When the process of the layer is finished, output results are written back to DDR memory. Each module implemented in the PL section parallelizes operations in the channel direction and performs pipeline processing. The number of input channels that can be processed in parallel depends on the amount of FPGA resources and memory bandwidth. In this implementation, the maximum number of channels to be processed in parallel Nmax is set to 32, referring to the literature [3]. If the number of input channels in the layer Nin and the number of output channels
228
Y. Araki et al.
in the layer Nout are larger than Nmax , the processing in one layer needs to be divided into multiple sublayers and communication with DDR memory occurs for each sublayer processing. We designed the FPGA circuit using HLS (High-Level Synthesis). The circuit was synthesized, placed, and routed with Vivado 2019.1 using IP created by HLS. The following sub-sections describe the designed modules for Pointwise Convolution and Depthwise Convolution in this research.
Fig. 3. CNN accelerator and peripherals
4.1
Pointwise Convolution Module
The pointwise convolution module (Conv pw module) performs convolution operations with a kernel size of 1×1. Figure 4 is an overview of Conv pw module. Normally, when performing convolution operations on streamed input pixels, a line buffer is required to hold the input pixels and form a window. However, the Conv pw module does not require a window to be formed, thus eliminating the need for a line buffer. The Conv pw module receives 4 channels of stream input, supplies each pixel to at most Nmax multipliers, and multiplies each by the corresponding weight kernel stored in the weight buffer. This module receives data from the 4 channel inputs and performs convolution operations with the values stored in the weight buffer. Here, up to NM AX convolution operations are executed in parallel. The computed results are sent to the output buffer everytime input channels. the accumulation is completed for Nmax 4 4.2
Depthwise Convolution Module
The depthwise convolution module performs convolution operations with a kernel size of 3 × 3 that are independent across channels. Figure 5 is an overview of Conv dw module. The Conv dw module has a convolution kernel and a line buffer that holds the stream inputs for each of the 4 channels. When the 3 × 3 windows are prepared, they are cut out and the Conv Kernels perform the convolution operation using the weight kernels stored in the weight buffer. Compared to normal convolution, there is no integration between channels, eliminating the need for a circuit to perform addition between channels.
Object Recognition System by Using FPGA with Low Power Consumption
229
Fig. 4. Conv pw module overviewConv pw module overview
5
Evaluation and Discussion of the Designed System
The evaluation was performed on a small FPGA SoC, Zynq-7020 (xc7z020). In this evaluation, our implementation is compared with the base implementation, and evaluated in terms of resource usage, latency of PL section, and on-chip power. 5.1
Evaluation of Resource Usage
Table 3 shows a comparison of resource usage between base implementation and our implementation. As Table 3 shows, the usage of DSP and BRAM has increased. This is because the implementation of the Conv dw and Conv pw modules has increased the number of multipliers compared to the base implementation. In addition, the HLS pragma ARRAY PARTITION in the Conv pw module increased the number of BRAM partitions used as weight buffers. The decrease in LUT and FF can be attributed to the fact that the original implementation had Nmax Conv Kernels, which were divided into 4 Conv Kernels and Nmax multiplications in this implementation, reducing the logic used for addition and other operations. Table 3. Resource usage (usage rate) Resource
Base-design
LUT
25830 (48.55 %) 18170 (34.15 %)
LUTRAM FF BRAM DSP
284 (1.63 %)
Our-design 882 (5.07 %)
Available 53200 17400
45286 (42.56 %) 19984 (18.78 %) 106400 92 (65.71 %)
120 (85.71 %)
140
161 (73.18 %)
181 (82.27 %)
220
230
Y. Araki et al.
Fig. 5. Conv dw module overview
5.2
Latency Evaluation
Latency was estimated by calculating the number of clock cycles required to process each layer based on the Vivado HLS synthesis report, and then calculated the total number of clock cycles required to process all layers. The result is simply the accumulation of the latency of each layer and the pipeline overlaps between layers are not considered, so the actual latency is expected to be somewhat smaller than the estimation in this evaluation. In the base implementation, multiple layers are processed together as one group (layer group) to reduce the number of communications with DDR memory. Our implementation also follows this approach, to form equivalent layer group structure to the base implementation. Table 4 shows the latency per inference. Figure 6 shows the total number of channels (Nin × Nout ) and the latency reduction ratio by this implementation. The horizontal axis shows the number of channels and the vertical axis is the reduction ratio. According to Table 4, our implementation reduces latency by about 30.8%. Figure 6 shows that in the first half of the layer groups, the larger the number of channels in the layer, the greater the reduction effect by depthwise separable convolution. However, there is not much correlation in the later layer groups. Especially for the layer groups after 7, the latency reduction rate was limited. This would be due the number of clock cycles for accumulation modules is dominant (Fig. 7). Since the group 5 consists of only max pooling, there was no change in latency.
Object Recognition System by Using FPGA with Low Power Consumption
231
Table 4. Latency per inference Implementation Latency (clock cycles) Base
57402524
Ours
39698222
Fig. 6. Latency reduction percentage by layer group
5.3
Power Consumption Evaluation
Finally, power consumption is evaluated. In this research, we compared and evaluated the results of power consumption analysis with those of the base implementation by executing the Report Power command after the Vivado place and route is executed. Table 5 shows the results of the base implementation and our implementation. Figure 8 shows a comparison of the breakdown of dynamic power consumption. Table 5 shows that the on-chip power can be reduced by about 0.5 W compared to the base implementation. Figure 8 shows that the power consumption reduction due to the replacement of the Conv module with the Conv dw and Conv pw modules is significant. These results indicate that depthwise separable convolution is effective in terms of FPGA power consumption. Table 6 shows the frame rate and the number of frames processed per energy (frame rate/power consumption) in the PL section, when operating at 100 MHz. The second row of the table shows the frame rate when inference with the YOLOv3-tiny-improved model is performed using only the Zynq Arm core CPU (Arm Cortex-A9). The inference process on the Arm core was implemented using Python. The power consumption of the Arm core is not included in the comparisons in Table 6, since it cannot be estimated with the Report Power command of Vivado. Reduced latency and power consumption have increased the number of frames processed per energy by a factor of about 1.8. The proposed method improves energy efficiency. Compared to the case using an Arm CPU, our implementation achieves a speedup of 26.72 times. In terms of frame rate, on the other hand, a real-time performance, which is 30 FPS for typical camera devices, was not
232
Y. Araki et al.
Fig. 7. Latency breakdown Table 5. Comparison of on-chip power HW Implementation Static (W) Dynamic (W) Total (W) Base
0.180
2.325
2.505
Ours
0.165
1.838
2.003
Fig. 8. Dynamic power breakdown Table 6. Energy efficiency comparison Implementation
Frequency (MHz) Frame rate (FPS) Energy efficiency (frames /J)
Software on arm 667
0.0942
–
Base
100
1.742
0.695
Ours
100
2.519
1.258
achieved. Further hardware optimization, for example use of compressed arithmetic format to alleviate DMA bandwidth limitation, should be addressed.
Object Recognition System by Using FPGA with Low Power Consumption
6
233
Conclusion
In this paper, we proposed a vehicle detection network model based on YOLOv3tiny that introduces depthwise separable convolution, whose effect on performance was investigated by implementing the system in an FPGA. As a result, latency and power consumption were reduced by approximately 30% and 20%, respectively, compared with the base implementations. This shows that depthwise separable convolution is effective in terms of speed and power in hardware implementation. However, the frame rate shows that it is not yet realtime processing, so further acceleration is needed while maintaining low power consumption.
References 1. Redmon, J., Farhadi, A.: YOLOv3: an incremental improvement (2018). https:// pjreddie.com/media/files/papers/YOLOv3.pdf 2. Redmon, J.C.: Yolo: real-time object detection. https://pjreddie.com/darknet/ yolo/ 3. Yu, Z., Bouganis, C.-S.: A parameterisable FPGA-tailored architecture for YOLOv3-tiny. Appl. Reconfig. Comput. Arch. Tools Appl. 330–334 (2020) 4. Wai, Y.J., Yussof, Z.B.M., Salim, S.I.B., Chuan, L.K.: Fixed point implementation of tiny-Yolo-v2 using OpenCL on FPGA. Int. J. Adv. Comput. Sci. Appl. (IJACSA) 9(10), 506–512 (2018) 5. Nguyen, D.T., Nguyen, T.N., Kim, H., Lee, H.-J.: A high-throughput and powerefficient FPGA implementation of YOLO CNN for object detection. IEEE Trans. Very Large Scale Integr. (VLSI) Syst.27(8) (2019) 6. Zhang, N., Wei, X., Chen, H., Liu, W.: FPGA implementation for CNN-based optical remote sensing object detection. Electronics 10(3), 2021 (2021) 7. Howard, A.G., et al.: MobileNets: efficient convolutional neural networks for mobile vision application. arXiv: 1704.04861v1 [cs.CV] (2017) 8. Li, Y., Han, Z., Xu, H., Liu, L., Li, X., Zhang, K.: YOLOv3-lite: a lightweight crack detection network for aircraft structure based on depthwise separable convolutions. Appl. Sci. 9(18) (2019) 9. Dang, L., Pang, P., Lee, J.: Depth-wise separable convolution neural network with residual connection for hyperspectral image classification. Remote Sens. 12(20) (2020) 10. Kuznetsova, A., et al.: The open images dataset v4: unified image classification, object detection, and visual relationship detection at scale. IJCV 128, 1956–1981 (2020) 11. Linder-Noren, E.: PyTorch-YOLOv3. https://github.com/eriklindernoren/ PyTorch-YOLOv3 12. Dallinger, D.: FPGA optimized dynamic post-training quantization of tiny-YoloV3 (2021). https://publik.tuwien.ac.at/files/publik 296008.pdf
Achieving Sustainable Competitive Advantage Through Green Innovation; the Moderating Effect of Islamic Environmental Ethics and Islamic Business Ethics Budhi Cahyono(B) and Marno Nugroho Faculty of Economics, Sultan Agung Islamic University (UNISSULA), Jl. Raya Kaligawe Km. 4, Semarang, Indonesia {budhicahyono,marnonugroho}@unissula.ac.id
Abstract. The role of ethics is very important in the practice of manufacturing activities in Indonesia, especially Islamic environmental ethics (IEE) and Islamic business ethics (IBE). This study will examine the moderating role of IEE and IBE in the relationship between green product innovation and green process innovation on sustainability competitive advantage. The number of respondents who participated were 145 Islamic entrepreneurs in Central Java. The sampling technique used is purposive sampling. The results show that green product innovation and green process innovation have a significant impact on sustainable competitive advantage. Meanwhile, IEE and IBE moderate the relationship between green innovation and sustainable competitive advantage. The limitations and agenda of future research are also presented in this article. Keywords: Green product innovation · Green process innovation · Sustainable competitive advantage · Islamic environmental ethics · Islamic business ethics
1 Introduction Purba-Rao (2004), says that 70% of industrial activity will be concentrated in the South East Asian region, because this area is a cheaper production house. Environmental issues will become more complex with the increasing activities of the manufacturing industry. Increased activity of the manufacturing industry appears because of the increased demands of economic growth. Environmental management is basically to assess influence on environmental performance and the performance of the company, as an indicator of success in implementing various environmental management variables. Studies on environmental management have tended to emphasize the relationship between variables towards attaining competitiveness level, the company’s performance, environmental performance, and competitive advantage. Concepts discussed focuses on the study of green customer, green purchasing, supply chain environmental management (SCEM), cleaner production, green product and green technology. Environmental management for companies is still regarded as a cost which impact on the emergence of cost, and ultimately © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 234–248, 2022. https://doi.org/10.1007/978-3-031-08812-4_23
Achieving Sustainable Competitive Advantage Through Green Innovation
235
weigh on the cost of the company’s products. But on the other hand has been a lot of studies whose results indicate a positive relationship between environmental management on company’s environmental performance and the company performance (Bekk et al. 2015). Differences in the findings may give different interpretations to the industry in the adoption of regulations and follow up with tangible actions. At the lowest level, environmental management is done when the damage occurred, while at the highest level towards the proactive action. Conflicts of study results indicate the emergence of pessimistic expectations towards efforts to proactively manage the environment. Meanwhile awareness in environmental management will be given priority if there is pressure from consumers, government, and the environmentalist. To create effectiveness study on the model of sustainability competitive advantage based products and processes that are environmentally friendly, then do a review of literature related to practices of environmental management (best practices) and sustainability competitive advantage, so as to obtain a model that provides a comprehensive picture of environmental management, then associated with a competitive advantage. Sustainability competitive advantage can be realized when the company delivers the same capabilities as its competitors, but with a lower cost (cost advantage). Capability to delivers benefits that exceed competitors’ products (differentiation advantage). Differentiation can be developed through product quality, technology and innovation, reliability, brand image, repuptasi company, robustness, and service to consumers, which is difficult for competitors to imitate (Chang 2011). The issue of the impact of the environmental protection of the company’s competitive advantage is not yet serious attention by the academic community. Under conditions of environmental regulations and the adoption of increasingly stringent environmental regulations, and increasing consumer pro-environment, the environmental management company will have an important role in the present time. Environmental management topic has been studied and lead to contradictory results of the study. Chang (2011) found an association between organizational culture based on environment and environment-based leadership of the organization’s identity based on environment and environment-based competitive advantage. In another study Geng et al. (2010) also found a significant effect of the application of cleaner production and corporate performance. Clean Production is distinguished in the low-cost and high-cost schemes, while performance is divided into financial performance and non-financial. Nevertheless, other studies have yielded conflicting results, such as the traditional view that believes that environmental activities will negatively impact the company’s performance, particularly the growth of sales and profit levels. This view is based that the need for high investment as a reflection in creating the product and the production process to achieve economic value and a better environment (Naffziger 2003). The environmental crisis in the 21st century arises because of the loss of human awareness of the preservation of nature. This condition needs an in-depth review to develop and reconstruct new ideas about environmental ethics and environmental business from an Islamic perspective. New thinking about environmental ethics and Islamic business as a form of accountability in religion, in the form of: no wastage or over-consumption of natural resources, no unlawful obstruction or destruction of any component of the natural resources, no
236
B. Cahyono and M. Nugroho
damage, abuse or distortion of the natural environment in any way, sustainable development of the earth, its resources, elements, and phenomena through the enhancement of natural resources, the protection and conservation of them and of all existing forms of life, bringing new life to the land through its reclamation, and the rehabilitation and purification of the soil, air, and water. (Al-Damkhi 2008). The role of ethics in corporate activities is very important, especially regarding Islamic environmental ethics (IEE) and Islamic business ethics (IBE). From some of the results of the study led to the finding that proactive environmental management has a significant impact on corporate performance, environmental performance, and competitive advantage. These findings are interesting, and of course the implication requires deepening studies so as to formulate a model of best practices in environmental management to create a sustainable competitive advantage. Research conducted attempted to use a comprehensive approach based sustainable competitive advantage relation with the environment and green product creation process within the organization. This study will provide guidance for SMEs in developing and implementing models of sustainable competitive advantage based on green innovation. This study aimed to examine the impact of green product innovation and green process innovation on sustainability competitive advantage with IEE and IBE as a moderating variable. The benefits expected from the results of this research is the role of green product innovation and green process innovation to improve competitive advantage in manufacturing industries.
2 Literature Review Green Innovation The revolution of environment management is divided into three stages (Rondinelli and Berry 2000), namely: (1) unprepared or model crisis, (2) reactive or cost models, and (3) proactive or model of business sustainability. Unprepared or model crisis occurred between the years 1960–1970, which focuses on tackling various environmental crisis is happening and try to control the range of the damage. In the second phase, the model of reactive happened in the year 1980 marked the desire of companies to adopt various government regulations in the field of environment at the time it began to grow exponentially, so the need to do efforts to minimize the costs of the complaint. In proactive environmental management era which started in the 1990s, companies started thinking about anticipated environmental impacts of the operation of the company to take measurements of the effort to reduce waste and pollution in connection with the emergence of various regulatory areas of the environment by finding positive efforts in order to achieve business excellence through total quality environmental management (TQEM). At this stage, the company seeks to make prevention of pollution and explore to create new opportunities in developing green products, green process, and green technology. Competitive advantage at small medium enterprises based on the principle of environmental management will emphasize the importance of a competitive advantage for SMEs which rely on environmental management, along with growing environmental issues at this time. The company not only focus on how to increase profits, but also that the company has a greater responsibility of maintaining environmental quality. This concept emphasizes a balance between the purpose of profit and creation of environmental
Achieving Sustainable Competitive Advantage Through Green Innovation
237
sustainability. Meanwhile in future studies will focus on the application of best practices in the field of environmental management as a basis for building competitive advantage of companies. The concept of green innovation has been studied by several experts. Green innovation is divided into two forms, namely green product innovation and green process innovation (Huang and Li 2015). Further studies focusing on the role of departments, agencies, environment offices living in each county or city in Central Java to determine the level of effectiveness of the institution’s role in the planning, implementation and monitoring of the environmental field. Following up on two previous studies, in 2009 assessed on environmental management models associated with environmental performance and corporate performance. The results indicate the existence of significant influence, even so many variables of the study showed a score below average. Various previous studies have contributed to SMEs in environment management, however previous studies are still partial, both for the scope and subject matter of the variables. Further research is focused on the subject matter of wider by developing variables relevant research, among others: the clean production, environment-based organizational culture, leadership-based environment, the identification of green organizations, green and competitive advantage. While the models that have been built in the first year further implemented and evaluated, and eventually used as instruments for SMEs in gaining competitive advantage.
3 Sustainability Competitive Advantage Competitive advantage can be realized when the company delivers the same capabilities as its competitors, but with a lower cost (cost advantage), or capable delivers benefits that exceed competitors’ products (differentiation advantage). Differentiation can be developed through product quality, technology and innovation, reliability, brand image, company reputation, robustness, and service to consumers, which is difficult for competitors to imitate (Mose 2010). The competitive advantage that comes from an internal factor is the crucial factor for success. Competitive advantage through the enrichment of management actions through the management structure, processes, culture and people in the organization. The resources are scarce and valuable at the same time can create a competitive advantage, especially those resources is difficult to imitate, changed and redeemed (Barney 1991). A company implements a differentiation strategy can achieve competitive advantage beyond its competitors, due to its ability to create opportunities to build consumer and brand loyalty through offering quality, advertising and marketing. While Barney (1991) prioritizes the company’s ability to describe the various barriers to entry to imitate relation to the prevention of its competitors and get an edge by technology and innovation, human resources, and organizational structure. The various strategies of competitive advantage offered by Iles (2008), consisting of strategy differentiation, cost leadership, and differentiation focus. Customer segmentation will provide a special opportunity to provide a product that is different from competitors that have broader targets. Serving segmentation limited impact on distribution costs are cheaper.
238
B. Cahyono and M. Nugroho
Core competence is measured by three-dimensional; shared vision, cooperation and empowerment. Sustainable competitive advantage is measured by flexibility and responsiveness. Organizational performance is measured by growth and profitability. LeonardBarton (2000) core competence as one of the which differentiates a firm from its milieu. Sanchez and Heene (1997) core competencies are usually the result of collective learning processes and are manifested in business activities and processes. (González-Benito and González-Benito (2005), core competency is a collection of competencies that are widespread in the corporation. Prasad et al. (2018), argue that the core competence en communication, involvement, and a deep commitment to working across organization boundaries.
4 Relationship Between Green Innovation and Sustainability Competitive Advantage Chen et al. (2006) says there is significant correlation between green product innovation and green process innovation on competitive advantage of companies in the semiconductor company, information hardware industry, optoelectronic industry, industry, communications, consumer electronics industry, and electronic component industry. Performance innovatin green product and process have a positive relationship with the company’s competitive advantage. These findings indicate that the company’s innovation in products and processes that will improve competitive advantage. Hasil penelitian juga menunjukkan bahwa green innovation yang meliputi green product innovation dan green process innovation memiliki pengaruh penting dalam meningkatkan kinerja lingkungan maupun kinerja perusahaan. Green product innovation dan green process innovation juga memiliki pengaruh signifikan terhadap keunggulan bersaing perusahaan. Based on empirical studies, the following hypotheses can be formulated: The results also show that green innovation which includes green product innovation and green process innovation has an important influence in improving environmental performance and company performance (Huang and Li 2015). Green product innovation and green process innovation also have a significant influence on the company’s competitive advantage (Chang 2011). Hypothesis 1: Green product Innovation have a positive impact on sustainability competitive advantage Hypothesis 2: Green process innovation had a positive impact on sustainability competitive advantage
Achieving Sustainable Competitive Advantage Through Green Innovation
239
5 Moderating Effect of IEE and IBE on Relationship Between Green Innovative and Sustainability Competitive Advantage Green innovation is the improvement of products or processes about energy-saving, pollution-prevention, waste recycling, green product designs, and corporate environmental management in the field of environmental management (Chen et al. 2006). This study divides green innovation into green product innovation and green process innovation. Green innovation can enhance the performance of environmental management to satisfy the requirements of environmental protection. A company devotes to develop green innovation can not only meet the environmental regulations, but also build up the barriers to the other competitors (Barney 1991; Chen et al. 2006). Previously, many companies thought investing in environmental management was an unnecessary investment (Porter and van der Linde 1995). Recently, there are more positive associations between corporate environmental ethics and green innovation. Companies should change their strategies and operations so that they can comply with the trend of environmentalism. Companies with high-environmental ethics are prone to increase resource productivity through green innovation to make up the environmental costs (Chen et al. 2006). Ultimately, green innovation can further raise resources productivity and make companies more competitive (Porter and van der Linde 1995). Companies require the motivation and ability to produce creative and innovative ideas to develop new products or processes (Chen and Huang 2009). Previous studies pointed out that the well-defined policies and processes in companies have positive effect on their innovation (Stewart 1994). Therefore, well-defined environmental policies can facilitate and integrate the operations among different departments in companies and solve the environmental problems (Porter and van der Linde 1995). Corporate environmental ethics highlights the role of proactive environmental management (Weaver et al. 1999b). The environmental ethics in a company can influence innovation of environmental technology and business operation (Greeno and Robinson 1992; Schlegelmilch et al. 1996). This study argues that corporate environmental ethics plays an important role in the green innovation of a company. Corporate environmental ethics is regarded as one kind of superior corporate culture to attain sustainable development. Hence, corporate environmental ethics of companies can stimulate their proactive environmental actions that can facilitate their green innovations (Chang 2011). In doing business, Islam uses principles that pay attention to the interests of the world and the hereafter. The study (Tlaiss 2014) states that the instrument for the survival and success of the company consists of: the interests of the world and the hereafter (fallah), carried out in earnest (itqan), related to the values of goodness and hard work (amal salih), honesty and hold fast to trust (sidik and trust), are fair and just (haqq and adl), and benevolence (ihsaan). Hypothesis 3: Islamic environmental ethics moderate the relationship between green product innovation and sustainable competitive advantage. Hypothesis 4: Islamic environmental ethics moderate the relationship between green process innovation and sustainable competitive advantage (Fig. 1).
240
B. Cahyono and M. Nugroho
Green Product innovation Sustainable Competitive Advantage Green process innovation
• Islamic Environmental Ethics • Islamic Business Ethics Fig. 1. Research model
6 Methods Population and Sampling Method The population in this study are all small and medium entreprise in Central Java industrial cluster. Cluster is an approach in which the SMEs was developed in groups joined in a holistic approach and integrated approach, include: establishment of raw materials and suppliers, production or transformation process, suppliers, service providers, other companies involved, and the supporting sector to another (accreditation, certification and calibration institutions; organization of quality control; research and development institutions and business associations). Relating to the research topic about the moderating role of Islamic environmental ethics and Islamic business ethics, the researcher uses purposive sampling, meaning that this research will use Muslim entrepreneur respondents to support the research activities carried out. Research respondents were managers in IKM, totaling 145 respondents in the manufacturing industry, including: batik, furniture, hijab, bread, and apparel industries. Research Variables and Measurement. The variables in this study consisted of independent variables, green product innovation and green process innovation. Green product innovation measured using three indicators: raw materials with minimal pollution, raw materials with low energy consumption and raw materials efficiently. Green process innovation is measured by four indicators: the production process to reduce waste, the production process is able to recycle, the production process is able to reduce the consumption of water, electricity, and oil, and the production process is able to reduce the use of raw materials. Sustainability competitive advantage is measured using eight indicators: excellence in production costs, product quality, product development, managerial ability, profit, number of production, the initial originator, and the image of the company. The moderating variable of Islamic environmental ethics is measured by six indicators: environmental management as a form of responsibility to God, creative environmental treatment and environmental
Achieving Sustainable Competitive Advantage Through Green Innovation
241
conservation, the principle of balance in environmental conservation, humans have the nature to be responsible for the environment, stay away from environmental destruction, and develop methods to protect the environment. environment from damage. Meanwhile, the Islamic business ethics variable is measured by six indicators: honesty, trustworthiness, integrity, working hard, clever, and serious (Tlaiss 2014). Analysis of data using multiple regression analysis and interaction regression.
7 Result and Discussion The result showed two important findings: First, based on the average value of each indicator can be concluded that to get to the level of proactive environmental management shows far from expectations, considering all of the average of all the indicator value is below three or below the average score. These findings demonstrate that empirically proactively on environmental management in the manufacturing industry has not led to a proactive corporate environmental management. Second, companies participating in the study had a score below a proactive environment as much as 65.7%, which means indicates that the phenomenon of proactive environmental management has not received serious attention. Meanwhile, the number of companies that have a proactive environmental management value of more than one as much as 34.3%. Environmental performance indicators used in this study may help improve the reputation and organizational strategy related to the importance of changes in the future to manage the environment. Values of respondents for each indicator in product innovation is still below the average (on a scale of 1–5), it means that the connection with product innovation at SMEs is still not fully done. But in the first indicator has the value of the average value of the ultimate answer, that is equal to 2.6138 for the indicator use of raw materials with minimal pollution. While it’s an green process innovation variable average value of respondents also remain below the average value, it means that green process innovation activity and green production process is still not optimal. Indicator whose value tends to be high is on the fourth indicator, namely the production process is able to reduce the consumption of raw materials. Sustainable competitive advantages variable, there are several variables that can be used as an indicator of excellence at SMEs, which is an indicator of product quality, product development, and production cost advantages. Table 1. Means, deviation standard and cronbach alpha values Indicators
Mean
Std. deviation
Cronbach α
GProdI-1
2.6138
1.53743
0,978
GProdI-P2
2.3724
1.37913
GProdI-P3
2.6138
1.55092
GProcI-1
2.4345
1.48518
GProcI-2
2.3655
1.45685
GProcI-3
2.4759
1.47706
0,976
(continued)
242
B. Cahyono and M. Nugroho Table 1. (continued)
Indicators
Mean
Std. deviation
GProcI-4
2.4759
1.50038
Iee-1
4.0759
.85051
Iee-2
4.1931
.81901
Iee-3
3.9103
.78100
Iee-4
4.2621
.70731
Iee-5
4.2414
.82728
Iee-6
4.2207
.82890
Ibe-1
4.0345
.62820
Ibe-2
4.0276
.62299
Ibe-3
4.2552
.74330
Ibe-4
4.3862
.70893
Ibe-5
4.3379
.69941
Ibe-6
4.0690
.63086
SCA-1
2.4483
1.39906
SCA-2
2.5379
1.52307
SCA-3
2.5103
1.49591
SCA-4
2.2828
1.29477
SCA-5
2.4552
1.38934
SCA-6
2.1310
1.20908
SCA-7
2.1517
1.22095
SCA-8
2.4483
1.45264
Cronbach α 0,902
0,814
0,984
The correlation between variables shows a very high value, such as the correlation between the green product innovation with sustainable competitive advantage was 88.5%, while the correlation between green process innovation with sustainable competitive advantage by 84.3%. The value of R2 is 72.2%, meaning that 72.2% of competitive advantage in the manufacturing industry affected by green product innovation and green process innovation, while 19.8% of sustainable competitive advantage is influenced by other variables that have not been studied. The influence of green product innovation and green process innovation to sustainable competitive advantage can be seen in Table 2. Green product innovation has a positive and significant impact on sustainable competitive advantage. While the green process innovation also has a positive and significant impact on sustainable competitive advantage. These findings indicate that green products innovation activities will be able to increase the sustainable competitive advantage at SMEs. Similarly, improvements in the green process innovation will be able to increase the competitive advantage at SMEs. These findings can be used as guidelines for company management in an effort
Achieving Sustainable Competitive Advantage Through Green Innovation
243
to increase the sustainable competitive advantage. The results of interaction regression by including moderating variable IEE and IBE. The test is to answer hypothetical three which states that IEE and IBE moderating the relationship between green product innovation and green process innovation with sustainable competitive advantage. Adjusted R2 value of 0.722 indicates that the independent variable has a contribution of 72.22% in influencing the dependent variable (see Table 2). Table 2. Interction regression result Model 1 (Constant)
Unstandardized Coefficients B -35,387
Std. Error 10,959
Standardized Coefficients Beta
t
Sig
-3,229
,002
Prod
,023
,011
,780
2,069
,040
Proc
,541
IEE
,061
,271
,411
1,998
,048
,026
,104
2,325
,022
IBE
,898
,386
,875
2,327
,021
PD-IEE
,013
,004
,138
2,949
,004
PD-IBE
5,124
1,442
,317
3,552
,001
PR-IEE
,305
,119
,321
2,558
,012
PR-IBE
,208
,097
,169
2,158
,033
a. Dependent Variable: SCA b. Predictors: (Constant)„ Prod, Proc, IEE, IBE, PD_IEE, PD_IBE, PR_IEE, PR_IBE, c. F value = 47,678, Sign = 0,000 d. R2 = 0,737 e. Adjusted R2 = 0,722
Green product innovation has the highest correlation on sustainable competitive advantage, as well as green process innovation is also a high correlation with a sustainability competitive advantage. The interaction regression results show that IEE and IBE moderating the relationship between green products innovation and green process innovation on sustainability competitive advantage. This finding shows that IEE and IBE have an important role or as a reinforcement in the relationship between green product innovation and green process innovation with sustainability competitive advantage. Green product innovation has a positive and significant impact on sustainability competitive advantage, meaning that if the SMEs can perform a variety of product innovation refers to the context of environmentally friendly, for example by the use of raw materials that cause pollution at a minimum, the raw material with energy consumption low, and the use of raw materials efficiently now, it can increase sustainability competitive advantage through reduced production costs, increased product quality, product development capabilities, improving managerial capacity, increase profits, increase the amount of production, and improved corporate image. Green product innovation will adversely impact the reduction of production costs, managerial capabilities, and corporate profits. Green process innovation also has a positive and significant impact on
244
B. Cahyono and M. Nugroho
sustainability competitive advantage at SMEs. Green process innovation can be developed with a variety of activities related to the production process, namely the production processes that reduce waste, recycle, reduce energy consumption, and production processes that reduce the use of raw materials. Green process innovation will be able to increase the sustainability competitive advantage primarily on the indicator of product quality, product development, and the company as the initial trigger. Results of regression to test the interaction of variables moderating IEE and IBE indicated moderate the relationship between green product innovation and green process innovation with sustainability competitive advantage. These findings indicate that the IEE and IBE will be able to strengthen the relationship between the innovation of environmentally friendly products and process innovation with sustainable competitive advantage, meaning that the better of IEE and IBE for SMEs will be able to increase the relationship between green product innovation and green process innovation in terms of sustainable competitive advantage. Islamic approach in protecting the environment through education, regulations, and conservation practices (Majeri Mangunjaya and Elizabeth McKay 2012). Islam in understanding the environment by linking religion with other aspects of everyday life. Islamic legal and ethical thinkers state that Islam positions environmental conservation as one of the great goals of Islamic law, which is better known as maqasid sharia. Maqasid sharia emphasizes on five human interests, including: religion, life, culture, wealth, and knowledge. Sustainability competitive advantage can be realized when the company delivers the same capabilities as its competitors, but with a lower cost (cost advantage) for along time. Capable delivers benefits that exceed competitors’ products (differentiation advantage). Differentiation can be developed through product quality, technology and innovation, reliability, brand image, company reputation, robustness, and service to consumers, which is difficult for competitors to imitate (Mose 2010). The competitive advantage that comes from an internal factor is the crucial factor for success. Competitive advantage through the enrichment of management actions through the management structure, processes, culture and people in the organization. Barney (1991), states that the resources are scarce and valuable at the same time can create a competitive advantage, especially those resources is difficult to imitate, changed and redeemed. A company implements a differentiation strategy can achieve competitive advantage beyond its competitors, due to its ability to create opportunities to build consumer and brand loyalty through offering quality, advertising and marketing. The sources of competitive advantage according to Barney (1991) include: 1) technology and innovation, innovation plays an important role in the economic development of the country, because of innovative companies through the commercialization of research and development will create new values. The company will obtain important information from the value created. In this way they will create wealth for themselves, the country and even the world. Innovative companies will always engage in continuous research to create better products, services and ways of doing things. They are trying to continuously improve internal capabilities and other resources. Knight (2007), states that companies more productive firms are more efficient in the use of resources. 2) human resources, companies can create competitive advantage only by creating value in ways that are difficult to imitate by competitors. The company can develop a competitive
Achieving Sustainable Competitive Advantage Through Green Innovation
245
advantage by creating the values that are difficult to imitate by competitors. Sources traditional advantages such as financial and natural resources, technology and economies of scale can be used to create value. Competitive advantage through human resource can be done with the best practices approach to HRM strategy, among others: internal career opportunity, training systems, selection and socialization process, appraisal, employment security, employee participation, job description, profit sharing. 3) organizational structure, the organizational structure are arranged depending on the company’s goals. An effective organizational structure to be able to facilitate working relationships among the various parts of the organization and can improve work efficiency within each organizational unit. The organizational structure can guarantee individual membership applications to new levels of flexibility and high creativity. The strategy of competitive advantage Iles (2008); 1) strategy-differentiation; successful in the differentiation strategy should be supported by internal forces, such as: access to leading scientific research, highly skilled and creative product development team, strong sales team, and corporate reputation for quality and innovation. 2) strategycost leadership; internal strength needed in cost leadership strategy, among other things: access to the capital required making a significant investment in production assets, skills in designing products for efficient manufacturing, high level of expertise in manufacturing process reengineering, and efficient distribution channels. 3) strategy- focus; this strategy will emphasize the need to focus on a small number of targeted market segments. Customer segmentation will provide a special opportunity to provide a product that is different from competitors that have broader targets. Serving segmentation limited impact on distribution costs are cheaper. The effectiveness of proactive environmental management is needed to respond the problems in the growth of global environmental degradation (Marcus and Fremeth 2009). Environmental management is positioned as an activity that has the main purpose to protect the environment through technical, policy, and procedures used by companies to monitor and oversee the operational impact to the surrounding environment. In particular, according to (Peng and Lin 2008), created a proactive environmental management as a follow-up implementation of practices and innovations, such as: operational design, process and product, to prevent environmental impact. Companies that respond to environmental issues proactively by practicing environmental management has indicated several avenues of profit. For example: 3M innovation through 3P (Pollution Prevention Pays), through the development of programs that commitment to the reduction of resources, through reformulation, process modification, equipment redesign, recycling, and reuse. Environmental management is trying to balance the interests of business and the environment has been through a long stage, ranging from reactive to proactive (Sharma and Vredenburg 1998). Peng and Lin (2008), states that in researching environmental management practices proactive environmental management defines as: producing environmentally friendly products through minimization of environmental impact in the clean production, green marketing and proactive management administrative. Company performance: investment in the field of environment not only reduces costs, but also increased the demand of consumers who are sensitive to the environment. However, the advantages of investing in PCEM is not easily evaluated with traditional accounting measurement.
246
B. Cahyono and M. Nugroho
The development of green innovation also occurs in the field of marketing. Research shows that there is a positive impact of green brand equity on green brand image via green satisfaction and green trust as parallel mediators (Bekk et al. 2015). More than, proactive leaders are an important factor to achieve a successful environmentally sustainable development products, and must be added with technological expertise as the basis of green innovation. The definition of green product innovation is supported by the use of technology (de Medeiros et al. 2018). People’s behavior towards the environment is strongly influenced by the behavior and perceptions of the community towards the environment (Cerda Planas 2017). The magnitude of the leader’s influence in responding to the environment is also supported by Huang and Li (2015), in his research found that green creativity is the impact of green transformational leadership, green intrinsic and extrinsic motivation. Another finding shows that green management is the most decisive factor for radical product innovation than incremental product innovation. Meanwhile, government support as a formal institution provides a very strong mediation in the relationship between green management on radical product innovation than its effect on incremental product innovation. Where social legitimacy as an informal institution is very strong in mediating the relationship between green management on incremental product innovation and its effect on radical product innovation. This finding has important implications in explaining how companies use green management to facilitate products innovations. (Chen et al. 2015).
8 Conclusion The increasing green product innovation will further increase the sustainability competitive advantage. This means that if a company uses raw materials with minimal pollution, energy consumption is low, and the use of raw materials efficiently will be able to create excellence in production costs, opportunity creation of product quality, product development capabilities, managerial ability, profits, total production, and improved company image. Green process innovation has a significant impact on sustainability competitive advantage. This means that if the company’s production process capable of reducing waste, able to recycle, can reduce energy consumption and efficiency of raw materials will increase excellence in production costs, opportunity creation of product quality, product development capabilities, managerial ability, profit, number of production and improvement of corporate image. IEE and IBE moderate the relationship between green product innovation, green process innovation and competitive advantage. This means that to manage entreprise not only based on hard side, like: manage product, manage process, leadership, work environment etc., but the soft side is very important to support the entreprise success, especially on Islamic environmental value and Islamic business ethics. Sustainable competitive advantage have studied, and the higher of sustainability competitive advantage will be depend on Islamic concept, especially by Islamic environmental ethics and Islamic business ethics.
9 Limitation and Future Research The data used in this study using a questionnaire, so it is still not able to provide a comprehensive overview of the variable green product innovation and green process innovation
Achieving Sustainable Competitive Advantage Through Green Innovation
247
is environmentally friendly. In addition, questionnaires not provide the depth of data required and the answers tend not give the real result. Future research is recommended to create a model that more studies can illustrate the actual conditions. Research can be enlarge by add variables, like; green intellectual capital, green human capital, green structural capital, and green relational capital. As for the retrieval of data using direct interviews or focus group discussions with company managers, so that the preparation of environmental management models can be more comprehensive. Acknowledgment. The research team would like to thank: the Director General of Higher Education, Ministry of Education and Culture, and the Dean of the Faculty of Economics, Unissula, who have financed and encouraged the research and publication of this article. The team also thanked LPPM Unissula who have encouraged and facilitated this research activity. As well we thank the parties involved with implementation of research activities.
References Al-Damkhi, A.M.: Environmental ethics in Islam: Principles, violations, and future perspectives. Int. J. Environ. Stud. 65(1), 11–31 (2008). Doi: https://doi.org/10.1080/00207230701859724 Barney, J.B.: Firm resources and sustained competitive advantage. J. Manage. 17(1), 99–120 (1991). https://doi.org/10.1177/014920639101700108 Bekk, M., Spörrle, M., Hedjasie, R., Kerschreiter, R.: Greening the competitive advantage: antecedents and consequences of green brand equity. Qual. Quant. 50(4), 1727–1746 (2015). https://doi.org/10.1007/s11135-015-0232-y Cerda Planas, L.: Moving toward greener societies: moral motivation and green behaviour. Environ. Resour. Econ. 70(4), 835–860 (2018). Doi:https://doi.org/10.1007/s10640-017-0167-y Chang, C.H.: The influence of corporate environmental ethics on competitive advantage: the mediation role of green innovation. J. Bus. Ethics 104(3), 361–370 (2011). https://doi.org/10. 1007/s10551-011-0914-x Chen, Y.S.: The positive effect of green intellectual capital on competitive advantages of firms. J. Bus. Ethics 77(3), 271–286 (2008). https://doi.org/10.1007/s10551-0069349-1 Chen, Y.-S., Chang, C.-H., Yeh, S.-L., Cheng, H.-I.: Green shared vision and green creativity: the mediation roles of green mindfulness and green self-efficacy. Qual. Quant. 49(3), 1169–1184 (2014). https://doi.org/10.1007/s11135-014-0041-8 Chen, Y.S., Lai, S.B., Wen, C.T.: The influence of green innovation performance on corporate advantage in Taiwan. J. Bus. Ethics 67(4), 331–339 (2006). https://doi.org/10.1007/s10551006-9025-5 Cheng, C.C.J.: Sustainability orientation, green supplier involvement, and green innovation performance: evidence from diversifying green entrants. J. Bus. Ethics 161(2), 393–414 (2018). https://doi.org/10.1007/s10551-018-3946-7 de Medeiros, J.F., Vidor, G., Ribeiro, J.L.D.: Driving factors for the success of the green innovation market: a relationship system proposal. J. Bus. Ethics 147(2), 327–341 (2015). https://doi.org/ 10.1007/s10551-015-2927-3 Geng, Y., Xinbei, W., Qinghua, Z., Hengxin, Z.: Regional initiatives on promoting cleaner production in China: a case of Liaoning. J. Clean. Prod. 18(15), 1500–1506 (2010). https://doi. org/10.1016/j.jclepro.2010.06.028 González-Benito, Ó., González-Benito, J.: Environmental proactivity and business performance: an empirical analysis. Omega 33(1), 1–15 (2005). https://doi.org/10.1016/j.omega.2004.03.002
248
B. Cahyono and M. Nugroho
Huang, J.-W., Li, Y.-H.: Green innovation and performance: the view of organizational capability and social reciprocity. J. Bus. Ethics 145(2), 309–324 (2015). https://doi.org/10.1007/s10551015-2903-y Iles, A.: Shifting to green chemistry: the need for innovations in sustainability marketing. Bus. Strateg. Environ. 17(8), 524–535 (2008). https://doi.org/10.1002/bse.547 Majeri Mangunjaya, F., Elizabeth McKay, J.: Reviving an islamic approach for environmental conservation in Indonesia. Worldviews: Global Religions, Culture, and Ecology 16(3), 286–305 (2012). https://doi.org/10.1163/15685357-01603006 Peng, Y.S., Lin, S.S.: Local responsiveness pressure, subsidiary resources, green management adoption and subsidiary’s performance: evidence from taiwanese manufactures. J. Bus. Ethics 79(1–2), 199–212 (2008). https://doi.org/10.1007/s10551-007-9382-8 Prasad, S., Shankar, R., Gupta, R., Roy, S.: A TISM modeling of critical success factors of blockchain based cloud services. J. Adv. Manage. Res. 15(4), 434–456 (2018). https://doi.org/ 10.1108/JAMR-03-2018-0027Purba-Rao.(n.d.) Rondinelli, D.A., Berry, M.A.: Corporate environmental management and public policy: Bridging the gap (2000) Tlaiss, H.A.: How Islamic business ethics impact women entrepreneurs: insights from four Arab middle eastern countries. J. Bus. Ethics 129(4), 859–877 (2014). https://doi.org/10.1007/s10 551-014-2138-3
Digital Social Capital and Financial Inclusion for Small Medium Enterprises Mutamimah(B) and Pungky Lela Saputri Department of Management, Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia {mutamimah,pungkylelasaputri}@unissula.ac.id
Abstract. This study aims to examine the role of digital social capital in increasing financial inclusion for SMEs. The population in this study is all SMEs in Central Java. The sampling techniques of this study used purposive random sampling with the criteria of fashion SMEs that use digital transactions, and the customers are from financial institutions, so 205 respondents were obtained. For statistical analysis, it used descriptive analysis and moderated regression analysis. After conducting instrument quality and classical assumption tests, it showed that financial literacy and financial technology do not affect financial inclusion. However, after including digital social capital as a moderation variable, it showed that digital social capital could strengthen the effect of financial technology on financial inclusion. Keywords: Financial literacy · Financial technology · Financial inclusion · Digital social capital
1 Introduction Indonesian SMEs have good prospects, as Focus Economy Outlook 2020 shows that the contribution of SMEs to the Gross Domestic Product in 2020 amounted to Rp 1,100 trillion. But the industry was unable to develop correctly due to limited access to finance [1, 2]. Limited financial access (financial inclusion) is due to SMEs not having collateral and low financial literacy level. This further encourages the emergence of credit risk [3]. In this digital era, the financial sector follows the dynamics of technological developments such as financial technology used in business and financial transactions to facilitate SMEs and public financial access. This ease of access will encourage the development of SMEs. Financial technology (fintech) is a digital innovation in financial institutions [4]. Financial institutions provide financial technology services such as M-Banking, Internet banking, digital platforms, and marketplaces to facilitate business and financial transactions for SMEs and the public. Fintech is a technology-based financial innovation that can provide new business models, applications, processes, and financial services [5]. To facilitate financial access, financial literacy is needed. Financial literacy is the ability for a person to read, analyse, manage and communicate economic conditions that affect © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 249–259, 2022. https://doi.org/10.1007/978-3-031-08812-4_24
250
Mutamimah and P. L. Saputri
their well-being [6]. Financial literacy includes knowledge, understanding, and skills in financial management. If SMEs have a high level of financial literacy, SMEs will be easier to get financial access [7]. The results of previous research on the influence of financial literacy and financial technology on financial inclusion are still contrary to each other or inconclusive. Financial literacy positively influences financial inclusion [8, 9]. Financial literacy negatively affects financial access [10]. Financial technology has a positive influence on financial inclusion [11]. Fintech does not influence financial inclusion [12]. The existence of inconclusive research results is due to differences in behaviour. It indicates that other variables play a role in influencing the relationship between independent variables and dependent variables as moderating variables [13]. Therefore, this study adds the digital social capital variable as a moderation variable. Digital social capital can act as social collateral that can replace physical collateral required by financial institutions to increase financial inclusion [14]. Digital social capital is social networking and internetbased knowledge sharing [15]. Digital social capital is the actual and potential digital number that exists in digital trust, digital social cliques, digital social networking, and digital social obligation owned by a person or organization [16]. The existence of digital social capital will make it easier for SMEs to get access to financial products and services (financial inclusion). Thus, the purpose of this study is to analyse the role of digital social capital in moderating financial literacy and financial technology on financial inclusion for SMEs.
2 Literature Review 2.1 Financial Inclusion for Fashion SMEs Financial inclusion is an essential factor that can encourage the development of fashion SMEs in Indonesia. Financial inclusion is a condition where every public member has access to quality financial products and services on time, smoothly, and safely at affordable costs by their individual needs and capabilities [17]. Financial inclusion can open opportunities for business actors that are useful to access the availability of financial services, welfare in the use of financial products and services that can eventually be used and can also be utilized in the process of business activities in improving sales growth, growth in profits, capital, and employment. In Indonesia, the Islamic financial inclusion rate has decreased from 11.1% to 9.1%, while the same indicator in conventional financial institutions has reached 76.19% [18]. For SMEs, capital is still becoming the main problem [19], so the open access to finance allows them to expand their business [20]. The high level of financial inclusion for SMEs will make it easier for them to access capital and other conveniences in getting access to financial services to develop correctly. 2.2 Financial Literacy and Financial Inclusion Financial literacy is a skill that SMEs must own because the high financial literacy will make it easier for them to choose various alternative services from financial institutions. Financial literacy is the knowledge, understanding, financial skills that a person has
Digital Social Capital and Financial Inclusion for Small Medium Enterprises
251
as a basis for financial retrieval. Financial literacy helps empower and educate small businesses so that they have the knowledge and can evaluate various financial products and services to make financial decisions wisely [21]. Financial literacy has a positive influence on financial inclusion. The higher the knowledge, understanding, and skills of a person in the field of finance will make it easier for them to make financial decisions and get access to capital and financial services [8, 9, 22]. Thus, H1: Financial literacy has a positive influence on financial inclusion. 2.3 Financial Technology and Financial Inclusion In digital technology today, people will easily carry out economic activities and business transactions. Financial technology can be divided into two points of view, namely: the demand side and supply side [5, 23, 24]. Demand-side means that financial innovation is seen from the point of view of customers and the ability to use the internet in making real-time transactions fast and efficient. The Supply-side is the point of view of the government and financial institutions in providing financial access. Financial technology can increase financial inclusion [22, 25, 26]. H2: Financial technology has a positive effect on financial inclusion. 2.4 Digital Social Capital, Financial Literacy, and Financial Inclusion In today’s technological era, encouraging people to use social media to form social capital is known as digital social capital. Social capital is an intangible resource that encourages people to create social structures through reciprocity, trust, and cooperation for those in the community [27]. Through structural mechanisms formed, social capital can facilitate information sharing, collective action, and decision making in substitute physical collateral [28]. In today’s technological era, it strengthens the role of social capital in shaping networks, knowledge sharing, solution collection, and digital-based trusts. Digital social capital is an internet-based social connection and knowledge sharing [15]. The existence of digital social capital will increase knowledge, understanding, and financial skills for SMEs because the social capital community will facilitate the sharing of information, knowledge, trust, and collective action. Thus, it will make it easier to access financial services for MSMEs. Social capital as social collateral substitutes physical collateral required by financial institutions, making it easier for SMEs to get financial access [14]. H3: Digital social capital can strengthen the influence of financial literacy on financial inclusion. 2.5 Digital Social Capital, Financial Technology, dan Financial Inclusion Digital social capital is a community formed and processed digitally that becomes a valuable asset that binds them facilitates knowledge sharing, trust, collective solutions so that intangible resources are precious. In today’s technological era, networks, knowledge sharing, digital-based collaborative solutions will be more effective, efficient, and networking more widespread. Digital social capital departs from a community with the same goal where a lot of information exchange occurs. The relationship that exists is not
252
Mutamimah and P. L. Saputri
just quantity but company status and reputation, where technology also polishes the quality of relationships in digital social capital. It can be concluded that digital social capital is an internet-based social connection and knowledge sharing [15]. Through structural mechanisms and cognition, social capital can facilitate information sharing, collective action, and decision making in substitute physical collateral, increasing members’ contributions in sharing and being responsible for group lending [28]. Thus, digital social capital can improve financial technology owned by SMEs so that further, it will make them easy to get financial access from banks and non-banks. H4: Digital social capital can strengthen the influence of financial technology on financial inclusion (Fig. 1).
Fig. 1. Research Framework explains the effect of financial literacy and financial technology on financial inclusion, and the role of Digital Social Capital moderating the effects of financial literacy and financial technology on financial inclusion.
3 Research Method This research design is explanatory research that explains the relationships between variables through hypothesis testing. The population of this study was all fashion SMEs in Central Java. The sampling technique used was purposive random sampling with the criteria of fashion SMEs that use digital technology and operate for at least two years. The number of fashion SMEs cannot be known with certainty because cooperative service SMEs do not register some fashion SMEs; the sampling technique used the formula by Jacob Cohen as follows: t N = 2 + +1 F 19, 76 + 4 + 1 = 202, 6 N= 0, 1 = 203 respondents (minimum). Which means:
Digital Social Capital and Financial Inclusion for Small Medium Enterprises
253
N = sample size. F2 = effect size (0,1). u = the number of variables related to research (4) t = the power function of ‘u’, obtained from the table The value of t-table with t-significance of 1%, power of 0.95 and u = 5 is 19.76. This study used descriptive analysis and Moderated Regression Analysis (MRA). Four indicators measuring financial literacy [6] include behaviour, skill, attitude, and knowledge. Financial inclusion indicators include access, usage, welfare, and quality [29, 30]. Financial technology (fintech) indicators include digital financial services applications, digital payment systems, and digital transaction innovation [31]. Digital social capital indicators include social networking, trust, and internet-based knowledge sharing [15]. Data collection used questionnaires and interviews with SME managers. The statistical equations in this study are as follows: FinCL = α + β1 FinLt + β2 FinTc + β3 DC + β4 FinLtDSC + β4 FinTcDSC + ε Note: α = Constant. β = Coefficient. FinCL = Financial Inclusion. FinLt = Financial Literacy. FinTc = Financial Technology. DSC = Digital Social Capital. ε = error.
4 Results and Discussion 4.1 Respondents Profile Based on the sampling data results, it was obtained as many 205 fashion SMEs with the characteristics as follows: 36 male (17.6%) and 169 female (82.4%). This shows that the majority of respondents are female. Furthermore, 105 people with Senior High School level (51.2%), 18 people with Diploma level (8.8%), 74 people with undergraduate level (36.1%), and eight people with postgraduate level (3.9%). This shows that the majority of respondents had Senior High School level. This means that they are very productive in developing their fashion SMEs in today’s digital era. 4.2 Instrument Quality Test Results The test validity of financial literacy, financial technology, digital social capital, and financial inclusion variables showed significant values of 0.000 < 0.05. It means all variables are valid. Reliability test results show that Cronbach’s alpha value of financial literacy 0.758 > 0.05, financial technology 0.800 > 0.05, digital social capital 0.592 > 0.05 and financial inclusion 0.821 > 0.05, it can be concluded that all variables are reliable.
254
Mutamimah and P. L. Saputri
4.3 Classic Assumption Test Results Based on the classical assumption test, it showed that the normality test result had an Asymp. Sig (2-tailed) value of 0.082 > 0.05, meaning that the data is normally distributed. Based on the multicollinearity test, the financial literacy had a tolerant value of 0.498 < 1 and a Variance Inflation Factor value of 2.009 > 10, financial technology had a tolerant value of 0.551 < 1, and a Variance Inflation Factor value of 1.816 > 10, digital social capital had a tolerant value of 0.509 < 1 and a Variance Inflation Factor value of 1.966 > 10, meaning all variables passed the multicollinearity test. The results of the heteroscedasticity tests of financial literacy, financial technology, and digital social capital showed that there was no heteroscedasticity. 4.4 Descriptive Statistics Results Based on descriptive statistical analysis, financial technology showed a mean value of 14.66 between a minimum value of 4.00 and a maximum of 20.00 with a standard deviation of 4.02%. This indicates that the average value of financial technology is close to the maximum value, meaning that the average respondent has high financial technology. Financial inclusion has a mean value of 15.48 between a minimum of 4.00 and a maximum of 20.00, with a standard deviation of 3.43%. This shows that the average value of financial inclusion is close to the maximum value, meaning that the average respondent has a high financial inclusion. Financial literacy has a mean value of 11.40 between a minimum value of 5.00 and a maximum of 15.00, with a standard deviation of 2.48%. This shows that the average value of financial literacy is close to the maximum value, meaning that the average respondent has a high financial inclusion. Digital social capital has a mean value of 12.13 between a minimum value of 5.00 and a maximum of 15.00, with a standard deviation of 2.35%. This shows that the average value of digital social capital is close to the maximum value, meaning that the average respondent has a high digital social capital. 4.5 Hypothesis Test Results
Table 1. Descriptive statistics N
Minimum
Maximum
Mean
Std. Deviation
Financial Technology
205
4.00
20.00
14.66
4.02
Financial Inclusion
205
4.00
20.00
15.48
3.43
Financial Literacy
205
5.00
15.00
11.40
2.48
Digital Social Capital
205
5.00
15.00
12.13
2.36
Valid N (listwise)
205
Sources: primary data, 2021
Digital Social Capital and Financial Inclusion for Small Medium Enterprises
255
Table 2. Hypothesis test result Model
1
Unstandardized Coefficients
Standardized Coefficients Beta
t
Sig
2.113
0.036
B
Std. Error
(Constant)
8.964
4.243
Financial Literacy
0.288
0.454
0.208
0.633
0.527
Financial Technology −0.119
0.253
−0.140
−0.470
0.639
Digital Social Capital −0.100
0.358
−0.069
−0.278
0.781
Financial Literacy*Digital social capital
−0.011
0.037
−0.158
−0.286
0.775
Financial Technology*Digital social capital
0.042
0.020
0.904
2.062
0.040
R
0.735a
F
46.758
R Square
0.540
Sig
0.000b
Adjusted R Square
0.529
Std. The error of the Estimate
2.35543
a. Dependent Variable: Financial Inclusion b. Predictors: (Constant), Financial Literacy*Digital social capital, Financial Technology, Digital Social Capital, Financial Literacy, Financial Technology*Digital Social Capital
Hypothesis 1 states that financial literacy has a positive effect on financial inclusion. However, the analysis results showed that financial literacy had a coefficient value of 0.288 with a significant t-statistical value of 0.633 of 0.527 because the significance level was more significant than 0.05, so hypothesis 1 was rejected. The results showed that financial literacy did not affect financial inclusion at a significance level of 5%, meaning financial literacy could not increase financial inclusion. These results show that the level of knowledge, understanding and financial skills, and financial decision-making for fashion SMEs is still low, so it cannot increase financial access. The results of this study are not in line with some studies that stated financial literacy has a positive influence on financial inclusion [8, 9, 22]. Financial literacy owned by the community does not necessarily make them able to qualify at the analysis stage of granting access to capital from financial institutions. Financial literacy negatively affects financial inclusion [10]. Hypothesis 2 states that financial technology has a positive influence on financial inclusion. However, the analysis results showed that financial technology had a coefficient value of −0.119 with a significant t-statistical value of -0.470 of 0.639 because the significance level was more significant than 0.05, so hypothesis 2 was rejected. The results showed that financial technology did not affect financial inclusion at a significance level of 5%, meaning financial technology could not increase financial inclusion. The results of this study do not support the results of some studies that stated financial
256
Mutamimah and P. L. Saputri
technology could increase financial inclusion [22, 25, 26]. This means that the use of technology in conducting transactions and financial statements for fashion SMEs is still low. Hence, most fashion MSMEs cannot make financial statements and digital transactions optimally, so it cannot increase the ease of financial access to bank and non-bank financial institutions. Fintech does not influence financial inclusion [12]. Hypothesis 3 states that digital social capital can moderate the influence between financial literacy on financial inclusion. However, the analysis results showed that digital social capital has a coefficient value of −0.011 with a significant t-statistical value of −0.286 of 0.775. Its significance level is more significant than 0.05, so hypothesis 3 is rejected. The results showed that digital social capital could not strengthen the influence of financial literacy on financial inclusion. Digital social capital consisting of SME groups provides information for its members about various types of products and access to financial services. But this does not strengthen the existence of financial literacy to support financial inclusion because there is no opportunity for the person to have a better level of financial inclusion [32]. Thus, digital social capital cannot encourage SMEs to understand and increase financial skills in financial reports, digital business transactions, and financial decisions, so they cannot promote access to finance from banks and non-banks. Hypothesis 4 states that digital social capital can moderate the influence of financial technology on financial inclusion. The analysis results showed that digital social capital has a coefficient value of 0.042 with a significant t-statistical value of 2.062 of 0.040 since its significance level is less than 0.05, so hypothesis 4 is accepted. The results showed that digital social capital could strengthen the influence of financial technology on financial inclusion. This means that networks, knowledge sharing, and digital-based collective solutions will be more effective and efficient in increasing the use of digitalbased financial innovations, especially in financial transactions for fashion SMEs. Digital social capital is an internet-based social connection and knowledge sharing [15]. Even digital social capital can act as social collateral that facilitates financial access for banks and non-banks. Based on Table 2, the coefficient of determination (R Square) value is 0.540. This means that financial literacy, financial technology, and digital social capital can affect financial inclusion by 54%. The remaining 46% were affected by other variables not tested in the study.
5 Conclusion This study examined the role of digital social capital in moderating the influence of financial literacy and financial technology on financial inclusion. The results showed that financial literacy and financial technology did not affect financial inclusion. After using digital social capital as a moderating variable, it showed that digital social capital could not strengthen the influence of financial literacy on financial inclusion. But digital social capital was able to strengthen the impact of financial technology on financial inclusion because social capital as intangible collateral / social collateral can help SMEs to gain financial access to bank and non-bank financial institutions. This means that digital social capital can be used as a new alternative for SMEs to get financial access
Digital Social Capital and Financial Inclusion for Small Medium Enterprises
257
because digital social capital fosters trust for financial institutions against fashion SMEs when they need financial access. The managerial implication is that digital social capital can become social collateral when SMEs apply for access to capital. It needs to encourage the improvement of digital social capital and its ecosystem to reduce un-bankable communities to create financial inclusion. If SMEs get easy financial access, they can use the capital for developing their business. This research is still limited to 205 SMEs in Central Java, so future research needs to expand the research object to all provinces in Indonesia or other countries. In addition, this study only used financial literacy and fintech. It is hoped that future research can add other independent variables in influencing financial inclusion, which is then examined with other moderation variables, such as digital entrepreneurial networking.
References 1. El-Said, H., Al-Said, M., Zaki, C.: Access to finance and financial problems of SMEs: evidence from Egypt. Int. J. Entrep. Small Bus. 20(3), 286–309 (2013). https://doi.org/10.1504/IJESB. 2013.056891 2. Yoshino, N., Taghizadeh-Hesary, F.: Major challenges facing small and medium-sized enterprises in Asia and solutions for mitigating them (2016). https://doi.org/10.2139/ssrn.276 6242 3. Mutamimah, Hendar: Integrated financial technology model on financing decision for small medium enterprises development. In: 17th International Symposium on Management (INSYMA 2020) (2020), vol. 115, pp. 124–128. https://doi.org/10.2991/aebmr.k.200127.025 4. Aaron, M., Rivadeneyra, F., Sohal, S.: Fintech: is this time different? a framework for assessing risks and opportunities for central banks. In: Bank of Canada Staff Discussion Paper, pp. 1–32 (2017) 5. Financial Stability Board, “Financial stability implications from fintech: Supervisory and regulatory issues that merit authorities’ attention,” in Financial Stability Board, pp. 1–61 (2017) 6. Lusardi, A., Mitchell, O.S.: The economic importance of financial literacy: theory and evidence. J. Econ. Lit. 52(1), 5–44 (2014). https://doi.org/10.1257/jel.52.1.5 7. Financial Service Authority, “Regulation of Financial Service Authority No.76/PJOK.07/2016 about increasing financial literature and inclussion in the financial services sector for consumers and/or the community,” in Financial Service Authority, pp. 1–18 (2016) 8. Bongomin, G.O.C., Ntayi, J.M., Malinga, C.A.: Analyzing the relationship between financial literacy and financial inclusion by microfinance banks in developing countries: social network theoretical approach. Int. J. Sociol. Soc. Policy 40(11–12), 1257–1277 (2020). https://doi.org/ 10.1108/IJSSP-12-2019-0262 9. Grohmann, A., Klühs, T., Menkhoff, L.: Does financial literacy improve financial inclusion? cross country evidence. World Dev. 111, 84–96 (2018). https://doi.org/10.1016/j.worlddev. 2018.06.020 10. Allen, F., Demirguc-Kunt, A., Klapper, L., Martinez Peria, M.S.: The foundations of financial inclusion: understanding ownership and use of formal accounts. J. Financ. Intermediation 27, 1–30 (2016). https://doi.org/10.1016/j.jfi.2015.12.003 11. Nurohman, Y.A., Kusuma, M., Narulitasari, D.: Fintech, financial inclusion, and sustainability: a quantitative approach of muslims SMEs. Int. J. Islam. Bus. Ethics 6(1), 54–67 (2021). https://doi.org/10.30659/ijibe.6.1.54-67
258
Mutamimah and P. L. Saputri
12. Efan, K., Basana, S.R., Ottemoesoe, R.S.D.: The effect of financial inclusion on micro, small, and medium enterprise performance using fintech as a moderating variable. In: Contemporary Research on Business and Management– Noviaristanti (Ed.), pp. 153–156 (2020) 13. Baron, R.M., Kenny, D.A.: The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. J. Pers. Soc. Psychol. 51(6), 1173–1182 (1986). https://doi.org/10.1037/0022-3514.51.6.1173 14. Bongomin, G.O.C., Munene, J.C., Ntayi, J.M., Malinga, C.A.: Exploring the mediating role of social capital in the relationship between financial intermediation and financial inclusion in rural Uganda. Int. J. Soc. Econ. 45(5), 828–846 (2018). https://doi.org/10.1108/IJSE-082017-0357 15. Jafari, S.M., Moharrami, R.: Conceptualizing digital social capital. In: International Seminar on New Topics in Business Management Varna, Bulgaria 6 June 2019, pp. 1–12 (2019) 16. Kistyanto, A.: The digital social capital to career success and its implication to industrial revolution 4.0. In: Proceedings of the 2nd International Conference on Business and Management of Technology (ICONBMT 2020), vol. 175, pp. 97–101 (2021). https://doi.org/10. 2991/aebmr.k.210510.018 17. President of Republic Indonesia, “Regulation of President Republic Indonesia No. 82 of 2016 about Financial inclusive national strategy,” in Regulation of President Republic Indonesia, pp. 1–7 (2016) 18. Financial Services Authority: Indonesian financial literacy national strategy. In: Financial Services Authority, pp. 1–128 (2021) 19. Lakuma, C.P., Marty, R., Muhumuza, F.: Financial inclusion and micro, small, and medium enterprises (MSMEs) growth in Uganda. J. Innov. Entrepreneurship 8(1), 1–20 (2019). https:// doi.org/10.1186/s13731-019-0110-2 20. Beck, T., Demirgfic-Kunt, A., Maksimovic, V.: Financial and legal constraints to firm growth: does size matter? In: The World Bank Development Research Group Finance, pp. 1–58 (2002) 21. Lusardi, A., Tufano, P.: Debt literacy, financial experiences, and overindebtedness (2009). https://doi.org/10.1017/S1474747215000232 22. Ali, M.M., Devi, A., Furqani, H., Hamzah, H.: Islamic financial inclusion determinants in Indonesia: an ANP approach. Int. J. Islam. Middle East. Financ. Manag. 13(4), 727–747 (2020). https://doi.org/10.1108/IMEFM-01-2019-0007 23. Awrey, D.: Toward a supply-side theory of financial innovation. J. Comp. Econ. 41(2), 401– 419 (2013). https://doi.org/10.1016/j.jce.2013.03.011 24. Financial Stability Board: Fintech credit : Market structure, business models and financial stability implications. In: Financial Stability Board, pp. 1–48 (2017) 25. Ozili, P.K.: Theories of financial inclusion. In: Uncertainty and Challenges in Contemporary Economic Behaviour, pp. 89–115 (2020) 26. Lubis, A., Dalimunthe, R., Situmeang, C.: Antecedents effect of financial inclusion for the people of North Sumatera. Budapest Int. Res. Critics Inst. Humanit. Soc. Sci. 2(4), 401–408 (2019). https://doi.org/10.33258/birci.v2i4.602 27. Putnam, R.D.: Making Democracy Work, pp. 1–181. Princeton University Press (1993) 28. Grootaert, C., Van Bastelar, T.: Understanding and measuring social capital : a multidisciplinary tool for practitioners. In: The International Bank for Reconstruction and Development/The World Bank, pp. 1–265 (2002) 29. Bongomin, G.O.C., Ntayi, J.M., Munene, J.C., Nabeta, I.N.: Social capital: mediator of financial literacy and financial inclusion in rural Uganda. Rev. Int. Bus. Strateg. 26(2), 291–312 (2016). https://doi.org/10.1108/RIBS-06-2014-0072 ˇ 30. Cihák, M., Demirgüç-Kunt, A., Feyen, E., Levine, R.: Benchmarking financial systems around the world (2012). https://doi.org/10.1596/9780821395035_ch01
Digital Social Capital and Financial Inclusion for Small Medium Enterprises
259
31. Financial Service Authority: Regulation of the Financial Service Authority No.13 /POJK.02/2018 about digital financial innovation in the financial services sector. In: Financial Service Authority, Republic Indonesia, pp. 1–16 (2018) 32. Arifin, A.Z., Yanuar, Y., Sawitri, N.N.: Financial inclusion on small medium enterprise. In: 16th International Symposium on Management (INSYMA 2019), vol. 308, pp. 21–23 (2019). https://doi.org/10.2991/insyma-19.2019.6
Knowledge Absorptive Capacity Toward Sustainable Organizational Reputation in Digital Era Mulyana Mulyana(B) and Erlinda Ramadhani Permata Putri Department of Management, Universitas Islam Sultan Agung, Semarang, Indonesia [email protected], [email protected]
Abstract. Business relationships by leveraging digital platforms can make it easier for organizations to communicate, share information and absorb knowledge. This paper will try to integrate the relationship between variables that affect the creation of sustainable organizational reputation. Relational capability and information communication technology (ICT) can help organizations to share information and knowledge. Likewise, knowledge absorptive capacity from external sources is seen as being able to encourage the speed of organizational innovation that has the potential to improve organizational performance. Furthermore, the better the performance of the organization, it will create a sustainable organizational reputation. This paper still requires further empirical studies, especially applied to the object of SMEs in order to obtain the truth of the concept developed. Keywords: ICT · Organizational innovation · Knowledge absorptive capacity · Sustainable organizational reputation
1 Introduction The business environment is changing so dynamically that organizations need speed in adopting knowledge from external sources. The speed and ability to adapt knowledge are very important to support innovation and company performance [1]. Knowledge absorptive capacity is the ability to utilize knowledge from external sources, through acquisition, assimilation, transformation, and exploitation [2]. Organizations that have knowledge absorptive capacity will be able to create a competitive advantage [3, 4]. Absorptive capacity is part of the company’s decision to allocate resources for innovative activities [5]. Knowledge absorptive capacity is needed to support organizational innovation which is considered capable of improving organizational performance [5]. Organizational innovation to realize sustainable organizational reputation needs ICT support and relational capability. The integration of ICT, relational capability, and knowledge absorptive capacity is seen as capable of supporting the realization of organizational innovation and having an impact on sustainable organizational reputation. Previous researchers have examined absorptive capacity with the main focus on information technology [6] and knowledge creation [7]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 260–268, 2022. https://doi.org/10.1007/978-3-031-08812-4_25
Knowledge Absorptive Capacity
261
Studies on digitization show that there are still obstacles when implementing digitization in business practices, namely lack of digital capabilities [8, 9], and sometimes the benefits of digitizing performance are less clear [10], so accuracy in utilizing digital platforms in business practice is very important to create a competitive advantage. This study tries to build the concept of the relationship between ICT, relational capability and knowledge absorptive capacity to support organizational innovation and create sustainable organizational reputation. The conceptual model developed still requires further empirical evidence.
2 Literature Review 2.1 Sustainable Organizational Reputation There is diversity in the definition of the concept of organizational reputation and its measurement. Corporate reputation as a collective assessment of the company’s ability to provide valuable results to a group of stakeholders [10]. Furthermore, the company’s reputation is measured through indicators of ability to attract, develop & retain top talent, ability to cope with changing economic environment, financial soundness, long-term investment value/potential for future profit, quality of management, quality of products & services, innovativeness, and social responsibility. Company reputation is the overall evaluation of stakeholders of the company overtimes [11]. Organizational reputation is the overall evaluation by customers of the company based on their reactions to goods, services, communication activities and interactions with the company [12]. Organizational reputation includes five dimensions: customer orientation, good employer, reliable and financially strong company, product and service quality, social and environmental responsibility. Therefore, sustainable organizational reputation is a collective assessment of the company’s ability to deliver the results achieved to stakeholders in a sustainable manner, which is measured through indicators of the ability to cope with changes in the economic, financial environment, future profit potential, and management quality. 2.2 Relational Capability and Knowledge Absorptive Capacity Relational capability can be developed through relational capacity, namely the speed of access to knowledge, innovation, organizational support, and the ability to coordinate and communicate [18]. Knowledge absorptive capability is built to achieve superior organizational performance [14]. Absorptive capacity supports organizational innovation as an effort to create organizational value [5]. Exploitation capacity as a dimension of knowledge absorptive capacity shows the use of knowledge for commercial purposes that allows the creation of new organizational capabilities [2]. Therefore, if the relational capability with partners is getting better, it will be easy to increase knowledge absorptive capacity.
262
M. Mulyana and E. R. P. Putri
2.3 ICT and Knowledge Absorptive Capacity Information Communication Technology (ICT) is considered by organizations as a means of creating added value, especially increasing productivity and growth [24]. The use of telephone, internet, cellular technology in business practice as a form of face-toface with business partners [25, 26], and the use of digital platforms that are in accordance with organizational needs are very important in order to improve their performance. Knowledge absorptive capability is built to achieve superior organizational performance [14]. Absorptive capacity supports organizational innovation as an effort to create organizational value [5]. Furthermore, knowledge absorptive capacity is grouped into four dimensions: acquisition, assimilation, transformation and exploitation [2]. Acquisition capacity is the identification and acquisition of external knowledge through the intensity, speed and direction of the organization’s efforts. Assimilation capacity consists of understanding external knowledge gained by turning it into organizational routines. Transformation capacity is a combination of assimilation and exploitation dimensions to transform external knowledge into organizational routines with the aim of applying knowledge through adaptation and organizational needs. Exploitation capacity is the use of knowledge for commercial purposes that allows the creation of new organizational capabilities. Organizations that master ICT will easily increase their knowledge absorptive capacity. 2.4 Relational Capability and Organizational Innovation Relational capability can be built through two approaches: 1) relational capacity includes speed of access to knowledge, innovation and organizational support, 2) coordination and communication capabilities with partners [18]. Relational capability is the ability to coordinate and communicate with partners business in order to create long-term relationships that have the potential to improve business performance [19]. Relational capability affects the quality of relationships with partners and has an impact on business performance [20]. Relational capability is demonstrated through the organization’s ability to interact, create trust and commitment relationship with clients [21]. Organizations that are able to build good relationships with partners will share knowledge and information to support organizational innovation. 2.5 ICT and Organizational Innovation The development of ICT has changed business patterns [6]. Utilization of ICT has been recognized to be the key to corporate growth [22], and ICT applications can help to track the role of organizational functions [23]. ICT is seen as capable of creating added value at various levels of the company which leads to increased productivity and growth [24]. Business practices by utilizing telephone, internet, cellular technology to change face-to-face with business partners [25, 26], so that accuracy in choosing and using digital platforms for business activities is very important in order to improve company performance. On the other hand, companies that have the ability to develop technology will tend to have superior performance [27]. Furthermore, technological capability is needed when organizations develop new products using new technologies
Knowledge Absorptive Capacity
263
to meet dynamic market needs [28]. Innovation is understood as a means of changing organizations to respond to environmental changes [15]. Innovation is a process that starts from ideas and results development in the form of new products, processes and services [16]. Innovation involves the adoption of new products or processes to increase competitiveness and overall profitability. Organizations that are good at ICT will easily get new knowledge to support strengthening innovation. 2.6 Knowledge Absorptive Capacity and Organizational Innovation Absorptive capacity is the organization’s ability to acquire, absorb, transform, and utilize external knowledge [2]. In addition, absorptive capacity is an important factor that contributes positively to the knowledge transfer process [29]. Absorptive capacity allows the use of a large external network and has an effect on improving innovation performance [30]. Absorptive capacity as the company’s ability to acquire, transform and utilize knowledge from outside [31]. Furthermore, knowledge absorptive capacity and organizational innovation can improve organizational performance [6]. Therefore, knowledge absorptive capacity that is built through acquisition, assimilation, transformation and exploitation can increase the organization’s ability to absorb new knowledge from outside and potentially increase organizational innovation. 2.7 Organizational Innovation and Sustainable Organizational Reputation Innovation is defined as an organization’s ability to generate, accept and implement new ideas, processes, products or services [32]. Organizational innovation occurs due to pressure from the external environment, such as competition, deregulation, resource scarcity, and customer demand [33]. Organizations adopt innovations to ensure that the organization is adaptive and makes changes to maintain or improve its performance [34]. Companies facing environmental challenges and uncertainties can achieve superior performance when synergizing technical and managerial innovation in their organizational structure [35]. Organizational reputation is a collective assessment of the company’s ability to provide valuable results to a group of stakeholders [36]. Furthermore, the company’s reputation measurement can use indicators of ability to attract, develop & retain top talent, ability to cope with changing economic environment, financial soundness, long-term investment value/potential for future profit, quality of management, quality of products & services, innovativeness, social responsibility. Therefore, organizational innovation that is carried out continuously and directed in accordance with the dynamics of the environment will be able to create a sustainable organizational reputation. 2.8 Relational Capability, Knowledge Absorptive Capability and Sustainable Organizational Reputation Focus on customers is very important for companies in facing business competition [37], and ensuring the creation of customer loyalty [38, 39]. The company’s ability to provide the best value for customers is the main key to product performance [40]. Therefore, companies must be able to develop long-term relationships with customers on the basis of
264
M. Mulyana and E. R. P. Putri
shared satisfaction [41]. The ability to build harmonious relationships with partners has an important role for knowledge absorptive capacity. Knowledge from external sources is a driving force for the creation of absorptive capability [42]. Absorptive capability includes the way to exploit, integrate and implement new ideas within the organization [43]. Likewise, customer orientation is a driving force for product innovation and harmonious relationships with customers enable companies to provide the best value for customers and contribute to company performance [44]. Therefore, the company’s success in building good relationships with partners will be the driving force for the creation of knowledge absorptive capability and organizational innovation that allows the realization of a sustainable organizational reputation. 2.9 ICT, Knowledge Absorptive Capability and Sustainable Organizational Reputation ICT is one of the determinants for the sustainability of the company’s growth [22], and is able to create added value that leads to increased productivity and company growth [24]. Likewise, the right use of e-commerce can expand market reach, access new customers and create cost efficiency for the company [45]. Digital marketing has an important role to create harmonious relationships and respond quickly to customer needs that have an impact on company performance [46]. E-business can be used to share information and knowledge efficiently without any distance limitations and is positively related to organizational innovation [47]. Knowledge absorptive capacity includes four dimensions: acquisition, assimilation, transformation and exploitation [48]. Knowledge absorptive capacity and organizational innovation are able to drive organizational performance [17]. Furthermore, exploratory and exploitative learning skills can help companies identify, evaluate and select information and technology to be adopted [49]. The higher the ability of information technology, the higher the possibility of the company to be involved in the exploration of innovation [50]. Organizational reputation is the overall evaluation by customers of the company on the basis of their reactions to goods, services, communication activities and interactions with the company [12]. Therefore, the proper use of ICT and relational capability will be the driving force for knowledge absorptive capability and organizational innovation. Furthermore, the increase in innovation performance will have an impact on organizational performance which allows the creation of a sustainable organizational reputation. The conceptual model developed is presented in Fig. 1 as follows: The organization’s ability to build harmonious relationships with business partners will create trust and relationship commitment, thereby facilitating access to knowledge to support innovation and absorb knowledge. Likewise, ICT makes it easier for organizations to communicate and interact with business partners to share the knowledge needed for innovation. Organizations that are able to have strong ICT and are able to build relationships with business partners will be able to absorb knowledge well and will encourage organizational innovation activities. The success of organizational innovation will create competitive advantage and superior performance and have an impact on sustainable organization reputation.
Knowledge Absorptive Capacity
265
Relational Capability Knowledge Acquisition
Knowledge Assimilation
Knowledge Transformation
Knowledge Absorptive Capability
Organizational Innovation
Sustainable Organizational Reputation
Knowledge Exploitation
ICT
Fig. 1. Conceptual model
3 Conclusion Relational capability and ICT have an important role to encourage knowledge absorptive capacity and organizational innovation. Furthermore, knowledge absorptive capacity includes the dimensions of knowledge acquisition, knowledge assimilation, knowledge transformation and knowledge exploitation. Knowledge absorptive capacity that is effective and efficient will be a driving force for the growth of organizational innovation and the potential for the creation of sustainable organizational reputation. Furthermore, empirical testing requires a more detailed approach. Researchers still need to explore the indicators of the research variables proposed in the conceptual model. After exploring all the indicators, a factor analysis test was conducted to find groups of indicators forming the variables. The last, it needs testing the relationship among variables by using the Structural Equation Modelling (SEM) as for determining the significance of the relationship among the variables.
References 1. Mahmoud-jouini, S.B.: Absorptive capacity and source-recipient complementarity in designing new products: an empirically derived framework. J. Prod. Innov. Manag. 25(5), 473–490 (2008, forthcoming) 2. Zahra, S.A.: Absorptive capacity: a review, reconceptualization, and extension. Acad. Manag. Rev. 27(2), 185–203 (2002) 3. Jansen, J.J.P., Van Den Bosch, F.A.J., Volberda, H.W.: Managing potential and realized absorptive capacity: how do organizational antecedents matter? Rep. Ser. Acad. Manag. J. 46(6), 1–41 (2005) 4. Vonderembse, M.A., Ragu-Nathan, T.S., Sharkey, T.W.: Absorptive capacity: enhancing the assimilation of time-based manufacturing practices. J. Oper. Manag. 24(5), 692–710 (2006). https://doi.org/10.1016/j.jom.2005.05.004
266
M. Mulyana and E. R. P. Putri
5. Cohen, W.M., Levinthal, D.A.: Absorptive capacity: a new perspective on learning and innovation absorptive capacity: a new perspective on learning and innovation. Adm. Sci. Q. 35(May), 128–152 (1990). https://doi.org/10.2307/2393553 6. Liu, H., Wei, K.K.: Effects of supply chain integration and market orientation on firm performance Evidence from China. Int. J. Oper. Prod. Prod. Manag. 33(3), 322–346 (2013). https:// doi.org/10.1108/01443571311300809 7. Su, Z., Ahlstrom, D., Li, J., Cheng, D.: Knowledge creation capability, absorptive capacity, and product innovativeness. R D Manag. 43(5), 473–485 (2013). https://doi.org/10.1111/ radm.12033 8. Day, S.: Closing the marketing capabilities gap. J. Mark. 75(July), 183–195 (2011) 9. Kane, Palmer, D., Phillips, A.N., Kiron, D.: Is your business ready for a digital future?. MIT Sloan Manag. Rev. 56(4), 37–44 (2015) 10. Peter, Leeflang, S.H., Verhoefa, P.C., Dahlström, P.: Challenges and solutions for marketing in a digital era. Eur. Manag. J. 32, 1–12 (2014) 11. Wilson, M.G.: Management decision. Corp. Reput. Manag. Living Brand Decis. 39(2), 99– 104 (2001) 12. Walsh, G., Beatty, S.E.: Customer-based corporate reputation of a service firm: scale development and validation. J. Acad. Mark. Sci. 35, 127–143 (2007). https://doi.org/10.1007/s11 747-007-0015-7 13. Camisón, C., Forés, B.: Knowledge absorptive capacity: new insights for its conceptualization and measurement ✩. J. Bus. Res. 63(7), 707–715 (2010). https://doi.org/10.1016/j.jbusres. 2009.04.022 14. Cumming, B.S.: Innovation overview and future challenges innovation overview. Eur. J. Innov. Manag. 1(1), 21–29 (1998) 15. Thornhill, S.: Knowledge, innovation and firm performance in high- and low-technology regimes. J. Bus. Ventur. 21(5), 687–703 (2006). https://doi.org/10.1016/j.jbusvent.2005. 06.001 16. Hitt. M.A. (Guest ed.): Introduction to the special issue strategic entrepreneurship: entrepreneurial strategies for wealth creation. Strateg. Manag. J. 491, 479–491 (2001). https:// doi.org/10.1002/smj.196 17. Kittikunchotiwut, P.: The role of social capital on absorptive capacity and organizational innovation. J. Bus. Retail Manag. Res. (October), 27–39 (2015) 18. Smirnova, M., Naudé, P., Henneberg, S.C., Mouzas, S., Kouchtch, S.P.: Industrial Marketing Management The impact of market orientation on the development of relational capabilities and performance outcomes: the case of Russian industrial firms. Ind. Mark. Manag. 40(1), 44–53 (2011). https://doi.org/10.1016/j.indmarman.2010.09.009 19. Nuryakin, N., Aryanto, V.D.W., Setiawan, M.B.: Mediating effect of value creation in the relationship between relational capabilities on business performance. Contaduria y Adm. 63(1), 1–21 (2018). https://doi.org/10.22201/fca.24488410e.2018.1178 20. Rungsithong, R., Meyer, K.E., Roath, A.S.: Relational capabilities in Thai buyer-supplier relationships. J. Bus. Ind. Mark. 32(8), 1228–1244 (2017). https://doi.org/10.1108/JBIM-022017-0027 21. Farida, N.: Network capability, relational capability and Indonesian manufacturing SME performance: an empirical analysis of the mediating role of product innovation. Eng. Manag. Prod. Serv. 13(1), 41–52 (2021). https://doi.org/10.2478/emj-2021-0003 22. Shao, J., Taisch, M., Ortega-Mier, M.: A grey-DEcision-MAking Trial and Evaluation Laboratory (DEMATEL) analysis on the barriers between environmentally friendly products and consumers: practitioners’ viewpoints on the European automobile industry. J. Clean. Prod. 112, 3185–3194 (2016). https://doi.org/10.1016/j.jclepro.2015.10.113
Knowledge Absorptive Capacity
267
23. Verdouw, C.N., Wolfert, J., Beulens, A.J.M., Rialland, A.: Virtualization of food supply chains with the internet of things. J. Food Eng. 176, 128–136 (2016). https://doi.org/10.1016/j.jfo odeng.2015.11.009 24. Rezagholizadeh, M.: The impact of information and communication technology (ICT) on economic growth in the OIC countries. Econ. Environ. Stud. 17(2), 255–276 (2017) 25. Yoo, Y., Boland, R.J., Lyytinen, K., Majchrzak, A.: Organizing for innovation in the digitized world. Organ. Sci. 23(5), 1398–1408 (2012). https://doi.org/10.1287/orsc.1120.0771 26. Coreynen, W., Matthyssens, P., Van Bockhaven, W.: Boosting servitization through digitization: pathways and dynamic resource configurations for manufacturers. Ind. Mark. Manag. 60, 42–53 (2016). https://doi.org/10.1016/j.indmarman.2016.04.012 27. Lavie, D., Rosenkopf, L.: Balancing exploration and exploitation in alliance formation. Acad. Manag. J. 49, 797–818 (2006) 28. Hsieh, Tsai, K., Wang, J.: The moderating effects of market orientation and launch proficiency on the product advantage – performance relationship. Ind. Mark. Manag. 37, 580–592 (2008). https://doi.org/10.1016/j.indmarman.2007.03.003 29. Soh, P.H., Roberts, E.B.: Technology alliances and networks: an external link to research capability. IEEE Trans. Eng. Manag. 52(4), 419–428 (2005). https://doi.org/10.1109/TEM. 2005.850727 30. Powell, W.W., Koput, K.W., Smith-doerr, L.: Interorganizational and the collaboration locus of innovation: networks of learning in biotechnology. Adm. Sci. Q. 41, 116–145 (1996) 31. Duchek, S.: Designing absorptive capacity? An analysis of knowledge absorption practices in German high-tech firms. 19(4), 1–22 (2015). https://doi.org/10.1142/S1363919615500449 32. Hurlay, R.F.: Innovation, market orientation and organizational learning: an integration and empirical examination, 42–54 (1998) 33. Mahmoud, M.A., Blankson, C., Owusu-frimpong, N., Nwankwo, S.: Market orientation, learning orientation and business performance the mediating role of innovation. Int. J. Bank 34(5), 623–648 (2016). https://doi.org/10.1108/IJBM-04-2015-0057 34. Damanpour, Walker, R.M., Avellaneda, C.N.: Combinative effects of innovation types and organizational performance: a longitudinal study of service organizations. J. Manag. Stud. 46(June), 650–675 (2009). https://doi.org/10.1111/j.1467-6486.2008.00814.x 35. Han, J.K., Kim, N.: Market orientation and organizational performance: is innovation a missing link? Market orientation and organizational performance: is innovation a missing link? J. Mark. 62(4), 30–45 (1998) 36. Fombrun, C.J., Gardberg, N.A., Sever, J.M.: The reputation quotient: a multi-stakeholder measure of corporate reputation. Nat. Publ. Gr. 7(4), 241–255 (2000). https://doi.org/10. 1057/bm.2000.10 37. Payne, A., Frow, P.: For customer framework relationship management. J. Mark. 69(4), 167– 176 (2005) 38. Kumar, V., Shah, D.: Building and sustaining profitable customer loyalty for the 21st century. J. Retail. 80, 317–330 (2004). https://doi.org/10.1016/j.jretai.2004.10.007 39. Verhoef, P.C.: Understanding the effect of efforts on customer retention and customer share development. 67(October), 30–45 (2003) 40. Kumar, N., Scheer, L.: From market driven to market driving. Eur. Manag. J. 8(April), 129–142 (2000) 41. Reinartz, W.J., Kumar, V.: On the profitability of long-life customers in a noncontractual setting: an empirical investigation and implications for marketing. J. Mark. 64(4), 17–35 (2000). https://doi.org/10.1509/jmkg.64.4.17.18077 42. Kostopoulos, Papalexandris, A., Papachroni, M., Ioannou, G.: Organizational unit ACAP: social context and performance absorptive capacity, innovation, and financial performance. J. Bus. Res. 64(12), 1335–1343 (2007). https://doi.org/10.1016/j.jbusres.2010.12.005
268
M. Mulyana and E. R. P. Putri
43. Lichtenthaler, U.: Absorptive capacity, environmental turbulence, and the complementarity of organizational learning processes. Acadey Manag. J. 52(4), 822–846 (2009) 44. Augusto, M., Coelho, F.: Market orientation and new-to-the-world products: exploring the moderating effects of innovativeness, competitive strength, and environmental forces. Ind. Mark. Manag. (2007). https://doi.org/10.1016/j.indmarman.2007.09.007 45. Kaynak, E., Tatoglu, E., Kula, V.: An analysis of the factors affecting the adoption of electronic commerce by SMEs: Evidence from an emerging market. Int. Mark. Rev. 22(6), 623–640 (2005). https://doi.org/10.1108/02651330510630258 46. Wang, F.: Digital marketing capabilities in international firms: a relational perspective. Int. Mark. Rev. 37(3), 559–577 (2020). https://doi.org/10.1108/IMR-04-2018-0128 47. Soto-Acosta, P., Popa, S., Palacios-Marqués, D.: E-business, organizational innovation and firm performance in manufacturing SMEs: an empirical study in Spain. Technol. Econ. Dev. Econ. 22(6), 885–904 (2016). https://doi.org/10.3846/20294913.2015.1074126 48. Zahra, S.A.: Harvesting family firms ‘organizational social capital: a relational perspective Shaker A. Zahra. J. Manag. Stud. 47(2), 345–366 (2010). https://doi.org/10.1111/j.14676486.2009.00894.x 49. Lin, H.E., McDonough, E.F., Lin, S.J., Lin, C.Y.Y.: Managing the exploitation/exploration paradox: the role of a learning capability and innovation ambidexterity. J. Prod. Innov. Manag. 30(2), 262–278 (2013). https://doi.org/10.1111/j.1540-5885.2012.00998.x 50. Rosenkopf, L., Nerkar, A.: Beyond local search: boundary-spanning, exploration, and impact in the optical disk industry. Strateg. Manag. J. 306(October 2000), 287–306 (2001). https:// doi.org/10.1002/SMJ.160
The Role of Holistic Value Co-creation Capability in Improving Sustainable Relationship Ken Sudarti(B) , Wasitowati, and Ari Pranaditya Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia {kensudarti,wasitowati}@unissula.ac.id, [email protected]
Abstract. This study examines a model that links Holistic Value Co-Creation Capability (HVCC) with Sustainable Relationship (SR). HVCC is the perception of Sharia insurance customers that Sharia insurance salesmen can participate actively, to create not only transactional values but also religious values. This study was conducted on 177 sharia insurance customers in Central Java. The sampling technique used was purposive sampling. The regression analysis technique is used to observe the relationship between variables using SPSS version 16 software. The results showed the perception of sharia insurance customers on the ability of salesmen to be actively involved in creating holistic value co-creation had been proven to strengthen customers’ desire to establish long-term relationships. The capability of the “people” element in insurance services can be used as a solid basis as a unique and inimitable differentiation for the sustainability of the service industry that offers religious products. The results of this study have succeeded in completing the concept of value co-creation by internalizing religious values, especially Islamic values. Keywords: Holistic Value Co-creation Capability · Sustainable relationship
1 Introduction This study focuses on value creation when frontline staff and customers meet in interactive marketing activities due to the inherent inseparability of services. The concept of value creation is the most significant factor for the company’s success and has been believed to be an essential source of competitive advantage [1]. This concept is a derivative of the Service-Dominant Logic concept, which involves producers, consumers, suppliers, and other stakeholders in the service system. These groups integrate various resources and collaborate by sharing to establish a co-creation value system [2]. [3] stated that in an increasingly saturated market condition and limited resources, companies should no longer focus on optimizing internal resources but must be able to explore external resources, including involving customers in the co-creation value. Collaboration between internal and external resources results in optimal value creation. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 269–279, 2022. https://doi.org/10.1007/978-3-031-08812-4_26
270
K. Sudarti et al.
Co-creation Value can be developed from 4 values, namely functional, social, emotional, and economic values [4]. Another opinion was expressed by [5–7]. Based on the literature review on value co-creation, the search for meaning in the value creation process proposed by previous researchers has not reached the religious aspect. Religious values are essential in consumer buying behavior [8], including halal products. Therefore, this study aims to refine the dimensions of the value co-creation concept by adding one more dimension, namely Religious Value Co-Creation (RVCC), which is defined as the interaction between salesmen and insurance customers to establish co-creation value by reinforcing mutual beliefs and Sharia product knowledge. The religious dimension becomes very important when an organization offers products based on religious values. This dimension can presumably improve the interest of Islamic insurance customers to establish sustainable relationships. Through the additional religious dimension, the co-creation values will strengthen the beliefs of both parties; thus, they can provide wider benefits. This uniqueness is believed to bring strong differentiation through values superiority that has reached the level of non-transactional motives; therefore, it strengthens sustainable mutual relationships.
2 Literature Review The theory of value (TOV) is the foundation of Service-Dominant Logic (SDL) which then derives the concept of Value Co-Creation (VCC) as a value construction. Value is customer-centric and is co-created by the company and the customers. [2] stated that value co-creation can be started by creating meaning through interaction, collaboration, mutual exchange, evaluation of job performance, and resource integration. SDL postulates that when customers engage in exchanging shared values, they actively create meaning from the process, thereby increasing value [7]. Religious value co-creation is value creation related to religious values. The value obtained from religion is related to its religious commitment [8]. The religious commitment shows how far a person believes in his religious values and practices them in everyday life, including the desire to do da’wah through the buying process. Da’wah has the potential to form harmony between humans to create group cohesiveness [9]. Da’wah contains elements of two-way communication. When someone does da’wah, he not only spreads religious values but will mutually get feedback from his da’wah material. Therefore, RVCC contains elements of contribution and elements of collaboration. The command to give charity is contained in the Holy Qur’an. An-Nissa verse 114. And among the most important alms is the charity of knowledge. As stated by the Prophet Muhammad narrated by Ibn Majah: “The most important charity is when a Muslim learns a knowledge, then teaches it to other Muslim brothers.“ 2.1 Sustainable Relationships Individuals are willing to carry out sustainable relationships due to their engagement with the products they consume and with the organization. The concept of engagement is derived from the partnership theory developed by [10]. Engagement is defined as a state of being involved, focused, fully centered, or captivated by something, so that it grabs
The Role of Holistic Value Co-creation Capability
271
his attention and creates attraction [11]. Engagement is a unidimensional concept that involves emotional, cognitive, and behavioral aspects. Individuals who wish to establish long-term relationships will be more ready to get involved and participate because there are customer interactions and creative experiences with the company [11]. [12] stated that when a relationship is considered satisfying and has an emotional bond, it will strengthen the level of engagement. Nowadays, customers are no longer seen as passive recipients of marketing cues but as proactive parties to create shared value. Customer engagement will create experiences and co-creation value and contribute to innovation [13]. Customers who are willing to maintain a relationship with an organization are satisfied and loyal and want to recommend the company’s products and services to others [14]. This statement is reinforced by [15], which stated that a good relationship that is carried out through customer-company interactions contributes to creating satisfied customers who will not only make repeat purchases but also be committed to the company and recommend products and services to other customers. The indicators used to measure Sustainable Relationships are: willing to continue the Relationship, Being the first choice, having no desire to move, willing to increase involvement. 2.2 Holistic Value Co-Creation (HVCC) A co-creation value will be completely valuable when it contains two kinds of benefit, namely the benefit of the world and the hereafter, by involving religious elements. However, prior researchers’ value co-creation has not reached religious aspects. Therefore, this study attempts to complement the value co-creation concept offered by [6], namely religious value co-creation through a religious dimension. Hopefully, by adding a religious dimension, this concept will become broadly and holistically meaningful. Functional Value Co-creation Capability (FVCC). Functional value is related to how far the product has the desired benefit [4]. This value is based on the assumption that individuals are rational problem solvers [16], including their members’ information needs, leading to financial savings and high quality of services. The FVCC in this study is more directed to frontline staff’s ability to actively participate in creating functional value of halal products to be more beneficial, qualified, and innovative [17]. Therefore, when consumers believe that frontline staff can interact with them to strengthen their belief in the benefits of halal products, consumers will be more confident in their choice and willing to continuously use the products in a long-term period. Based on this explanation, the proposed hypothesis is: H1: The higher the functional value co-creation capability, the stronger the sustainable relationship. Social value co-creation capability (SVCC). It is a process where the service providers and customers create social value for themselves through mutually beneficial dialogue and interactions. Social value is a perceived utility obtained from an alternative association with more particular social groups [18] (Sheth et al. 1991). Efforts to establish social identity are related to how individuals see themselves in product choices according to their social identity, especially for a more prominent identity. For the consumers, the prominent self-identity role will determine the consumption based on social
272
K. Sudarti et al.
expectations of a particular identity [19], such as feeling accepted, how an individual is perceived, impression to others, and social approval [4]. SVCC, in this study, is more directed to the salesman’s capability of actively participating in assuring the consumers that there are many consumers in their groups who use halal products. It will further strengthen their desire to continue their relationship for a long-term period, and there is no hesitation from themselves to move to other products. Based on these explanations, the proposed hypothesis is: H2: The higher the social value co-creation capability, the stronger the sustainable relationship. Emotional Value Co-creation Capability (EmVCC). This emotional value is related to customers’ pleasure and enjoyment from using products and services [18]. Since the product is specifically designed to provide enjoyment, the emotional value derived from using products is likely to affect the significance of product user identity and probably recommend the product to others [19]. This emotional value concerns the enjoyment in use, the desire to continue using the product, feeling comfortable, protected, safe, pleased, relieved, and proud of using the products [20]. In this study, EmVCC is more focused on the salesman’s capability to actively improve consumers’ confidence that halal products will make them feel protected and comfortable. This perception will strengthen consumers’ desire to make halal products their primary choice and not move to consume non-halal products. Based on this explanation, the proposed hypothesis is: H3: The higher the emotional value of co-creation, the stronger the sustainable Relationship. Economic Value Co-Creation capability (EVCC). The economic dimension of customer values addresses monetary aspects such as price, resale price, discount, investment, and more. It refers to product value that is stated in monetary units. It becomes essential because the consumer also attempts to minimize costs and other sacrifices that are probably involved in purchasing, ownership, and using certain products [4]. In this study, EVCC is more directed to frontline staff’s capability to actively create co-creation value. Thus, they can produce the most profitable economic cost agreement. The increased consumer confidence resulting from interaction with frontline staff will strengthen their desire to continuously use halal products. Based on those explanations, the proposed hypothesis is: H4: The higher the economic value co-creation, the stronger the sustainable Relationship. Religious Value Co-Creation Capability (RVCC). A value is phenomenologically determined by the customer [21]; thus, personality traits play an essential role in the appraisal process. Therefore, an opened personality will support determining the result and the process of sharing mutual experience, particularly in the new object [22]. Religious value co-creation is a value creation related to religious values. The value obtained from religion is related to its religious commitment [8]. The religious commitment shows how far someone believes in religious values and applies them in everyday life. In this study, RVCC is associated with frontline staff capability of actively participating in establishing values in halal and haram laws of the product, resulting in
The Role of Holistic Value Co-creation Capability
273
strengthening and broadening knowledge of halal products. Strengthening consumers’ beliefs that consuming haram products is a big sin and will not bring any blessing will further increase their beliefs that halal products should become their first choice [23]. Based on those explanations, the proposed hypothesis is: H5: The higher the religious value co-creation, the stronger the sustainable Relationship. The empirical model that shows the relationship between variables can be found in Fig. 1.
Fig. 1. Empirical model
3 Research Methods The population used in this study was sharia insurance customers. The sampling technique used was purposive sampling. Non-random sampling was used in the sampling technique because the number of the population was unknown, so it was not possible to make a sampling frame as the basis for using random sampling. By using a measuring instrument SEM-PLS. The samples collected were 177 of the 200 target respondents. The number of samples was determined based on the opinion [24], which stated that the sample size was sufficient for social research between 100 to 200 respondents. The sample of Islamic insurance customers in this study was taken based on the following criteria: (1) had joined Sharia insurance for at least one year, (2) was able to make decisions independently, (3) knew Sharia insurance. Questionnaires were delivered by officers who had been trained beforehand to respondents who were selected as sample members. The description of the respondents consisted of 68% men and 32% women aged between 32 years to 65 years. The last education of respondents consisted of high school graduates (28%), Bachelors (72%).
274
K. Sudarti et al.
3.1 Analysis Techniques Regression analysis was used to test the empirical research model. The goodness of fit model was determined to determine whether the variation in the independent variable could explain the variation in the dependent variable. A model has well goodness of fit model if the F test produces a p-value that does not exceed 0.05. The coefficient of determination was then determined to determine the percentage of variation in the independent variable that explained the variation in the dependent variable. In addition, the variance inflatoir factor (VIF) test in multiple regression analysis was also used to explain that there was no multicollinearity in the specified regression model. The test results showed that the VIF exceeded 10, so it was considered as there was no multicollinearity in the regression model. Regression analysis was performed using SPSS software version 16.00.
4 Finding 4.1 Reliability Dan Validity Investigation of the internal consistency of latent variables using Cronbach’s alpha required that all variables in the model exceed the recommended minimum of 0.6 [25]. To test the indicator validity was done by calculating the p-value in the t-test to the correlation coefficient of the indicator item score with the total score. The p-value produced a value less than 0.05 indicating high validity. The results of data analysis showed that the Cronbach alpha of all constructs ranged from 0.650 to 0.948 indicating good reliability. Operational definitions and indicators in this study are described in Table 1. below: 4.2 Hypothesis Testing Results Regression analysis showed the excellent goodness of fit model because the Anova test produced an F value = 18,441 and a p-value of 0.000. The normality assumption test with Kolmogorov-Smirnov Z = 1.061, p-value = 0.11 indicated that the data was not normal, but this did not mean anything with a sample of 177 or more than 100. A VIF value less than 10 indicated that there was no multicollinearity. The coefficient of determination for this model was shown by R2 0.34. It means that 34% of the data variation in Sustainable Relationships could be explained by data variations in the Functional Value co-Creation, Social Value co-Creation, Emotional Value co-Creation, Economics Value co-Creation, and Religious Value co-Creation. Meanwhile, the remaining 66% explained the variation of other variables outside the model. Regarding hypothesis testing, the regression model confirmed the regression of Functional Value co-creation and Sustainable Relationship (β = 0.412, p-value = 0.034 or
BT
---->
RI
0.192
0.055
0.0022*
CEx
--->
BT
---->
RI
0.262
0.055
0.0003*
PSC
--->
BT
---->
RI
0.175
0.060
0.0060*
BR
--->
BT
---->
RI
0.080
0.047
0.0538
Note: * p < 0.01
The mediation test procedure proposed by Sobel (1982) was adopted to examine the mediating effect of BT (Table 2). Regarding the Sobel test, the antecedents of CEn, CEx, and PSC affect RI through BT. This is indicated by the p-value of the three antecedents which are Rp. 3.000.000 3 peoples (3%).
4 Result The test results between product knowledge and e-impulse buying show the t-count value is 3.754 > 1.984, which means that the t-count value is greater than the t-table value and sig 0.00. This means that product knowledge has a significant positive effect on e-impulse buying. It can be concluded that the first hypothesis which states that product knowledge has a significant positive effect on e-impulse buying can be accepted. The test results between shopping lifestyle and e-impulse buying, the t-count value is 3.848 > 1.984, which means that the t-count value is greater than the t-table value and sig 0.00. This means that the shopping lifestyle has a significant positive effect on e-impulse buying. Based on these results it can be concluded that. The second hypothesis which states that the shopping lifestyle has a significant positive effect on e-impulse buying is acceptable. In the test results between product knowledge and positive emotion, the t-count value is 4.241 > 1.984, which means that the t-count value is greater than the t-table value and sig 0.00. This means that product knowledge has a significant positive effect on positive emotion. Based on these results, it can be concluded that the third hypothesis which states that product knowledge has a significant positive effect on positive emotion is acceptable. From the test results between shopping lifestyle and positive emotion, the t-count value is 4.483 > 1.984, which means that the t-count value is greater than the t-table value and sig 0.00, this means that shopping lifestyle has a significant positive effect on positive emotion. Based on these results, it can be concluded that the fourth hypothesis which states that the shopping lifestyle has a significant positive effect on positive emotion can be accepted. From the test results between positive emotion and e-impulse buying, the t-count value is 3.881 > 1.984, which means that the t-count value is greater than the t-table value and sig 0.00. This means that positive emotion has a significant positive effect on e-impulse buying. Based on these results, it can be concluded that the fifth hypothesis which states positive emotion has a significant positive effect on e-impulse buying.
320
L. Nurcholis and N. Sa’adah
The results of the path analysis show the direct effect of product knowledge on eimpulse buying is 0.313, while the indirect effect through positive emotion is 0.128. The results of these calculations indicate an indirect effect > sig. 0.05. This means that if positive emotion can become a mediating variable between the effect of product knowledge on e-impulse buying, it means that the higher the product knowledge a Shopee user in Semarang has, the more positive emotion they will have so that it has a positive impact on e-impulse buying. The direct effect of shopping lifestyle on e-impulse buying is 0.324, while the indirect effect through positive emotion is 0.135. The results of these calculations indicate that the indirect effect > sig. 0.05. This means that positive emotion can be a mediating variable between the influence of shopping lifestyle on eimpulse buying, meaning that the higher the shopping lifestyle owned by Shopee users in Semarang, the better positive emotion will have a positive impact on e-impulse buying (Tables 1 and 2). Table 1. Regression test results of model 1 Model
Unstandardized coefficients
Standardized coefficients
T
Sig.
B
Std. error
Beta
1 (Constant)
3.651
1.022
3.573
.001
Product knowledge
.456
.108
.405
4.241
.000
Shopping lifestyle
.402
.090
.428
4.483
.000
t
Sig.
Table 2. Regression test results of model 2 Model
Unstandardized coefficients
Standardized coefficients
B
Beta
Std. Error
1 (Constant)
3,776
,740
5,102
,000
Product knowledge
,299
,080
,313
3,754
,000
Shopping Lifestyle
,258
,067
,324
3,848
,000
Positive emotion
,268
,069
,317
3,881
,000
Source: Processed primary data, 2021
E-Impulse Buying Improvement with Product Knowledge
321
5 Discussion Product Knowledge has a Significant Positive Effect on E-impulse Buying. It can be seen that the Shopee user already has product knowledge as a satisfactory value, which means that a consumer is satisfied when they know information on an item presented by the seller in the Shopee application. When the Shopee users have good product knowledge, then the Shopee users will achieve an e-impulse buying, which is a sudden purchase through an online purchase made by someone. In this study, it is known that the highest average on the e-impulse buying indicator is “ignorance” which means that Shopee users do not care about the consequences of E-impulse buying. When consumers already know, they tend to buy an item without thinking about its usefulness of a good product, they tend to buy an item without thinking about the usefulness or benefits that will be provided from an item. The results of this study are in line with research conducted by [8] research by [10] which states that product knowledge has a positive and significant effect on impulse buying. A Shopping Lifestyle has a Significant Positive Effect on E-impulse Buying. The level of shopping lifestyle of Shopee users tends to be high, which is characterized by spontaneous purchases through the application. Shopee users believe that the products offered through the application are marked by the highest average indicator, namely “Brand and Quality” which means they have trusted a brand and the best quality will be given when they purchase the Shopee application. Shopee application users have a good shopping style, namely being able to use money wisely by purchasing goods through the Shopee application. This can affect a person’s level of e-impulse buying. When the level of consumer shopping lifestyle increases, the level of consumer e-impulse buying will increase. The results of this study are in line with previous research conducted by [12] which states that the shopping lifestyle has a positive and significant effect on impulse buying. Product Knowledge has a Significant Positive Effect on Positive Emotion. Based on the answers to the questionnaires filled out by the respondents, consumers who use the Shopee application have good knowledge about the products they will buy. In addition, consumers are satisfied with the information presented by the seller in the Shopee application. This will have an impact on the level of positive emotion of a consumer, namely a person’s mood and response to something that happens and comes from an environment that can cause positive emotions. Consumers using the Shopee application feel satisfied when making purchases through the Shopee application, this is manifested in the results of this study which show the highest indicator of positive emotion, namely feeling satisfied when shopping. Good product knowledge will have an impact on positive emotions when making purchases through the Shopee application. This research is in line with research conducted by [10] and also research from [11] which states that product knowledge has a positive effect on positive emotion. A Shopping Lifestyle has a Significant Positive Effect on Positive Emotions. Consumers using the Shopee application can have a good shopping style, they are confident in what they buy through the Shopee application. Consumers choose the best brand and quality
322
L. Nurcholis and N. Sa’adah
for the goods they buy, when the goods have been received by consumers, the level of consumer positive emotion will be good due to their trust in Shopee which can provide the best quality. The shopping style of Shopee users can follow changes that are expected to affect positive emotions when they shop. This can have a good effect on a purchase decision. This research is in line with previous research conducted by [12] which found that the shopping lifestyle has a significant positive effect on positive emotions. Positive Emotion has a Significant Positive Effect on E-impulse Buying. From the results of the answers to questionnaires that have been filled out by respondents, it is found that the Shopee user community has high positive emotion toward e-impulse buying through the Shopee application as evidenced by the high feeling of satisfaction when shopping to always produce. With positive emotions when making a purchase, consumers can spend their money through the Shopee application. Consumers tend to have spontaneity when purchasing the Shopee application, they spend their money without thinking more carefully. However, consumers will feel satisfied when buying something through the Shopee application, this has a positive impact on Shopee, if consumers have good positive emotions, it will affect the level of e-impulse buying, namely buying something that happens either planned or not. This study is in line with the results of previous research conducted by [13], which found that positive emotion had a positive and significant effect on impulse buying. The Positive Emotion can Mediate the Effect of Product Knowledge on E-impulse buying. Product knowledge is closely related to knowledge of a product, both raw materials and use values that are beneficial to consumers. If consumer product knowledge is good and correct, then consumers can have a good level of e-impulse buying. Thus, this condition indicates that if consumers have product knowledge by having good positive emotions, the effect on e-impulse buying will be greater. The Positive Emotion can Mediate the Effect of a Shopping Lifestyle on e-Impulse Buying. This shows that a good consumer shopping lifestyle will tend to spend their money wisely to increase the level of positive emotion which has an impact on increasing e-impulse buying. When consumers have positive emotions, they tend to have a positive shopping style so that consumers can spend their money in a planned or spontaneous way. So, this condition means that if a consumer has a strong shopping lifestyle and high positive emotion, then the effect on e-impulse buying will be greater.
6 Conclusion Based on the research and discussion that has been done regarding the analysis of product knowledge and shopping lifestyle on e-impulse buying with positive emotion as a mediating variable in the Shopee user community in Semarang City, the following conclusions can be drawn: if the Shopee user community in the city of Semarang has product knowledge and a strong shopping lifestyle by having good positive emotion, the effect on e-impulse buying will be greater.
E-Impulse Buying Improvement with Product Knowledge
323
References 1. Turban, E., Lee, J.K., King, D., McKay, J., Marshall, P.: Electronic Commerce: a Managerial Perspective 2008. Commerce A Manage Perspectve 5th (ed.), pp. 1–53 (2008) 2. Novak, T.P., Hoffman, D.L., Yung, Y.F.: Measuring the customer experience in online environments: a structural modeling approach. Mark. Sci. 19, 22–42 (2000) 3. DeLone, W.H., McLean, E.R.: Measuring e-commerce success: applying the DeLone and McLean information systems success model. Int. J. Electron. Commer. 9, 31–47 (2004) 4. Liu, S. An empirical study on e-commerce’s effects on economic growth. In: 2013 Conference on Education Technology and Management Science (ICETMS 2013), pp. 957–960 (2012). https://doi.org/10.1109/NCES.2012.6543371 5. Ramzy, O., Eldahan, O.: An empirical investigation of E-commerce in Egypt: the impact of culture on online purchasing. Glob. Bus. Rev. 17(5), 1011–1025 (2016) 6. Bauboniene, Z., Guleviciute, G.: E-commerce factors influencing consumers’ online shopping decision. Soc. Technol. 5, 74–81 (2015) 7. Verhagen, T., Van Dolen, W.: The influence of online store beliefs on consumer online impulse buying: a model and empirical application. Inf. Manag. 48, 320–327 (2011) 8. Liang, Y.-P.: The relationship between consumer product involvement, product knowledge, and impulsive buying behavior. Procedia - Soc. Behav. Sci. 57, 325–330 (2012) 9. Tirmizi, M.A., Rehman, K.U., Saif, M.I.: An empirical study of consumer impulse buying behavior in local markets. Eur. J. Sci. Res. 28, 522–532 (2009) 10. Armilawati, N.L., Kusuma, I.G., Agung, A.P.: The effect of online shopping experience on risk perception and trust for building online repurchase intention. Int. J. Contemp. Res. Rev. 11(05), 21787–21797 (2020) 11. Hodgins, M., Dadich, A.: Positive emotion in knowledge creation. J. Heal. Organ. Manag. 31, 162–174 (2017) 12. Nurlinda, R.A., Christina, D.: Peran posituve emotion sebagai mediator hedonic shopping dan shopping lifestyle terhadap impulse buying di lazada. J. Ris. Manaj. dan Bisnis Fak. Ekon. UNIAT 5(1), 231–244 (2020) 13. Yulia Hermanto, E.: Pengaruh fashion involvement terhadap impulse buying behaviour masyarakat surabaya dengan hedonic shopping motivation dan positive emotion sebagai variabel intervening pada merek zara. J. Manaj. Pemasar. 10, 11–19 (2016) 14. Kusuma, I.D., Untarini, N.: Pengaruh pengetahuan produk terhadap niat beli dengan sikap sebagai variabel intervening. J. Ilmu Manaj. 2, 1573–1583 (2014) 15. Peter, J.P., Olson, J.C.: Consumer Behavior and Marketing Strategy. McGraw-Hill (McGrawHill/Irwin, London (1999) 16. Armstrong, G., Gary, M., Kotler, P., Opresnik, M.O.: Marc O. Marketing: an Introduction (2019) 17. Tambuwun, M.: Shopping lifestyle as intervening relation between hedonic motive and gender on impulse buying. Int. J. Bus. Finance Manag. Res. 4, 9–16 (2016) 18. Chusniasari, C., Prijati, P.: Pengaruh shopping lifestyle, fashion involvement dan hedonic shopping terhadap impulse buying pelanggan. J. Ilmu dan Ris. Manaj. 4, 1–21 (2015) 19. de Kervenoael, R., Aykac, D.S.O., Palmer, M.: Online social capital: understanding e-impulse buying in practice. J. Retail. Consum. Serv. 16, 320–328 (2009) 20. Dholakia, U.M.: Temptation and resistance: an integrated model of consumption impulse formation and enactment. Psychol. Mark. 17, 955–982 (2000) 21. Shen, K.N., Khalifa, M.: System design effects on online impulse buying. Internet Res. 22, 396–425 (2012) 22. Park, E.J., Kim, E.Y., Forney, J.C.: A structural model of fashion-oriented impulse buying behavior. J. Fash. Mark. Manag. 10, 433–446 (2006)
324
L. Nurcholis and N. Sa’adah
23. Semuel, H.: Respons lingkungan berbelanja sebagai stimulus pembelian tidak terencana pada toko serba ada (toserba) (studi kasus carrefour surabaya). J. Manaj. dan Kewirausahaan 7, 152–170 (2005)
Transformational Performance of Police of the Republic of Indonesia Through Smart Working Siti Sumiati(B) and Verri Dwi Prasetyo Department of Management, Faculty of Economic, Universitas Islam Sultan Agung (UNISSULA), Semarang, Indonesia [email protected]
Abstract. The Police of the Republic of Indonesia is one of the institutions that carry out the function of service to the community. The Police of the Republic of Indonesia are required to provide the best service to the community by showing good, professional, and reliable performance in their fields. Based on the research gap and the existing gap phenomenon, this research will develop a model for improving the hard-working and the need for achievement of the Jepara Precinct Police personnel, which is expected to support the personnel’s performance. The population in this study were all members of the Jepara Police, totaling 356 people. The sampling method used the purposive sampling method. The sample in this study were members of the Police who participated in training and particular competencies and the main competencies of the Police totaling 110 people. Measurement using Likert 1 to 5 and data processed using SEM PLS. The results show that Need for Achievement and Need for Power has a significant positive relationship with Hard Working. Meanwhile, Need for Affiliation has a significant negative relationship with Hard Working. Need for Achievement, Need for Power and Need for Affiliation have a significant positive relationship with Smart Working. Hard Working does not have a substantial relationship to HR Performance while Smart Working has a significant positive relationship to HR Performance. Keywords: Need for achievement · Need for power · Need for affiliation · Hard working · Smart working · Human resource performance
1 Introduction The work environment is changing rapidly due to the pandemic. Technology helps people do jobs from anywhere. However, staying connected to work combined with a work environment that requires physical presence for hard workers can lead to negative outcomes. The highly dynamic and sudden changes in the work environment cause people to work more from home and be less able to make work-related decisions. For decades, hard workers have been considered a valuable asset to organizations.
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 325–336, 2022. https://doi.org/10.1007/978-3-031-08812-4_31
326
S. Sumiati and V. D. Prasetyo
Meanwhile, workaholics who are accustomed to working long hours were rewarded. The result was creating a trend of long hours working because of consumerism or society’s need for recognition. Hard workers work long hours every day, so they have less time for other things that affect their quality of life [1]. Motivation is the urge to behave and act in determining individual movement or behavior towards goals [2]. Motivation refers to a good movement within or outside a person that drives an individual’s desire to carry out activities to achieve goals [3]. Previous research is crucial as a basis for the preparation of this research. It helps find out the results conducted by previous researchers that can support subsequent research activities. The relationship between motivation and performance is more substantial when the motivation is correlated to ability than when it comes to the need for achievements. Thus, the study’s findings support further research to examine the impact of the need for achievements on performance [4]. Differences in research results found that achievements do not necessarily affect performance. Need of achievement shows a different effect on individual performance related to their personality [5]. Achievements support individual performance; the higher the desire to achieve something, the higher the performance [6]. In comparison, motivation influences performance, but the significance is minimal, so it cannot be generalized properly [7]. This result is different from the results, which show that motivation influences performance. The type of motivation that individuals have will affect their work style. The human resources of the Jepara Police Precinct are a strategic asset to achieve the vision, mission, and objectives of the Indonesia National Police. The role of human resources in the organization is substantial, so there is a need for professional, modern, and accountable management. It is through the provision, education and training, use, maintenance, and termination of personnel and sound strategic planning to achieve personnel performance. This study aims to describe and analyze the relationship between the need for achievement, the need for power, and the need for affiliation to hardworking and smart working. This study also described and analyzed the relationship between hardworking and smart working on the performance of police personnel at the Jepara Police.
2 Literature Research 2.1 Police Personnel Performance According to [9], performance is the completion of work assigned to an individual by fulfilling several aspects/standards. Performance is the result of the work process achieved by a person or group of people in the organization based on their authority and responsibilities to accomplish organizational goals as measured by the standards set by the organization [10]. The factors that affect performance are (a) the quality of work results, which can be measured by timeliness, work accuracy, and work neatness; (b) The quantity of work, can be measured by the number of jobs completed; (c) timeliness, can be measured by the achievement of the time achieved in completing the work [11].
Transformational Performance of Police of the Republic of Indonesia
327
2.2 The Effect of Motivation on Smart Working and Hard Working The theory of needs by McClelland proposed the relationship between the need for achievement, affiliation, and power in the late 1940s. It includes the need for achievement, the need for affiliation, and the need for power [12]. Need for achievement (n-Ach) is one type of motivation that grows because of a need motive that distinguishes it from other needs [5]. Personnel with a high need for achievement have control over their behavior and like difficult challenges. On the contrary, employees who have a low need for achievement are easily satisfied with few challenges. The need for achievement has certain characteristics, as developed by McClelland. They are innovative, require feedback, responsibility, persistence, and like challenges [13]. Motivation will influence employees to work hard and ensure their organizational goals. Motivated employees will prefer to work hard and persevere to achieve the set targets [5]. Motivation influences inspiration, the need for autonomy and freedom, tolerance for ambiguity, hard work, persistence, and optimism [14]. In conclusion, individuals with a high need for achievement will desire to work hard. Previous research indicates that motivation towards achievement plays an important role in determining performance [3]. The results showed a significant and positive influence of the need for achievement on performance [6]. It is indicated by encouraging the need for higher performance which will increase performance. Need of power (Npow) is a person’s tendency to influence others to be considered a strong/influential person [15]. Individuals with power (nPow) influence affective responses to these motive-specific stimuli and alter the stimulus-driven learning process [16]. Individuals with a high need for power (nPow) want to control others, enjoy their power to advance personal interests, and consider power as a means to help others [17]. Previous research by [18, 19] A person with a Need for Achievement; Need for Affiliation, High Need for Power have control over their behavior and prefer difficult challenges, which will affect their performance [14, 16]. Individuals who require power usually take extreme risks, aim high for goals, seek to build alliances with others, have a desire to control events and the environment, and tend to work harder to make their wishes come true [20]. H1:Need for achievement has a significant effect on hard working H2:Need for achievement has a significant effect on smart working H3:Need for power has a significant effect on hard working 2.3 The Effect of Motivation on Smart Working and Hard Working Workers who engage in smart work increase their productivity compared to those who continue to work traditionally. Smart workers must develop self-determined behaviors and strong intrinsic motivation to work [21]. Individuals with a need for power crave positions that allow them to use their power in influencing others. They influence others to contribute more to the organization by working effectively and efficiently [22]. As a result, motivation is needed to implement smart working.
328
S. Sumiati and V. D. Prasetyo
Individuals with a need for affiliation will try to work in groups by creating friendly relationships and have a strong desire to be liked by others [3]. Employees reflect the desire to have close, cooperative, and friendly relationships with other parties. Employees with high motivation will try to do their best, have confidence in their ability to work independently, and be optimistic. They will also have a great responsibility for every action or deed they do [23, 24]. The need for affiliation encourages individuals to maintain good relations with co-workers and leaders above them to improve performance [14]. A person with this need for affiliation tends to collaborate with others and usually avoid high-risk or uncertain situations [23]. Smart workers must develop self-determined behavior and strong motivation to work [25]. H4:Need for power has a significant effect on smart working H5:Need for affiliation has a significant effect on hard working H6:Need for affiliation has a significant effect on smart working 2.4 The Role of Hard Working on Personnel Performance Improvement Hard-working reflects the behavioral tendency to work hard and produce new and appropriate jobs. Hard-working improves organizational effectiveness and efficiency. It is summed up as enthusiasm and passion for working seriously to achieve excellent and maximum results. The indicators in this study are happy to be creative, seek involvement, passionate about work, overwork, and work compulsively. The tendency to work smart can encourage them to perform well. A good performance is shown by completing work on time and completing tasks according to the specified quantity. In conclusion, employees who can work smart can impact performance improvement. H7: Hardworking has a significant effect on HR performance 2.5 The Role of Smart Working on Personnel Performance Improvement The adoption of smart working requires intervention across organizational structures, workplace layouts, work practices, and levels of human behavior [26]. The indicators used are software collaboration, changing HR behavior and practices, and reconfiguring the work environment [27]. The implementation of smart working initiated by the company will enable workers to work more effectively [28]. Smart working facilitates the network’s building where people feel more unrestrained from hierarchical boundaries, communicate better, and work collaboratively with greater autonomy [25]. The context of smart working is not only centered on technology but, furthermore, face-to-face and virtual interactions, both in physical and digital workplaces. The interactions allow work to be done more efficiently, effectively, and timelessly [26]. H8: Smart working has a significant effect on HR performance
3 Research Method The type of research used is explanatory research. This research determined the population consists of all members of the Jepara Police, totaling 356 people. The sampling
Transformational Performance of Police of the Republic of Indonesia
329
method used the purposive sampling method for 110 members of the Police who took part in training and special vocational competencies and main competencies of the Police. The data were obtained through questionnaires using interval measurements, and the score was in the form of a Likert scale of 1 to 5, from strongly agree to a scale of 1 for strongly disagree. The data were then analyzed using structural equation modeling through the Partial Least Square (PLS) approach, which the result can answer the hypothesis.
4 Research Findings and Discsussion Descriptive analysis data showed that male respondents are dominated by 97 respondents (88.18%) compared to female respondents. It gives its advantages to members of the Jepara Precinct, where the police were also the right person to handle cases involving physical and heavy issues. Respondents with a high school education background ranked first with 54 respondents or 49.09%, followed by a bachelor’s education background (S1) with 49 respondents. Thus, this condition impacts the organization because, with a high school education background that dominates the members of the Jepara Precinct, the retirement period for the police will be longer. The entry age for high school graduates is certainly younger so that the police population is maintained and avoids zero growth (the number of retirees and the registrant is the same/same logged-in member). The tenure of the personnel is dominated by 11 to 15 years of service, which shows that members of the Jepara Precinct have deeply studied equality, internalized the code of ethics, and animated the Bhayangkara philosophy. Jepara Police personnel who belong to BRIPKA are the most respondents, namely 43 respondents or 37.8%. It benefits the BRIPKA of Jepara Police as the personnel who have served for a long time, more than 15 years, have sufficient work experience and can continue their education to the next level. A validity test is used to measure the validity of a questionnaire. The result shows that AVE value is above 0.5 for all constructs contained in the research model. Therefore, all indicators in this study are declared valid. Cronbach Alpha results are above 0.6. Thus, it is reliable if the Cronbach Alpha value is > 0.60. The results of Composite Reliability between constructs and their indicators show promising results, namely above 0.70 > 0.5. In other words, it has a good reliability value and can be used for further research processes. Reliable means the indicators used in actual research are under the real conditions of the research object. After the estimated model meets the Outer Model criteria, the next step is testing the structural model (Inner model). The Adjusted R-Square values for the Hard-Working construct is 0.054. It means that Need for Achievement, Need for Power, and Need for Affiliation can explain the Hard-Working variance of 5.4%. The remaining 94.6% are explained by other variations not included in the model. The R-value is also found in Smart Working for 69.8%, influenced by Need for Achievement, Need for Power, and Need for Affiliation. The remaining 30.2% is influenced by other variables not included in the model. The R-value is also encountered in HR performance influenced by Hard Working and Smart Working, 61.9%. The remaining 38.1% is influenced by other variables not included in the model. Following is the hypothesis testing:
330
S. Sumiati and V. D. Prasetyo Table 1. Hypothesis test Sampel Asli (O)
T Statistik (|O/STDEV|)
P Values
Ket
H1
Need of achievment -> Hard working
−0,174
1,166
0,244
Ditolak
H2
Need of achievment -> Smart working
0,511
6,037
0,000
Diterima
H3
Need of affiliation -> Hard working
−0,222
2,198
0,028
Diterima
H4
Need of affiliation -> Smart working
0,149
2,011
0,045
Diterima
H5
Need of power -> Hard Working
0,433
3,166
0,002
Diterima
H6
Need of power -> Smart 0,287 working
3,105
0,002
Diterima
H7
Smart working -> Kinerja SDM
14,794
0,000
Diterima
H8
Hard working -> Kinerja 0,048 SDM
0,786
0,432
Ditolak
0,790
Source: Processed primary data, 2021
To determine whether a hypothesis is accepted or not, it can be found out by comparing t count with t table. The condition is if the t count > t table, then the hypothesis is accepted. The test used a two-tailed test with a probability (α) of 0.05, and the degree of freedom of the test is Df = (nk) = (110–6) = 104. It results in t table value for df 104 t table two-tailed test (two-tailed) found a coefficient of 1.98. So that the equation formed based on the table above is: Equation 1: Y1 = –0,174 X1 + 0,433 X2 –0,222 X3 Equation 2: Y2 = 0,511 X1 + 0,287 X2 + 0,149 X3 Equation 3: Y3 = 0,048 Y1 + 0,790 Y2 Table 1 shows that: First, The need for achievement does not have a significant effect on hard-working.. Stressors factors result in the incapability of the need for achievement in influencing performance at work. The achievement and performance measurement within the National Police unit is complex. Besides administrative demands, field performance must also be fulfilled, which is in contact with cognitive and behavioral aspects in completing tasks. Work behavior begins with motivation accompanied by positive work attitudes, perceptions, values, and abilities or competencies possessed by members of the National Police. However, due to various circumstances, demands, the influence of the environment, and personality conditions that tend to be weak, it will result in modest daily life in the work environment and outside the task. In such conditions, despite having the ability, high work motivation will not appear in superior performance but will lead to poor performance and can endanger himself and others.
Transformational Performance of Police of the Republic of Indonesia
331
Second, Need for achievement has a significant positive effect on Smart Working. This study demonstrates that the higher the desire for achievement, the higher the ability to work intelligently. Therefore, the smart work behavior of the Jepara Police is formed from the desire to accomplish a goal. Third, The need for power has a significant positive effect on hard work. The higher the desire for power, the higher the ability to work hard. In other words, the hard-working behavior of the Jepara Police is assembled from the desire to maintain power. Forth, The need for affiliation has a significant negative effect on hard work. The higher the desire for affiliation, the lower the ability to work hard. In sum, the desire for affiliation will lower the hard-working behavior of Jepara Police Members. Sixth, The need for affiliation has a significant positive effect on Smart Working. The higher the desire for affiliation, the higher the ability to work smartly. It can be concluded that the smart work behavior of Jepara Police members is formed from the desire to be affiliated. Seventh, The need for affiliation has a significant positive effect on smart working. The higher the desire for affiliation, the higher the ability to work smartly. In conclusion, the smart work behavior of Jepara Police members is constructed from the desire to be affiliated. Eighth, Hard-working does not have a significant positive effect on HR performance. The more elevated the hard-working behavior of the Jepara Police members will not affect their performance. Specifically, the performance of Jepara Police personnel is not formed from hard-work behavior. It is because the workload of the Jepara Police is related to the safety and comfort of working in pandemic conditions. The police’s work is a field that cannot be done online or with remote workers. Working in a tense and life-threatening pandemic atmosphere will play a role in determining the degree of stress that occurs. The performance output during the pandemic will, of course, be different from the performance output during normal conditions. Nonetheless, the indicators for calculating and assessing the IKU IKK of the Polres work remain to affect performance achievements. Smart working has a significant positive effect on HR performance. The higher the smart work behavior of the Jepara Police, the higher the performance. Notably, the performance of the Jepara Police personnel is constructed from the ability of the personnel to work smartly. The obtained original sample estimate value shows that the highest value affecting hard work is the need for power, 0.433. Meanwhile, the highest value affecting Smart Working is the need for achievement, 0.511. The highest value affecting HR performance is Smart Working, equal to 0.790.. The following is a diagram of the statistical T value based on the output with SmartPLS Version 3: This indirect effect analysis is intended to determine the effect of the hypothesized variables. The indirect effect is the effect caused by the intermediate variable. Each variable is presented in Table 2.
332
S. Sumiati and V. D. Prasetyo Table 2. The indirect effect Sampel Asli (O)
Rata-rata Sampel (M)
Standar Deviasi (STDEV)
T Statistik (|O/STDEV|)
P values
Ket
−0.008
−0.006
0.015
0.578
0.563
Ditolak
Need of −0.011 affiliation -> Hard Working -> Kinerja SDM
−0.009
0.015
0.715
0.475
Ditolak
Need of power -> Hard Working -> Kinerja SDM
0.021
0.021
0.029
0.712
0.477
Ditolak
Need of Achievment -> Smart Working -> Kinerja SDM
0.403
0.392
0.070
5.725
0.000
Diterima
Need of 0.118 affiliation -> Smart Working -> Kinerja SDM
0.115
0.062
1.910
0.057
Ditolak
Need of power -> Smart Working -> Kinerja SDM
0.237
0.072
3.148
0.002
Diterima
Need of achievment -> Hard Working -> Kinerja SDM
0.227
Source: Processed primary data, 2021
The indirect effect in this research model only appears in the influence of the need for achievement on HR performance through smart working and the need for power on HR performance through smart working. Thus, in this study, smart working is an intervening
Transformational Performance of Police of the Republic of Indonesia
333
Fig. 1. Output bootstrapping
variable in the need for achievements and power on HR performance. Smart working can mediate the need for achievement and need for power on HR performance but cannot moderate the need for affiliation on HR performance. The hard-working variable cannot mediate the need for achievement, power, and affiliation on HR performance. Therefore, in this study, hard-working is not an intervening variable. Based on the indirect influence path analysis, the path that influences HR performance most is the path of need for achievement on HR performance through smart working and the path of need for power on HR performance through smart working.
5 Conclusion The research conducted on 110 Jepara Police personnel showed that hard work is built by the need for power. In contrast, the need for affiliation has a significant negative relationship with hard work. Smart working is made by the need for achievement, need for power, and need for affiliation. Meanwhile, HR performance is built by smart working. Improving the hard-working behavior of Jepara Police personnel can be realized by increasing the desire to have power and paying attention to the pattern of affiliation that occurs within the organization to minimize the adverse impact on HR performance. Improved smart work behavior can be done through MC Clelland’s motivation, namely the desire to achieve, the desire for power, and the desire to be affiliation. The performance of Jepara Police Human Resources can be improved through smart work behavior that can utilize existing resources to do work effectively and efficiently.
6 Managerial Implication Organizations must improve the ability of personnel to innovate. It is accomplished by providing rewards for personnel who innovate or renew their work. The Jepara Police
334
S. Sumiati and V. D. Prasetyo
are expected to enhance personnel leadership skills to be more decisive in influencing others. Jepara Police is expected to improve partnerships to establish emotional involvement between personnel and work more in teams. It can be done by having joint activities regularly so that all components and elements in the organization can mingle and socialize. Jepara Police is expected to improve the ability of its members to work excessively. Dealing with a high workload will affect emotional resilience and self-resistance to stress. It can be done by providing religious assistance and guidance and holding relaxing and refreshing activities to maintain a balance in the work-life of members. Improved reconfiguration of the work environment aims to make personnel feel comfortable carrying out community activities and services. It can be realized by rearranging the work environment, adding work infrastructure, revitalizing or modernizing work tools and facilities. The Jepara Police are expected to emphasize neatness in every activity carried out by personnel. It is by disciplining the placement of work tools, using the ergonomic and applicable infrastructure.
7 Theoretical Implication This research contributes to improving HR performance through smart working. The supporting study states that the smart work context allows work to be done more efficiently, effectively, and timelessly [26]. This research contributes to the theory of motivation, including the three dimensions of McClelland’s motivation. They are Need for Achievement, Need for Power, and Need for Affiliation, which can improve the ability of human resources to work smartly. The development of research on the role of Mc Leland’s theory on hard work has differences in this study’s results with research by [5, 14] It states that motivated employees will desire to work hard and diligently to achieve the set targets. Then the results of this study contradict the study results by [14]. The Need for affiliation can improve performance. This research contributes to the development of research related to the role of hard-working on HR performance.
8 Research Limitation and Future Research Agenda The limitations of quantitative research are the lack of secondary data availability, possible bias due to homogeneous respondents, and static organizations because they have solid regulatory attachments. The coefficient of need for achievements is small so that it cannot affect hardworking; It is an attractive black box to describe. The coefficient of need for affiliation is negative, causing a lowering effect on hard-working. The study was conducted in the same organization, so the possibility of bias is high and requires further research to be generalized more broadly. Future research is expected to dig deeper into the need for affiliation on hard work. The research on McClelland’s motivational dimensions can be carried out again in a more dynamic and heterogeneous organization, and it generalizes broader. This study relates the role of Need for Achievement, Need for Power, and Need for Affiliation to hard-working and smart working but has not analyzed the linkage of these three dimensions directly to the performance of police personnel.
Transformational Performance of Police of the Republic of Indonesia
335
References 1. Fragouli, E., Ilia, I.: Working smart and not hard Key to maximize employee efficiency?. Int. J. Inf. Bus. Manag. 11(2), 74–111 (2019) 2. Nhat, N.C., Van, D.N., Huu, T.H.: Effects of motivation on employees’ performance at petrovietnam nghe an construction joinst stock corporation. Am. J. Bus. Manag. 2(2), 212–217 (2013). https://doi.org/10.11634/216796061302307 3. Rybnicek, R., Bergner, S., Gutschelhofer, A.: How individual needs influence motivation effects: a neuroscientific study on McClelland’s need theory. Rev. Manage. Sci. 13(2), 443– 482 (2017). https://doi.org/10.1007/s11846-017-0252-1 4. Van Iddekinge, C.H., Aguinis, H., Mackey, J.D., DeOrtentiis, P.S.: A meta-analysis of the interactive, additive, and relative effects of cognitive ability and motivation on performance. J. Manage. 44(1), 249–279 (2018). https://doi.org/10.1177/0149206317702220 5. Jaquays, B.T.D.: Motivation McClelland’s Needs Theory, vol. 8, no. 9, pp. 1–7 (2018) 6. Groening, C., Binnewies, C.: Achievement unlocked! - the impact of digital achievements as a gamification element on motivation and performance. Comput. Human Behav. 97, 151–166, November 2018 (2019). https://doi.org/10.1016/j.chb.2019.02.026 7. Sareen, P., Joshi, P.: Organizational learning and motivation: assessing the impact on employee performance. IRA-Int. J. Manag. Soc. Sci. (ISSN 2455–2267), 5(2), 355 (2016). https://doi. org/10.21013/jmss.v5.n2.p13 8. Bao, C., Nizam, D.I.: The impact of motivation on employee performance in the electronics industry in China. Int. J. Account. Bus. Manag. 3(2), 29–45 (2015). https://doi.org/10.24924/ ijabm/2015.11/v3.iss2/29.45 9. Shields, J., et al.: Managing Employee Performance & Reward: Concepts, Practices, Strategies. Cambridge University Press (2015) 10. Choi, I.: Moving beyond mandates: organizational learning culture, empowerment, and performance. Int. J. Public Adm. 43(8), 724–735 (2020). https://doi.org/10.1080/01900692.2019. 1645690 11. Robbins, S.P., Judge, T.A.: Organization Behaviour. Pearson/Prentice Hall (2007) 12. McClelland, D.C.: Studies in Motivation (1955) 13. McClelland, D.C.: The achievement motive in economic growth. In: The Gap Between Rich And Poor, pp. 53–69. Routledge (2019) 14. Nguyen, P.T., Yandi, A., Mahaputra, M.R.: Factors that influence employee performance: motivation, leadership, environment, culture organization, work achievement, competence and compensation (a study of human resource management literature studies). Dinasti Int. J. Digit. Bus. Manag. 1(2), 1–16 (2020). https://doi.org/10.31933/DIJDBM 15. Braden, P.A.: McClelland ’ s Theory of Needs, pp. 1–2 (2000) 16. Janson, K.T., Bleck, K., Fenkl, J., Riegl, L.T., Jägel, F., Köllner, M.G.: Inhibited power motivation is associated with the facial width-to-height ratio in females. Adapt. Hum. Behav. Physiol. 148, 148–162 (2019) 17. Abbah, M.T.: Employee motivation: the key to effective organizational management in Nigeria. IOSR J. Bus. Manag. 16(4), 01–08 (2014). https://doi.org/10.9790/487x-16410108 18. Pratiwi, R., Widodo.: Coercive intellectual leadership antecedent towards organizational performance. Qual. - Access Success 22(182), 35–40 (2021) 19. Barton, M., Schaefer, R., Canavati, S.: To be or not to be a social entrepreneur: motivational drivers amongst american business students. Entrep. Bus. Econ. Rev. 6(1), 9–35 (2018). https://doi.org/10.15678/EBER.2018.060101 20. Erciyes, E.: A new theoretical framework for multicultural workforce motivation in the context of international organizations. SAGE Open 9(3), 1–12 (2019). https://doi.org/10.1177/215824 4019864199
336
S. Sumiati and V. D. Prasetyo
21. Bucea-Manea-toni¸s, R., et al.: The relationship between eco-innovation and smart working as support for sustainable management. Sustain 13(3), 1–17 (2021). https://doi.org/10.3390/ su13031437 22. Kocur, D., Mandal, E.: The need for power, need for influence, sense of power, and directiveness in female and male superiors and subordinates. Curr. Issues Pers. Psychol. 6(1), 47–56 (2018). https://doi.org/10.5114/cipp.2018.72200 23. Steinmann, B.: The role of the need for affiliation and the behavioral manifestation of implicit motives in effective leadership: a dimensional approach (2017) 24. Steinmann, B., Otting, S.K., Maier, G.W.: Need for affiliation as a motivational add-on for leadership behaviors and managerial success. Front. Psychol. 7, 1–18 (2016). https://doi.org/ 10.3389/fpsyg.2016.01972 25. McEwan, A.: Smart Working: Creating the Next Wave. CRC Press (2016) 26. Iannotta, M., Meret, C., Marchetti, G.: Defining leadership in smart working contexts: a concept synthesis. Front. Psychol. 11, 1–11 (2020). https://doi.org/10.3389/fpsyg.2020. 556933 27. Decastri, M., Gagliarducci, F., Previtali, P., Scarozza, D.: Understanding the Use of Smart Working in Public Administration: The Experience of the Presidency of the Council of Ministers (2015) 28. Neri, M., et al.: Smart working: una prospettiva critica, no. September 2016 (2017)
The Role of Tawhidic Paradigm in Knowledge Creation Process Nurhidayati(B) and Andhy Tri Adriyanto Department of Management, Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia [email protected], [email protected]
Abstract. This study aims to determine the role of the Tawhidic paradigm in knowledge creation for organizations. The tawhidic paradigm refers to Islamic rules that become the basis for the creation of new knowledge in individuals and provides illustrations that help link the spirit of Tawhid with the conventional SECI model of knowledge creation which is a process that emphasizes the tacit or explicit knowledge creation process and utilizes these processes to build knowledge networks within the organization. This model is also based on resource-based view that asserts that humans are valuable resources based on their own intellectual, emotional, and spiritual assets in terms of knowledge creation. The model should ensure that knowledge process creation is integrated and balanced with the principles of the world (ad-dunya) and hereafter (al-akhirah). Tawhidic paradigm becomes a catalyst and guide that knowledge creation will be achieved in the corridor of goodness (‘amr bil ma’aruf ) and prevention of evil (nahi anil munkar). The proposed model of tawhidic paradigm in the creation of knowledge will make individuals as members of organizations continue to learn within the framework of worship. This proposed model will enrich organizational insights into individual knowledge creation and the processes involved. Keywords: Knowledge creation · Islamic worldview · Tawhidic paradigm · SECI · Resource-based view
1 Introduction In the Knowledge Era which indicates speed and high competition, the key to the success of a company in winning the competition in the industry is to ensure that the company has an asset to become competitive [1, 2]. A competitive advantage such intangible assets of knowledge ownership is able to obtain by creating from self-experience and collecting from external sources. Competitive advantage is defined as the company’s ability to earn a consistent return on investment above average for the industry and when the company implements a unique value creation strategy that is difficult to be imitated by its competitors [1, 2]. Typical knowledge with competitive advantage properties such as rare, valuable, non-substitutable, and imperfectly imitable, able to become strategic assets that are manageable by an organization to generate long-term profits. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 337–347, 2022. https://doi.org/10.1007/978-3-031-08812-4_32
338
Nurhidayati and A. T. Adriyanto
Knowledge has both a tacit and explicit dimension and, in an organization, the creation of new knowledge results from a consistent dialogue between the tacit and the explicit knowledge [3]. Thus, knowledge creation should pay more attention to the organizational context. Knowledge management is a complex concept and has many aspects to support knowledge creation, knowledge transfer, and application of knowledge in an organization. This knowledge is embedded and carried through several entities, including organizational culture and identity, routines, policies, systems, and human resources. Knowledge management consists of identifying and utilizing collective knowledge within an organization [2]. Individuals who have a good level of knowledge management will be able to turn it into a value-creating activity [4]. The existence of this value certainly needs to go through several stages in knowledge management, one of which is knowledge creation as the initial stage in knowledge management. Knowledge creation is a process of creating knowledge through a certain series accompanied by media and the transfer of knowledge to other individuals or groups. According to Kaba and Ramaiah [5], knowledge creation is the first stage in the knowledge chain. Knowledge creation is seen as the most vital and important element in knowledge management activities [6]. Thus, most organizations try to be competitive at their best by developing knowledge creation that will help them in achieving organizational goals. In knowledge creation, an organization is based on the practices of the SECI model developed by Nonaka and Takeuchi [7]. Their work suggested that the creation of knowledge in organizations can be represented by a four-stage spiral model abbreviated the model as SECI for four modes of knowledge conversion including - Socialization, Externalization, Combination, and Internalization. This SECI model emphasizes the tacit or explicit knowledge creation process and utilizes this process to build knowledge networks within the organization [8]. The SECI model can help to know how to make both types of knowledge usable at all levels of the organization. Huang et al. [9] stated that the application of the SECI model would enrich the insights of organizations into their knowledge creation and the processes involved. The practical implications of the application of the SECI model in an organization developed by Nonaka and Takeuchi [7] still need to be improved, such as creating a knowledge vision, building strong interactions, and building knowledge networks with other organizations. In implementing this practice, of course, there are still some weaknesses, one of which is the lack of involvement of religious values in it. Recently, religious paradigms such as tawhidic paradigms in Islam taught gradually increase and obtain the attention of scholarly to relate with organizational management and performance. However, the study literature lacks studies on these values and the impact of urgent organizational issues such as knowledge management. Religious value is a life value that reflects the development of religious life based on belief (aqidah), worship, and morals. In the practice of knowledge creation, it is necessary to involve religious values as the basis for developing Islamic knowledge for each individual. This value can create Islamic knowledge that can influence behavior and culture in the organization. To involve religious values in knowledge creation, Islam has a tawhidic paradigm in making rules and implementing everything within the limits of aqidah that have been determined in Islamic law. The tawhidic paradigm provides an
The Role of Tawhidic Paradigm in Knowledge Creation Process
339
understanding to perform actions that are in accordance with aqidah, worship, and morals [10, 11]. In everyday life, individuals must get good religious values, live as servants and caliphs of Allah, increase values in life, and focus on the afterlife. The tawhidic paradigm in the application of knowledge creation practices must be able to harmonize social interests in the organization and interact with the environment to deliver studies on Islamic religious rules into the concept of knowledge creation that can be transferred by interacting and taking part in group activities where skills and values will understand each other. Thus, this article proposes the role of the tawhid paradigm in knowledge creation. The discussion of this paper begins with explaining the knowledge creation of the SECI Model. Then, the paper discusses the major components of the Islamic worldview, and tawhidic paradigm, and a critical section of this paper present the integration between tawhidic paradigm and the knowledge creation process. Finally, the paper concludes the research.
2 Literature Review 2.1 Knowledge Creation As an initial phase of knowledge management, knowledge creation has been identified to provide value and competitive advantage to organizations [1, 2]. Knowledge creation can also be identified as the creation of new ideas and the process of developing knowledge to replace old ideas which can be regarded as a social not mechanical, interactions and dynamic, not static, processes between individual member organizations and the environment. Thus, the process is not only the transfer of information and data [3, 12]; the process, including ‘input-process-output” as a cognitive process, needs organizational learning in three domains such as; supportive learning environments, concrete learning process, and practices experimentation, and knowledge-oriented leadership [13]. How knowledge is created, Nonaka and Toyama explain that activity based on the Structuration Theory. Generally, knowledge is created by synthesizing the contradiction between an organization’s internal resources and the environment [12]. In their works, [12] suggested the structuration theory that views humans as role-taking and norm fulfilling beings who act based on their images of what reality is, and treats all institutions and social practices as structures. On the other hand, the interactions between humans as agents and the environment continuously provides action and reaction in defining and reproducing social action. Knowledge is created through such interactions between human agency and social structures that will create and enlarge knowledge through the conversion process of tacit and explicit knowledge. Previous studies concluded that knowledge created in social interaction amongst individuals and organizations in a dynamic learning environment requires human action, interpretation, and understanding [3, 12, 14]. According to [7], knowledge creation refers to the process of overcoming individual limitations caused by existing information and past experiences by reaching new perspectives or observing new environments and new knowledge. The knowledge creation process is a learning process to create new and more complex knowledge which involves individuals and groups in organization. In order to help managers, implement the four phases of the knowledge creation process in an organization that is socialization, externalization, combination, and internalization, [7, 13] proposed five practices
340
Nurhidayati and A. T. Adriyanto
namely, sharing tacit knowledge, creating concepts, and justifying concepts, and building archetypes (Fig. 1).
Fig. 1. Organizational knowledge creation process
2.2 Islamic Worldview Al-Maududi [15] termed it as Islami Nazariyat (Islamic Vision), Sayyid Qutb uses the term al-Tasawwur al-Islami (Islamic Vision), Atif al-Zayn mentions Al Mabda’ AlIslami (Islamic Principle), al-Attas [16] named it Rukyatul Islam Lil Wujud (Islamic Worldview). If these definitions are combined, then Islamic Worldview is defined as “Aqidah Fikriyah” or belief based on reason, the principle of which is the unity of tawhid that believes God is Almighty which is there is no other God except Allah, which is formed in mind and the heart of every Muslim and influences his views on all aspects of life. If the structure of knowledge in the worldview supports intellectual activity in an epistemological sense, then will be born a definition of knowledge that not only emphasizes great concepts but also generates in mind the intellectual framework of general workings of the theory of knowledge. The concept of knowledge then produces the concept of truth because the main goal of science is true knowledge, which in turn leads to the concept method. After all, scientists want to know how true knowledge is attained. Previous studies [16, 17] stated that the Muslim Worldview of Islam, namely: believing that Islam is an absolute, true, and perfect religion, being a living system, and a set of eternal and universal teachings, studying Islam, practicing Islam, preaching Islam, and be patient with Islam. From this sentence, it can be interpreted that we as Muslims must feel that we have, maintain, maintain, build, and develop our living system so that it can be a means to worship Allah. Muslims must increase their commitment to their respective scientific fields by
The Role of Tawhidic Paradigm in Knowledge Creation Process
341
increasing their creations, achievements, and reputations with full diligence, attention, and dedication [17]. According to Islam taught the original ownership of knowledge is God, which was later revealed to the Prophet Muhammad SAW and those written in Holy Book Al-Qur’an. Furthermore, Holy Book Al-Qur’an will be the main guidance book for Muslims to practice anything including creating new knowledge. 2.3 Tawhidic Paradigm The tawhidic paradigm is one of the views of Islam that is used as the basis for making rules and applying everything within the limits of aqidah that have been determined in Islamic law. The direct impact of the tawhidic paradigm is to carry out the mandate to promote good and prevent evil [18]. In short, it can be said that the tawhidic Paradigm views all God’s creatures with an equal view; it implies that God is the supreme being. Putting God in the highest position leads to the consequence that all of God’s creatures are equal. The term ‘Tawhid’ in Arabic is defined as the Oneness of Allah [19]. Paradigm is important in the Islamic faith because it is related to kalima shahada, a statement that demands conformity of one’s thoughts, understandings, feelings, actions, and decisions towards it. The tawhidic paradigm provides an understanding to perform actions that are in line with faith (‘aq¯ıd¯ah), worship, and moral in life as a servant and caliph of Allah [19]. In everyday life, Muslims must get good values, live as servants and caliphs of Allah, increase values in life, and focus on the afterlife. Briefly, the basic principles of the world and hereafter are described in the figure below [19]:
Fig. 2. Tawhidic paradigm
342
Nurhidayati and A. T. Adriyanto
The tawhidic paradigm aligns the relationship with God vertically, where God as the creator always has a position above His creatures and balances the relationship between humans and nature horizontally; and The Tawhid paradigm provides an explanation of how God’s power over the single event in the world can be seen through His verses and this able to be analogized as an ‘open book’. Meanwhile, the Word of God mentioned in Al-Qur’an is a ‘written book’ that guides humans in living their lives, including in creating knowledge. Whatever the process of human activities, including the creation of knowledge, those need to understand the dynamics of nature and the Al-Qur’an, as the basis of tawhidic, both are assimilated and cannot be separated. The tawhidic paradigm encourages people to think through the process of discovery, the creation of knowledge, and all events in nature must be based on goodness and avoid bad values, both in the input, process, and application of knowledge results. According to Islam, knowledge aims to guide the human intellectual in elaborating and understanding the ultimate questions of life.
3 SECI and Tawhidic Paradigm The tawhidic paradigm in this proposed conceptual model recognizes that science is a gift from God given to humans as perfect creatures of God (Al-Qur’an Sura At-Tin: 4) who have unique and valuable assets. The tawhidic paradigm in this model links knowledge creation with the SECI model and the Resource-Based View theory. The SECI model describes the stages and mechanisms of knowledge conversion [7, 12, 13], while the Resource-Based View describes people as unique and valuable organizational assets to win business competition [1, 2]. Humans have intellectual, emotional, and spiritual assets that form the basis for creating knowledge based on divine values. The tawhidic paradigm, which is rooted in the Oneness of God, states that humans have a vertical relationship with God (habluminallah) and a horizontal relationship with humans (habluminannas). In the context of the creation of knowledge, the relationship of habluminallah guides the initiates, motives, and intention to create the knowledge which has to root in faith, belief in God, worship, and Morals. Moreover, in the relationship of habluminallah it is also stated that humans have roles as servants (‘abd) and caliphs (khalifah), and the process of creating knowledge through the SECI model must be based on values related to the purpose of creation and the role of humans in the world. Thus, the reason for knowledge creation should follow the command and the prohibitions that God mandates to human beings and have to generate goodness (Good, Righteous, and Lawful) and avoid (Bad, Sin, and Prohibited) destruction for human beings and the universe. Tawhidic paradigm becomes å religious foundation to lead the input-processoutput of knowledge creation will always be in Command and Blessing from Allah and bring benefit to others. The summary of the role tawhidic paradigm in the knowledge creation process is shown in Table 1.
The Role of Tawhidic Paradigm in Knowledge Creation Process
343
Table 1. The role of tawhidic paradigm in the knowledge creation process Phase
Tawhidic paradigm Vertical relation of human with God (habluminallah)
1.Socialization 2.Externalization 3.Combination 4. Internalization
Horizontal relation of human with human (habluminannas) Command
Human assets
Combination
Prohibited
• • • •
Faith, • Good • Bad • 1.Socialization Belief in God, • • Sin Intellectual, – Tawhidic Worship, Righteous • • Emotional, Paradigm Morals Prohibited • Spiritual −> • Lawful motive/intention • Human as (niat), respect servants 2.Externalization (‘abd) – Tawhidic • Human as Paradigm −> caliphs patient (sabr) (khalifah) 3.Combination – Tawhidic Paradigm −> trust (amanah) 4.Internalization – Tawhidic Paradigm −> reward, ownership
The explanation of the role of tawhidic paradigm in the knowledge creation process of SECI below [12–14, 19, 20]: Socialization phase, where the converting tacit to tacit knowledge the social conditions in an organization usually indicates the willingness to share knowledge individually with an internal and external organizational member, high employee commitment and loyalty, more cooperation rather than competition, and organization facilitates as knowledge partner thus leadership in organization should able to lead the initial motivation of sharing knowledge amongst the employees. The employees probably gather information from other departments. They might collect work-related information and ideas from either informal or formal relationships with other employees to find problem-solving and better improvement in their work.
344
Nurhidayati and A. T. Adriyanto
The point that should be noticed from tawhidic paradigm: • The initial motive is to share or collect knowledge and information only for worship to Allah and obtaining ridho from Him. Allah says in Sura Al-Alaq 1–5: “In the name of your Lord Who created, created man from a clot of congealed blood and your Lord is Most Generous Who taught by the pen taught man what he did not know”. Allah also promises to raise people who have the knowledge, in Sura Mujadalah 11: “Allah will raise those who have believed among you and those who were given knowledge, by degrees. And Allah is Acquainted with what you do”. • The difference in ideas, perceptions and background potentially result in conflict between employees, thus respectful toward the differences and privacy is a must. In the externalization phase, where the converting tacit to explicit knowledge the social conditions in an organization usually indicate the high group commitment, and often the company obtains little pressure from shareholders thus the leadership in the organization should be able to create and organize the knowledge sharing culture. The organization needs to facilitate a such situation through creative and constructive discussion forums of sharing knowledge and enlarge, open, transparent and constructive two-way communication amongst the group members. The leadership in organization should able be to become wise and good communicators to listen to obstacles during the converting tacit to explicit knowledge. The point that should be noticed from tawhidic paradigm: • Obtaining new knowledge is a challenging process and needs more time to crystalized the tacit into the explicit knowledge, thus in the context of tawhidic paradigm, it needs to patience, persistence, and perseverance (sabr) and not in a hurry in demanding knowledge. Allah says in Sura Thaha 114: “Hence the High God is the real king, and do not hurry to read The Qur’an before its completion revealed it to you, and say: “My Lord, add to me knowledge”. Allah is always with those people such mentioned in Al-Qur’an Sura Al- Baqarah 153: “O, you who believe, seek help through patience and prayer. Surely, Allah is with those who are patient”. Combination phase, where the converting explicit to more complex and systemic explicit knowledge the social conditions in an organization usually indicate less interdepartmental rivalry, high employees’ commitment, and loyalty, and less informational abuse thus organization should be able to create and organize the knowledge sharing management culture especially relate with the technology used of data-base. The organization in a such situation needs to set up a quality of the informational system to reduce overlapping and redundancy of functional responsibilities provide free access
The Role of Tawhidic Paradigm in Knowledge Creation Process
345
to corporate information using technologies and data bases and play a role in consultative decision making. Practically, the employees engage in developing criteria and in conducting experiments on a new concept. The point that should be noticed from tawhidic paradigm: • The complexity and systematic of explicit knowledge sharing will be responded to by increasing the accessibility of the knowledge. In the context of twahidic paradigm, thus controlling such access should be paid attention as a crucial issue. As the explicit knowledge becomes the property and intangible assets of an organization then, trust (amanah) becomes specific values that need to be shared amongst all the organizational members. Including trustworthiness in ownership of property and intellectual rights and abuse is used. Allah says in Sura An-Nisa 58: “Indeed, Allah commands you to return trusts to their rightful owners, and when you judge between people, judge with fairness. What a noble commandment from Allah to you! Surely Allah is All-Hearing, All-Seeing”.
Sura Al Anfal 27: “O, you who have believed, do not betray Allah and the Messenger or betray your trusts while you know the consequence.” Internalization phase, where the converting explicit to tacit knowledge is closely related to learning by doing. The social conditions in an organization usually indicate little fear of mistakes. Thus, organizations should encourage and open opportunities for employees to learn new knowledge through trial and error in work practices and these will be supported with policies that intensively appreciate the practice of learning by doing. The employees often use new knowledge that they obtain from learning by doing as the source for the next time work projects. The point that should be noticed from tawhidic paradigm: • In the internalization phase, employees will manage and change newly explicit knowledge based on individual capacity and contribution. Thus, equality of rewarding (adl’) becomes a strategic policy of compensation. Awarding and compensating need to be selected to employees based on their contribution as property intellectual rights including the participation of employees during learning by doing. This is based on Al-Qur’an Sura Jaatsiyah 22: “For Allah created the heavens and the earth for a purpose, so that every soul may be paid back for what it has committed. And none will be wronged” The proposed model of Tawhidic paradigm in the knowledge creation process of SECI presents in Fig. 3.
346
Nurhidayati and A. T. Adriyanto
Faith
Belief in God Patient
Respect
Socialization
Externalization
Internalization
Combination
Equality Worship
Amanah
Morals
Fig. 3. The proposed model of Tawhidic paradigm in knowledge creation process of SECI
4 Conclusion and Future Research In conclusion, this paper shows that it is necessary to pay attention to moral, social, cultural, and religious values in the process of knowledge creation. This is because these values can shape organizational identity and assist organizations in making decisions and solving problems. These values can serve as a guide in behavior to realize the organization’s vision. In the organization, these values can also be used as the basis for the organization to form knowledge, systems, policies, and strategies in running the organization. One of the values that need to be instilled in individuals in the organization is religious values. Where this value can create new knowledge that can encourage individuals to behave in accordance with the rules or teachings in Islam. This religious value is based on morals, worship, belief, and faith so that individuals have guidelines in creating Islamic knowledge, and organizations can form business strategies that are following Islamic teachings. For future research, this research needs to be followed up with the development and application of technology sourced from the existing provisions of Islamic or sharia law.
References 1. Halawi, L.A., Aronson, J.E., McCarthy, R.V.: Resource-based view of knowledge management for competitive advantage. Electron. J. Knowl. Manag. 3(2), 75 (2005) 2. Porter, M.E.: Competitive Advantage. The Free Press, New York (1999)
The Role of Tawhidic Paradigm in Knowledge Creation Process
347
3. Pemsel, S., Widén, K.: Creating knowledge of end users’ requirements: the interface between firm and project. Proj. Manag. J. 41(4), 122–130 (2010) 4. Alavi, M., Leidner, D.E.: Knowledge management and knowledge management systems: conceptual foundations and research issues. MIS Q. 25(1), 107–136 (2001) 5. Kaba, A., Ramaiah, C.K.: Demographic differences in using knowledge creation tools among faculty members. J. Knowl. Manag. 21, 857–871 (2017). https://doi.org/10.1108/JKM-092016-0379 6. Mehralian, G., Nazari, J.A., Ghasemzadeh, P.: The effects of knowledge creation process on organizational performance using the BSC approach: the mediating role of intellectual capital. J. Knowl. Manag. (2018). https://doi.org/10.1108/JKM-10-2016-0457 7. Nonaka, I., O Nonaka, I., Ikujiro, N., Takeuchi, H.: The Knowledge-Creating Company: How Japanese Companies Create the Dynamics of Innovation, vol. 105. Oxford University Press, USA (1995) 8. Warkentin, M., Bapna, R., Sugumaran, V.: E-knowledge networks for inter-organizational collaborative e-business. Logist. Inf. Manag. 14, pp. 149–162 (2001). https://doi.org/10.1108/ 09576050110363040 9. Huang, Y., Basu, C., Hsu, M.K.: Exploring motivations of travel knowledge sharing on social network sites: an empirical investigation of US college students. J. Hosp. Mark. Manag. 19(7), 717–734 (2010) 10. Isa, M.Y.B.M., Mohamed, Z.B.: Insaniah (humanistic) economics of waqf: From IR 4.0 to a human-centered tawhidic society 5.0. Euro. J. Islam. Finan. 15 (2020) 11. Ismail, Y., Mhd, S.: The role of tawhidic paradigm in the transformation of management system. In: National Seminar on Islamic Management Systems Transformation, PWTC, Kuala Lumpur, October 2011 12. Nonaka, I., Toyama, R.: The knowledge-creating theory revisited: knowledge creation as a synthesizing process. In: The essentials of Knowledge Management, pp. 95–110. Palgrave Macmillan, London (2015) 13. Song, J.H., Uhm, D., Yoon, S.W.: Organizational knowledge creation practice: comprehensive and systematic processes for scale development. Leadersh. Organ. Dev. J. (2011) 14. Nonaka, I., Toyama, R., Konno, N.: SECI, Ba and leadership: a unified model of dynamic knowledge creation. Long Range Plan. 33(1), 5–34 (2000) 15. Maududi, A.A.: The Process of Islamic Revolution. Islamic Publication, Lahore, Pakistan (1977) 16. Al-Attas, S.M.A.: An Exposition of the Fundamental Elements of The Worldview of Islam. Kuala Lumpur, Malaysia: the International Institute of Islamic and Civilization (1995) 17. Faruqi, A.: Al Tawhid: Its Implications for Thought and Life (Issues in Islamic Thought). Herndon, Virginia, USA: International Institute of Islamic Thought (IIIT) (1994) 18. Anor Salim, F.A., Maidin, A.J., Mhd Sarif, S., Zainudin, D.: The developing entrepreneurship training curriculum based on Tawhidic paradigm and legal principles: a case study of Malaysia. Econ. Manag. Sustain. 4(2), 30–39 (2019). https://doi.org/10.14254/jems.2019.4-2.3 19. Sarif, S.M., Ismail, Y.: Sustaining knowledge management development through the Ulu Al-Albab approach. Int. J. Bus. Econ. Law 18(16), 37–44 (2019) 20. Andreeva, T., Ikhilchik, I.: Applicability of the SECI model of knowledge creation in Russian cultural context: theoretical analysis. Knowl. Process. Manag. 18(1), 56–66 (2011)
Islamic Human Values for Career Adaptability and Career Success of Millennial Generation Ardian Adhiatma(B) , Salsya Vivi Feronica Althof, and Meita Triantiani Department of Management, Faculty of Economics, Unissula Jln., Kaligawe Raya Km. 4, Semarang, Indonesia [email protected], {salsyavivifa, meitatriantiani}@std.unissula.ac.id
Abstract. The purpose of this study is to explore Islamic Human Value which is believed by the young millennial generation to determine career adaptability and career success. Career success is very important for today’s Muslim millennial generation because they have high expectations regarding work-life balance. To achieve career success, it is necessary to have career adaptability so that the Muslim millennial generation is able to prepare themselves to face unexpected transitions or changes. In addition, the Muslim millennial generation also needs to have Islamic values, namely Islamic human values. These values are important so that the millennial generation has good planning in building a career. This research is qualitative research by conducting structured interviews on 8 respondents. The results show that there are 6 Islamic human values that can be applied as the basis for achieving career adaptability and career success. Keywords: Islamic human values · Career adaptability · Career success · Millennial generation
1 Introduction In the current era, career aspiration in society has faced a paradigm shift. This paradigm shift also occurs in the Muslim millennial generation who have high motivation to achieve career success. This is motivated by the many job opportunities that exist today. Career aspiration can be interpreted as the ambition of the Muslim millennial generation which is currently starting to fill jobs. Career aspiration can be seen from various perspectives such as orientation, attitude, and behavior perspectives. In the orientation perspective, the Muslim millennial generation has high perfectionism towards the profession they dream of. They choose their dream profession based on Islamic religious principles, Islamic principles, qualifications, expertise, comfort both in terms of salary and work environment, and service among others. This side of perfectionism can be seen in their standard of career success, which is being able to provide maximum work results and feel happy with the work they are doing. Then from the perspective of attitude, the Muslim millennial generation has a strong stimulus or hope and can determine the line according to their career choice. The form of stimulus for the millennial Muslim © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 348–354, 2022. https://doi.org/10.1007/978-3-031-08812-4_33
Islamic Human Values for Career Adaptability and Career Success
349
generation in a career is family and close friends. From a behavioral perspective, the Muslim millennial generation is a generation that is smart, creative, hardworking, and has a good strategy in building their careers. This hard work can be seen from his efforts to achieve career success because he has high expectations regarding work- life balance. As stated by Smola and Sutton [1] that the millennial generation is looking for a different psychological contract and emphasizes a better work-life balance between work goals and personal goals. Millennials score higher on affiliation than any other generation and they are highly motivated by a cooperative workplace [2]. The right choice of work and a supportive work environment can help Muslim millennials build their careers. Currently, there are many career choices for the Muslim millennial generation such as content creators, YouTubers, online sellers, social entrepreneurs, and so on. In contrast to the career choices in the pre-millennial era, which were not as many as the millennial era. This can be influenced by the increasing number and variety of jobs available. Although there are many career choices for Muslim millennials, they may have the desire to change careers. This is because the Muslim millennial generation is identical to easily changing careers when they feel their work is not in accordance with their abilities and desires. Although the ease of changing careers is open, one of the important things that the millennial generation must have in a career is career adaptability. Career adaptability is the ability of individuals to prepare themselves in the face of various kinds of obstacles and unexpected changes and adapt to their new environment. To achieve this career adaptability, individuals need to have hope and hardiness within themselves [3]. Hope in the form of motivation to be able to achieve career success and hardiness in the form of good self-control when facing various kinds of problems or difficulties that occur. This can help millennials in dealing with a transitional environment. Please note, career adaptability is the most important factor that can help the Muslim millennial generation to have career success [4]. To help prepare for this transition, it is necessary to apply Islamic human values. Where this value can be used as the basis for forming attitudes and behavior in accordance with the principles and rules of the Islamic religion. Thus, the millennial generation can achieve career adaptability and be able to achieve career success. This study aims to explore the Islamic Human Values that they consider determining career success and career adaptability.
2 Literature Review 2.1 Career Adaptability Career adaptability is the most important characteristic that can effectively help individuals to deal with uncertainty in their careers. Career adaptability can also help individuals to prepare themselves to face various kinds of obstacles at work and adapt to their environment. Xu et al. [5] in their research stated that individuals who have a high level of career adaptability will have good problem-solving abilities, positive self-efficacy, good self-awareness, have clear learning goals, and good work performance. Career adaptability can be achieved if the individual has hope and hardiness. Hope is considered important in individual careers, especially in the Islamic millennial generation because today the business environment is experiencing uncertainty and demands self-disclosure,
350
A. Adhiatma et al.
adaptability, and resilience [3]. Individuals who have a high level of hope will have the motivation to achieve their goals in a career. Then individuals who have a high level of hardiness will be able to solve problems related to their work environment and be able to deal with various kinds of changes that occur unexpectedly. Usually, individuals who have a high level of hardiness are individuals who can control themselves when experiencing difficulties and will make it an opportunity within themselves [3]. Recent research on career adaptability conducted by Koen et al [6] states that career adaptability is the most important determinant of career outcomes. This finding examines career adaptability as an important component in the phenomenology of strategic career management among individuals. 2.2 Career Success Judge et al [7] Career success is a collection of achievements that arise from the work experience of an individual. Several researchers who researched the careers of Gattiker and Larwood [8]; Judge et al [9]; Nabi [10] In general, have conceptualized career success consisting of extrinsic and intrinsic outcomes because they measure this construct using objective indicators and subjective reactions. Extrinsic outcomes of career success (objective career success) include observable outcomes such as pay and promotions. Therefore, it will be relatively easier to observe than the intrinsic outcome of career success (subjective career success), which depends on one’s self-assessment. Career success is a combination of achieving the appropriate level of financial stability while doing the work you love and finding happiness and satisfaction in your chosen life and career. 2.3 Religious Personal Value Rehman and Shabbir [11] in their research show that religion is the most important dimension in shaping the knowledge, beliefs, and attitudes of individuals. Religiosity is shown to determine the extent to which individuals follow and apply values, practices, principles, and beliefs in religion [12]. Religious values are the main driving force for individuals to build and choose a career in accordance with the principles and rules of the Islamic religion [13]. This can affect individuals in behaving and behaving. As Muslims, it is necessary to apply Islamic principles and values in various aspects of life, including in building a career to achieve career success. These principles and values can be used as the basis that will become the core of human beings to always be honest, loyal, love, peaceful, etc [14]. In addition, the Islamic values possessed by Muslim millennials can help them to become better able to deal with stressful situations because these values tend to refer to goals that exist to motivate their actions [15]. Then when going to do an activity always involve Allah SWT in any circumstances because Allah SWT will always help and provide the best way out for His servants who always remember Him. Applying these Islamic values can also facilitate us in carrying out Allah’s commands and can also facilitate building a career for the future.
Islamic Human Values for Career Adaptability and Career Success
351
3 Conceptual Models (Fig. 1).
Fig. 1. Conceptual model
4 Methodology This research is a type of exploratory research [16]. This research method used qualitative data analysis methods, by conducting structured interviews. The interview material is about what Islamic Values they believed in choosing a career, treating career adaptability, and determining career success. This study used 8 sources informant consisting of 4 men and 4 women with an age range of 20–22 years. All of them are in the last year of the Bachelor’s degree program. The data were analyzed using the data display, data reduction, and conclusion method
5 Finding The results of the study in the form of a list of Islamic Human Values and the meanings understood by the informants are presented in Table 1. Table 1. Islamic human values No. Key findings
IHV informants
1
Informant 1 Istiqomah Informant 2 Value Informant 3 Informant 4 Informant 5 informant 6 Informant 7 Informant 8
Have a firm stance in every activity and never give up and look for solutions that are in accordance with Islamic principles
Categorization
(continued)
352
A. Adhiatma et al. Table 1. (continued)
No. Key findings
IHV informants
Categorization
2
It is necessary to seek the truth first when receiving information or news and consider it from an Islamic perspective
Informant 1 Tabayyun Informant 2 Value Informant 3 Informant 4
3
Always start activities by praying and balancing everything Informant 1 Tawazzun for the life of this world and the hereafter Informant 4 Value Informant 5 informant 6
4
Help each other between fellow Muslims so that everything Informant 1 Ta’awun feels easy and to increase solidarity Informant 2 Value Informant 3 Informant 5 informant 6
5
Have an attitude of tolerance between others and always be Informant 1 Tasamuh patient in dealing with problems because they believe that Informant 2 Value Allah SWT will always provide the best solution Informant 3 Informant 4 Informant 5 informant 6 Informant 7 Informant 8
6
Uphold the principle of honesty in carrying out various activities
Informant 1 Trust Value Informant 4 Informant 5 Informant 7
As an Islamic millennial generation, of course, they want to have a successful career that is in accordance with the teachings and rules of the Islamic religion. To be successful in a career, it is necessary to be prepared from within to be able to face various kinds of tasks and be able to adapt to the work environment. However, to prepare for various kinds of tasks (career adaptability) and to be successful in a career, of course, it is necessary to understand and believe in Islamic human values in everyday life. This study found several Islamic human values that can be applied and used as a basis for building a successful career and preparing themselves to face various kinds of tasks (career adaptability) such as the value of istiqomah, the value of ta’awun, the value of tasamuh, the value of trust, the value of tabayyun and the value of tawazzun. These values are very important to be applied in everyday life. Of the six values, there are several values that are very important as determinants of career success and career adaptability, namely the value of istiqomah, the value of tawazzun, and the value of trust.
Islamic Human Values for Career Adaptability and Career Success
353
The value of istiqomah in Islamic terminology is a strong stance. This value can be applied by always trying to solve problems and never giving up on finding solutions to problems by paying attention to the principles of Islam and the teachings of the Islamic religion. To achieve success in a career, of course, you will face various kinds of obstacles and various kinds of problems in completing tasks. For this reason, the value of istiqomah is very important to be able to achieve success in a career. Then the value of tawazzun, this value can be applied by always praying to ask for ease and smoothness in every activity. By praying, Allah SWT will facilitate all the affairs of His servants both in work and in an effort to achieve career success. trust value, this value can be applied by always being honest. Upholding the principle of honesty in everyday life is an attitude that is very commendable and very liked by Allah SWT. Not only being honest but the value of this trust can also be applied by always being responsible for what he is doing. That way, we can easily achieve success in career (Fig. 2).
Fig. 2. Islamic human values in career success and career adaptability
6 Conclusion and Future Research In conclusion, this paper concludes the importance of Islamic human values in career adaptability and career success. The application of Islamic human values such as istiqomah, amanah, tawazzun, ta’awun, tasamuh, and tabayyun values can help the millennial generation in preparing themselves to face transitions or unexpected changes in achieving career success. Ownership of these values is important so that the millennial generation has good planning in running a career. This is because career planning by applying Islamic human values can facilitate us in carrying out Allah’s commands and can also facilitate building a career for the future. For future research, we will validate the framework by using quantitative approach.
354
A. Adhiatma et al.
References 1. Smola, K., Sutton, C.D.: Generational differences: revisiting generational work values for the generational differences: revisiting generational work values for the new millennium. J. Organs. Behav. 23(4), 363–382 (2002). https://doi.org/10.1002/job.147 2. Wong, M., Gardiner, E., Lang, W., Coulon, L.: Generational differences in personality and motivation: do they generational differences in personality and motivation do they exist and what are the implications for the workplace ? J. Manag. Psychol. 23(8), 878–890 (2008). https://doi.org/10.1108/02683940810904376 3. Othman, R., Kamal, N.M., Alias, N.E.: Positive psychological traits and career adaptability among millennials. Int. J. Acad. res. Buses. soc. Sci. 8(9), 1420–1433 (2018). https://doi.org/ 10.6007/IJARBSS/v8-i9/4706 4. Wang, Z., Fu, Y.: Social support, social comparison, and career adaptability: a moderated mediation model. Soc. Behav. Pers. 43(4), 649–660 (2015). https://doi.org/10.2224/sbp.2015. 43.4.649 5. Xu, C., et al.: The role of career adaptability and resilience in mental health problems in Chinese adolescent children and youth services review the role of career adaptability and resilience in mental health problems in Chinese adolescents. Child. Youth Servant. Rev. 112(30), 104893 (2020). https://doi.org/10.1016/j.childyouth.2020.104893 6. Koen, J., Klehe, U., Van Vianen, A.E.M., Zikic, J., Nauta, A.: Job-search strategies and reemployment quality the impact of career adaptability. J. Vocat. Behav. 77(1), 126–139 (2010). https://doi.org/10.1016/j.jvb.2010.02.004 7. Judge, T.A., Higgins, C.A., Thoresen, C.J., Barrick, M.R.: The big five personality traits, general mental ability, and career success across the life span. Pers. Psychol. 52(3), 621–652 1999 8. Gattiker, U.E., Larwood, L.: Predictors for managers career mobility, success, and satisfaction. Hum. Relative. 41(8), 569–591 (1988). https://doi.org/10.1177/001872678804100801 9. Judge, T.A., Cable, D.M., Boudreau, J.W., Bretz, R.D.: An empirical investigation of the predictors of executive career success. 48(3), 485–519 (1995). https://doi.org/10.1111/j.17446570.1995.tb01767.x 10. Nabi, G.R.: An investigation into the differential profile of predictors of objective and subjective career success. Career Dev. Int. 4(4), 212–224 (1999) 11. Rehman, A., Shabbir, M.S.: Adoption the relationship between religiosity and new product adoption. J. Islam. Mark. 1(1), 63–69 (2010). https://doi.org/10.1108/17590831011026231 12. Worthington, E.L., et al.: The religious commitment inventory — 10: development, refinement, and validation of a brief scale for research and counseling. J. Couns. Psychol. 50(1), 83–96 (2003). https://doi.org/10.1037/0022-0167.50.1.84 13. Aziz, S., Husin, M., Hussin, N.: role of perceived trust factors that influence individuals’ intentions to purchase family takaful mediating role of perceived trust, Asia Pacific. J. Mark. logistics. (2019). https://doi.org/10.1108/APJML-12-2017-0311 14. Marian, R., Salsali, M., Vanaki, Z., Ahmadi, F., Hajizadeh, E.: Professional ethics as an important factor in clinical competency. Nurs. Ethics 14(2), 203–214 (2007) 15. Hystad, S.W., Bye, H.H.: Safety behaviors at sea: the role of personal values and personality hardiness. Saf. Sci. 57, 19–26 (2013). https://doi.org/10.1016/j.ssci.2013.01.018 16. Creswell, J.W.: Research Design Qualitative, Quantitative and Mixed Methods Approach. Student Library (2014)
FAST: A Conceptual Framework for Reducing Fraud Financial Statement in Financial Business Practice Ahmad Hijri Alfian(B) , Verina Purnamasari, and Dian Essa Nugrahini Department of Accounting, Faculty of Economics, Universitas Islam Sultan Agung, Semarang, Indonesia {hijrialfian,verinapurnamasari,dianessan}@unissula.ac.id
Abstract. Fraud is one thing that is avoided in every business activity. The impact of fraudulent activity will be very damaging and systemic for the company. In research, stated that one way to minimize financial statements fraud is by prioritizing internal control function, namely the Sharia Supervisory Board in the Islamic finance or banking industry. The sharia business concept applies several good things in every practice of business activity, namely Siddiq, Amanah, Fathonah, and Tabligh. In conclusion, the company has done everything discussed above without realizing that they had practiced one of the sharia business systems or concepts that Rasulullah Muhammad SAW did in ancient times. By improving and consistent attitude of each company to continuously carry out activities included in the indicators of Fathonah, Amanah, Siddiq, and Tabligh, it is expected to effectively and efficiently reduce and even avoid one of the disgraceful acts, namely fraud financial statements.
1 Introduction Fraud is one thing that is avoided in every business activity. The impact of fraudulent activity will be very damaging and systemic for the company. Another effect is a fraud will cause an unconducive working environment. Fraud is also a great challenge for the entire business world, especially in the banking and financial sector, which significantly contributes to the economic development of a country [1]. Fraud perpetrated in the financial industry will certainly have a major impact on the customer’s assessment of the company. The company’s operations will be significantly disrupted due to the main resource of the financial or banking industry is the power of participation from customers. A survey conducted by [2] stated that an increasing percentage of victims in the financial service industry is up to 46%. This happens when the investment climate is increasing and in good condition. Fraud that will be discussed in this research is fraud in financial statements as one of fraud in “Fraud Tree,” which is stated by the Association of Certified Fraud Examiners (ACFE) [3]. Fraud in financial statements is one of three types of fraud included in the fraud scheme. Fraud in financial statements is the most common fraud in the financial services industry because it is related to the applied model and modus operandi. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 355–363, 2022. https://doi.org/10.1007/978-3-031-08812-4_34
356
A. H. Alfian et al.
Financial service fraud must be eliminated by implementing everyday skills to minimize its activity [4]. The way to reduce financial statement fraud is by performing data analysis [5]. Data analysis can be done by applying standard operational procedures (SOPs) as a preventive action in the company. Other ways to minimize fraud is by internal monitoring process, financial performance is continuously maintained in good condition, and consistently and transparently reporting all progress, obstacle, and company condition through the financial statement for the stakeholders. These ways will help the companies to limit any fraudulent financial statements or potential activities. In several business practice models that are recently growing rapidly, the sharia industry has become one of them. The Islamic finance industry, which prioritizes the cultivation of Islamic law in carrying out all related activities, is one of its main attractions. In [6] research, stated that one way to minimize financial statements fraud is by prioritizing internal control function, namely the Sharia Supervisory Board in the Islamic finance or banking industry. Furthermore, this cannot be implemented for every company because of different business models. Therefore, there are sharia business concepts that can be applied generally in every business, and even unconsciously, every business industry has implemented it. The sharia business concept applies several good things in every practice of business activity, namely Siddiq, Amanah, Fathonah, and Tabligh [7].
2 Literature Review 2.1 Fraud Theory There are various kinds of fraud theories that several experts have stated. The first theory is the fraud triangle theory discovered by Donald. R Cressey in 1958. In the fraud triangle theory, it was found that three main factors cause fraud, namely Pressure, Opportunity, and Rationalization [8]. The second fraud theory is the fraud diamond theory. The theory suggested that four things underlie a person or an agency to commit fraud are Incentives, Opportunity, Rationalization, and Capability (Wolfe and Hermanson 2004) in [9] research. The next fraud theory is the fraud pentagon. The theory explains five main factors for someone to perform a fraud scheme. These include Incentives, Opportunity, Rationalization, Competence, and Arrogance (Howart 2011) in [10]. Another theory still related to the fraud theory is the Social Control Theory. This theory discusses the commitment resulting from a person’s social relationships who invest time and energy and do not want to be involved in criminal acts [11]. It can be concluded that someone who commits fraud is only a small part of what we usually refer to as unscrupulous persons and is accompanied by supporting factors mentioned above. 2.2 Financial Statement Fraud Financial statement fraud generally refers to the intentional alteration of a company’s financial statement to portray a different company image that often misleads users of financial information [12, 13]. Unlike some misstatements in financial statements that are unintentional errors, these changes have been carefully planned for the benefit of the perpetrators. Financial statement fraud can be classified into three main categories; changes in accounting methods, distortion in managerial cost estimation, and acceleration or delay of revenues and expenses recognition.
FAST: A Conceptual Framework for Reducing Fraud Financial Statement
357
2.3 Siddiq According to [14], Siddiq or honesty means being honest with yourself, others, and God. Siddiq indicators include a sense of responsibility towards Allah SWT as well as in working and seeking the truth both at the individual and social levels. Do not lie; true in thought, word, and deed. As Allah has mentioned in the Qur’an “O you who have believed! Fear Allah, and be with those who are true” (At Taubah 9:119). Siddiq is the belief that one must fulfill one’s obligations to achieve success in this world and the hereafter. When someone wants to be a successful person, one must obey Allah and His provisions and rules. At all times in acting, the board of directors must be honest and transparent. [15] found that honesty and ethical leadership increase employees’ trust in leaders. Hence, it influences employee attitudes, behavior, and cognition, while [16] suggest that strong and effective co-leadership can be achieved when the honesty attribute is adopted. Thus, any form of manipulation and concealment of the truth is not allowed. Financial statements and annual reports should contain accurate information, whereas earnings management and misstatement of financial information should be prohibited. To show the truth, directors must support their facts and numbers with tangible evidence to avoid further questions about the information integrity. Sufficient documents must be maintained to create a clear audit trail for each financial transaction. Therefore, the financial statements prepared by the board of directors are more trusted and assist users in making the right decisions, especially when audited by an independent and credible audit firm, [17–19]. In addition, any forward-thinking appointments, plans, and communications must be documented. These details will be a useful reference, especially when there is a dispute between the affected parties. 2.4 Amanah According to [14], Amanah is a sense of responsibility, honor, courtesy, showing optimal results, and respect for others. Allah mentions in the Qur’an, “Indeed, Allah commands you to return trusts to their rightful owners; and when you judge between people, judge with fairness. What a noble commandment from Allah to you! Surely, Allah is AllHearing, All-Seeing.” (An-Nisa 4:58). The faith that a leader must be fair. Being Amanah is when one judges among people, i.e. judge fairly. Being Amanah implies that the board of directors must practice the principle of alis (fairness). Due to the concepts of justice and Amanah being related, directors must ensure that justice is served at the highest values in their organization. In Islam, justice does not mean equality. Justice refers to putting the right thing in the right place or in the hands of the right people. [20] Emphasized that someone who is Amanah will be sincere in fulfilling his commitments, duties, and responsibilities. A committed director is an important factor for the success of the organization [21]. Directors should not use company resources for personal gain and take advantage of company insider information. This will increase market frictions and capital allocation inefficiencies that emerge from information asymmetry and agency problems. [22].
358
A. H. Alfian et al.
2.5 Fathonah Fatanah or wisdom according to [14] includes intelligence in attitude and knowledge, discipline, proactive attitude, and the ability to make the best decisions. Allah says in the Qur’an, “Allah will elevate those of you who are faithful and raise those gifted with knowledge in rank. And Allah is All-Aware of what you do” (Al-Mujadalah 58:11) Muslims have been taught manners or certain social behaviors in life. Allah knows the hardships and distress that we must endure. When making decisions, realize that Allah is well aware of your entire situation. Undeniably, knowledge is important because it contributes to the invention, innovation, and creativity, which leads to success. In today’s era, those who have the knowledge and intelligently apply their knowledge will be at the forefront of developing the present and shaping the civilization of the future. Knowledge is also able to produce a talented, skilled, creative, and innovative workforce [23]. Knowledge will enable leaders to make wise judgments, substantial added value in decision making, avoid risky choices [24], and achieve the best optimal growth and performance for the company. 2.6 Tabligh Tabligh, according to [14], is the ability to communicate, be accountable and transparent, able to deal with pressure and the ability to cooperate and work in harmony. In the current context, Tabligh is not to convey revelation, but to convey Islamic teachings through the Qur’an and Sunnah. The task of conveying revelation was completed with the Prophet Muhammad as the final messenger. As followers of the Prophet Muhammad, it is our duty to remind every Muslim to adhere to the teachings and practices of Islam. ‘Abdullah bin’ Amr bin Al-’As said, Rasulullah SAW stated, “Convey from me even one verse of the Qur’an…” (Hadits Al Bukhari). One must also use appropriate and polite words in conveying messages to others. Prophet Muhammad SAW said, “Whoever believes in Allah and the Day of Judgment must speak what is good or remain silent”. Using the right words is important because they describe good behavior and the true meaning of the message, although it doesn’t necessarily mean that better words mean good while bad words contain lies. In spreading the da’wah, the Prophet Muhammad SAW responded well even though people talked about it harshly. Harsh words were returned in a soft and beautiful tone. Such good behavior indeed pleases many non-believers to return to Islam. In this regard, the board of directors must communicate well with internal and external parties of the company. This will create a good relationship with management and subordinates, showing that they are approachable and negotiable. For external parties, good communication reflects a good corporate image, thus increasing the confidence of stakeholders in trusting the board of directors’ ability to manage the company. [25, 26] found that the increased disclosure of financial information can substantially support company performance. Therefore, directors must ensure the company completes all disclosure requirements by regulatory agencies. They must also be responsive to all matters of concern to stakeholders.
FAST: A Conceptual Framework for Reducing Fraud Financial Statement
359
3 Conceptual Framework Based on the literature review described above, the conceptual framework is as follows. (Fig. 1).
Fig. 1. FAST for reducing fraud financial statement
Based on the above concept, it can be seen how Fathonah, Amanah, Siddiq, and Tabligh as one of the concepts of sharia business that can reduce the potential for fraudulent financial statements. Fathonah, which means intelligent, can be seen in the company’s financial performance. How the company can generate profits well so that the financial performance ratios will also be well illustrated. Good financial performance is also produced by wise company leaders thus the company can grow and maximize to get optimal results. Good financial performance ratios will make the company get good financial stability. Good financial stability will reduce the potential for fraudulent financial statements. As in [27] research’s that companies experiencing unstable financial conditions will try to cover it up by manipulating financial statements. Amanah, which means responsible and fair, can be seen from the stock price returns and the choice of a public accounting firm as the auditor of the company’s financial statements. Stock
360
A. H. Alfian et al.
price returns will describe how the response and assessment of investors to the company. A good form of responsibility from the company will get a good assessment from investors which are reflected in the increasing return of the company’s stock price. The company’s high stock price comes from really good company performance which will reduce the potential for fraud in the form of a conflict of interest [28]. Prevention of fraudulent conflicts of interest will also cover other potential frauds and does not rule out fraud in financial statements. The company’s choice to use the best auditor, namely the “Big Four” is also a form of accountability from the company to stakeholders. This is to ensure that financial statements containing company resources have been reported properly and follow generally accepted standards and describe the actual condition of the company. The company’s choice to use the “Big four” auditors will reflect audit quality, as in research [29] which has proven that audit quality has an effect on earnings manipulation in financial statements. As well as research conducted by [30] succeeded in proving that there is an influence between fraud detection in financial statement audits and audit quality. Siddiq is honesty. Honesty in the company can be reflected in the results of the audit in the form of an audit opinion. The audit opinion with the highest score, namely Unqualified Opinion, obtained by the company can be an indication that the company has been honest and good in submitting the company’s financial statements. As in the [31] research’s found that Unqualified Audit Opinions with Explanatory Language can be used to detect fraudulent financial statements. The number of independent commissioners owned by the company can also illustrate that the company is committed to activities in accordance with the company’s goals, and according to the regulations and corridors that have been determined. The number of independent commissioners will affect a person’s opportunity to commit fraud [8]. The number of independent commissioners also describes the level of internal control. Good internal control will reduce the potential for fraudulent financial statements [32]. Research from [33] also proves that the number of independent commissioners has a negative effect on the potential for fraudulent financial statements, which means that the greater the number of independent commissioners, the less likely the fraudulent financial statements will occur. The honesty that exists in the company can also be seen from the existing leadership. Good leadership will also have a good effect on all aspects of the company, such as employee attitudes and performance, including all policies taken. All good aspects will be reflected in one of the company’s assessment indicators, namely good corporate governance. As in the research from [34] proves that good corporate governance has a negative effect on financial statement fraud. The better the good corporate governance, the smaller the potential for fraudulent financial statements. Tabligh which means to convey can be seen from several indicators in Internet Financial Reporting (IFR). The times that have made the transition from manual to internet-based digital systems make companies have to innovate to use the web to convey all useful information, including financial statements. The indicators in the IFR become a benchmark for how companies want to be more transparent in conveying information so as to gain the trust of stakeholders. The application of Internet Financial Reporting is one of the applications of technological innovation in the company. As in [35] explains that companies have spent a lot of money to improve technology related to improving the
FAST: A Conceptual Framework for Reducing Fraud Financial Statement
361
quality of the information in financial reports and to show that there is no manipulation in it. The timing of the submission of financial statements can also indicate that the company has tried to establish good communication with users of financial statements. In [36] the authors state that independent commissioners encourage company management to submit financial reports in a timely manner while not committing fraud in the process of presenting financial statements. The timely submission of financial reports will help the parties to make decisions immediately.
4 Conclusion and Future Research The conclusion is the company has actually done everything that is written in the conceptual framework. Activities carried out without them realizing it are included in the sharia business system or sharia concept that has been practiced by Rasulullah SAW in ancient times. In the future, hopefully, there is an improvement and a consistent attitude from each company to continue carrying out activities that are included in the indicators of Fathonah, Amanah, Siddiq and Tabligh. The activities carried out are proven to effectively and efficiently reduce and even avoid fraudulent financial statements. Future research should be able to prove that the indicators of each variable from Fathonah, Amanah, Siddiq, and Tabligh really have an effect and succeed in reducing the potential for fraudulent financial statements by using a sample of companies in various types of industry sectors.
References 1. Repousis, S., Lois, P., Veli, V.: An investigation of the fraud risk and fraud scheme methods in Greek commercial banks. J. Money Laund. Control 22(1), 53–61 (2019). https://doi.org/ 10.1108/JMLC-11-2017-0065 2. PriceWaterhouseCoopers, Time to change tack a need for new thinking by the financial services industry (2016) 3. Selimo˘glu, S.K., Altunel, M.: Forensic accounting and fraud audit in Turkey (2008–2018): an academic literature review and classification. Contemp. Stud. Econ. Finance Anal. 102, 219–244 (2020). https://doi.org/10.1108/s1569-375920200000102017 4. Frimpong, K.: Back to basics: fighting fraud and austerity. J. Finance Crime 22(2), 219–227 (2015) 5. Aboud, A., Robinson, B.: Fraudulent financial reporting and data analytics: an explanatory study from Ireland. Account. Res. J. 35(1), 21–36 (2022). https://doi.org/10.1108/ARJ-042020-0079 6. Ngumar, S., Fidiana, F., Retnani, E.D.: Implications of Islamic governance on Islamic bank fraud. J. Reviu Akunt. dan Keuang. 9(2), 226 (2019). https://doi.org/10.22219/jrak.v9i2.8735 7. Misbach, I.: Perilaku Bisnis Syariah. J. Manaj. Dakwah 5, 33–44 (2017) 8. Skousen, C.J., Smith, K.R., Wright, C.J.: Detecting and predicting financial statement fraud: the effectiveness of the fraud triangle and SAS No. 99. Adv. Finance Econ. 13(99), 53–81 (2009). https://doi.org/10.1108/S1569-3732(2009)0000013005 9. Sujana, E., Yasa, I.N.P., Wahyuni, M.A.: Testing of fraud diamond theory based on local wisdom on fraud behavior. Adv. Econ. Bus. Manag. Res. 69(3), 12–15 (2019). https://doi. org/10.2991/teams-18.2019.3
362
A. H. Alfian et al.
10. Danuta, K.S.: Crowe’s Fraud Pentagon Theory dalam Pencegahan Fraud pada Proses Pengadaan melalui E-Procurement. J. Kaji. Akunt. 1(2), 161–171 (2017). https://doi.org/10. 33603/jka.v1i2.826 11. Lokanan, M.E., Aujla, I.: Mapping the individual and structural theories of financial crimes. J. Financ. Crime 28(2), 420–432 (2020). https://doi.org/10.1108/JFC-12-2019-0165 12. ACFE, Global Fraud Study: report to the nation on occupational fraud and abuse Asia Pacific Edition (2018) 13. Robinson, W.N., Aria, A.: Sequential fraud detection for prepaid cards using hidden Markov model divergence. Expert Syst. Appl. 91, 235–251 (2018) 14. Rahman, Z.A., Shah, I.M.: Measuring Islamic spiritual intelligence. Procedia Econ. Finance 31, 134–139 (2015) 15. Mozumder, N.: A multilevel trust-based model of ethical public leadership. J. Bus. Ethics, 1–18 (2018) 16. Donnelly, M., Graham, W.: Co-leadership: public sector case studies using reciprocal expectations. Dev. Learn. Organ. Int. J. 32(2), 13–15 (2018) 17. Jais, K., Nawawi, A., Salin, A.S.A.: Reduction of audit quality by auditors of small and medium size audit firms in Malaysia: a case of premature sign-off of audit documents. J. Account. Bus. Manage. 23(2), 1–12 (2016) 18. Husnin, A., Nawawi, A., Salin, A.S.A.: Corporate governance and auditor quality – Malaysian evidence. Asian Rev. Account. 24(2), 202–230 (2016) 19. Asmuni, A.I., Nawawi, A., Salin, A.S.A.: Ownership structure and auditor’s ethnicity of Malaysian public listed companies. Pertanika J. Soc. Sci. Humanit. 23(3), 603–622 (2015) 20. Abdullah, M.: The role of Islam in human Capital development: a juristic analysis. Humanomics 28(1), 64–75 (2012) 21. Suhaimi, N.S., Nawawi, A., Salin, A.S.A.: Determinants and problems of successful ERP implementations – Malaysian experience. Int. J. Adv. Oper. Manag. 9(3), 207–223 (2017) 22. Chen, Z., Huang, Y., Kusnadi, Y., Wei, K.J.: The real effect of the initial enforcement of insider trading laws. J. Corp. Financ. 45, 687–709 (2017) 23. Ahmad, N.M.N., Nawawi, A., Salin, A.S.A.P.: The relationship between human Capital characteristics and directors’ remuneration of Malaysian public listed companies. Int. J. Bus. Soc. 17(2), 347–364 (2016) 24. Amirudin, N., Nawawi, A., Salin, A.S.A.P.: Risk management practices in tourism industry – a case study of resort management. Manag. Account. Rev. 16(1), 55–74 (2017) 25. Andrikopoulos, A., Merika, A., Triantafyllou, A., Merikas, A.G.: Internet disclosure and corporate performance: a case study of the international shipping industr. Transp. Res. Part A Policy Pract. 47, 141–152 (2013) 26. Ashbaugh, H., Johnstone, K.M., Warfield, T.D.: Corporate reporting on the internet. Account. Horizons 13(3), 241–257 (1999) 27. Hafizah, N., Respati, N.W., Chairina, C.: Faktor-Faktor Yang Mempengaruhi Kecurangan Laporan Keuangan Dengan Analisis Fraud Triangle. J. Reviu Akunt. dan Keuang. 6(1), 811– 822 (2017) 28. Martantya, D.: Pendekatan Kecurangan Laporan Keuangan Melalui Faktor Risiko Tekanan dan Peluang. Diponegoro J. Account. 2(2), 1–12 (2013) 29. Indarti, Deteksi Potensi Resiko Kecurangan Pada Laporan Keuangan Dengan Menguji Pengaruh Kualitas Audit, Size, Leverage Dan Financial Target Terhadap Manajemen Laba Riil. J. Ilm. Akunt. Kesatuan, 5(2), 086–181 (2017) 30. Sarwoko, I.: Pengaruh Ukuran KAP dan Masa Perikatan Audit Terhadap Penerapan Prosedur Audit Untuk Mendeteksi Risiko Kecurangan Dalam Laporan Keuangan, Serta Implikasinya Terhadap Kualitas Audit. J. Akunt. XVIII(01), 1–20 (2014) 31. Nuranggraini, S., Yusuf, M.: Analisis Fraud Pentagon Dalam Mendeteksi Financial Statement Fraud. J. Akunt. 1–23 (2020)
FAST: A Conceptual Framework for Reducing Fraud Financial Statement
363
32. Rahayu, F.J., Prabowo, T.J.W.: Pengaruh Sistem Pengendalian Intern Terhadap Kualitas Laporan Keuangan Pemerintah Daerah. Diponegoro J. Account. 7(4), 1–14 (2018) 33. Tiffani, L., Marfuah, M.: Deteksi Financial Statement Fraud Dengan Analisis Fraud Triangle Pada Perusahaan Manufaktur Yang Terdaftar Di Bursa Efek Indonesia. J. Akunt. Dan Audit. Indonesia, 19(2), 112–125 (2009) 34. Muzdalifah, M.: Pengaruh good corporate governance terhadap fraud pada perusahaan perbankan. J. Ris. Perpajak. 3(1), 56–64 (2020) 35. Roszkowska, P.: Fintech in financial reporting and audit for fraud prevention and safeguarding equity investments. J. Account. Organ. Chang. 17(2), 164–196 (2020). https://doi.org/10. 1108/JAOC-09-2019-0098 36. Dewi, I.G.A.R.P., Wirakusuma, M.G.: Fenomena Ketepatwaktuan Informasi Keuangan Dan Faktor Yang Mempengaruhi Di Bursa Efek Indonesia. E-Jurnal Akunt. Univ. Udayana 8(1), 171–186 (2014)
Risk Management and Islamic Value: A Conceptual Development of Al-Adl Financing Risk Management Jumaizi(B) , Widiyanto bin Mislan Cokrohadisumarto, and Eliya Tuzaka Faculty of Economics, Universitas Islam Sultan Agung (UNISSULA), Semarang, Indonesia {jumaizi_pdim5,eliyatuzaka}@std.unissula.ac.id
Abstract. This study is based on the weakness of the previous studies in the Financial Risk management concept which was then reconstructed into a Financial Risk Management concept based on values derived from the Holy Qur’an and Hadith. A new concept called Al-Adl Financing Risk Management (AFRM) will be developed along with its new dimensions and measurement scale. This concept is expected to provide a new financial risk management model which is a synthesis of portfolio theory, towards the concept of risk management and the Islamic worldview in the concept of fair management. This new concept is motivated by the phenomenon of the low distribution of financing funds in mudarabah in Sharia Bank (Islamic Rural Banks or Bank Pembiayaan Rakyat Syariah (BPRS)). This is due to various obstacles such as legal requirements and moral hazard issues, high risk, lack of awareness, and low return. This condition becomes a challenge for BPRS in channeling its financing with low risk so that the objectives of increasing Repayment of Financing and Mudharabah Financing Performance can be achieved. Therefore, it is necessary to carry out adequate management, so that BPRS are required to have a good understanding of Financing Risk Management. This study is expected to produce a new conception and its dimensions, and it is necessary to test the validity of each dimension empirically. Keywords: Risk management · Islamic value · Al-Adl financing risk management
1 Introduction Organizational success (in this case, BPRS) is highly dependent on the ability of managers to understand and control the level of risk taken in managing their business strategy (Enterprise Risk Management). That investment will face both return and risk [1]. Here, Islamic banks or Islamic rural banks (BPRS) in distributing mudharabah financing are required to understand financing risk management. However, understanding and controlling the level of risk is not limited to knowledge and skills only, but attitudes and actions to make decisions appropriately and fairly in the workplace are very important. Mudharabah contracts (Trust Financing) are recommended by many Islamic economists. This is because it has advantages, among others, based on the values of © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 364–372, 2022. https://doi.org/10.1007/978-3-031-08812-4_35
Risk Management and Islamic Value: A Conceptual Development
365
justice and free from usury, Gharar and Maisyr. However, not many financial institutions have implemented those types of financing due to various barriers to moral hazards [2]. Recording moral hazard reports from customers who report losses in their financial statements is necessary to avoid payments to the Bank [3]. In addition, the contract has a high risk. The issue of high risk and multi-faceted business risk becomes the main obstacle in the implementation of Mudharabah. Despite the presence of the Investment Account Platform (IAP) which was initiated by the Malaysian Central Bank (Bank Negara Malaysia - BNM) to encourage risk-sharing, the role of the bank remains the same as a Mudarib who mobilizes funds from investors to ventures through bank financing facilities using exchange-based contracts instead of equity-based contracts [4]. [5] found 4 (four) main factors that hinder the implementation of mudharabah contracts, namely high risk, lack of awareness, regulatory constraints, and low return. Furthermore, [5] found that high risk, asymmetric information problems, moral hazards, and a difficult financing evaluation process are becoming the main reasons for Islamic banks to not offer financing products with profit-sharing contracts. Meanwhile, regulators also need better risk mitigation for the scheme. Based on the above background, it is necessary to intervene in the role of Financing Risk Management with Islamic values, which are based on the value of justice. Financing risk management capabilities need to be built with Islamic values in order to get a fair balance of rights and obligations of each partner (Al-Adl) so that partners can benefit from entering into Mudharabah contracts. From the weakness of the concept of some literature, it is necessary to intervene Al-Adl with Financing Risk Management. These elements of Islamic values are very much needed in building Financing Risk Management in order to get the best solution. This study becomes interesting because religion should be a reference in making decisions. Thus, this research explores how the best solutions in dealing with situations that cannot be predicted and are not oriented to individual interests but for the common good and welfare, in organizational life. With the implementation of Islamic values, decisions will benefit people. The purpose of this study is to provide input for managers in implementing financing fairly and managing financing risks properly and systematically reviewing the concept of financing performance by developing concepts and identifying the validity of the measurement factors of the Al-Adl Financing Risk Management (AFRM) indicator so as to provide new strength for future research. The urgency of the research is to intervene in the value that gives transcendental meaning, namely Al-Adl, a new concept of the literature on the performance of Mudharabah financing through Al-Adl Financing Risk Management (AFRM). This concept is expected to make a significant contribution to BPRS managers to improve their ability to evaluate financing and its implementation. The development of fairness principles in financial risk management for the management of mudharabah financing will improve the Repayment of Financing and Mudharabah Financing Performance BPRS in Indonesia.
366
Jumaizi et al.
2 Literature Review 2.1 Al-Adl Financing Risk Management Risk management is the identification, measurement, monitoring, control, and reporting of the risk [6]. In [7] the author argues that risk management is a comprehensive process by understanding risk and risk management is carried out by identifying risks related to bank transactions and operations then assessing and analyzing various types of risks and monitoring bank risks. [8] divided the risk management process into four dimensions: risk identification, qualitative risk assessment, quantitative risk assessment, risk mitigation and treatment. Therefore, there are several dimensions in the risk management process including understanding, identifying, assessing, mitigating and monitoring and handling risks. In [9], the author suggests that risk identification can be carried out in a methodological way to ensure that all important activities in the organization have been identified, and all risks that follow these activities have also been identified and categorized. Thus, business activities and decisions can be classified as follows: strategic management risk, operational management risk, financial management risk, knowledge management risk and compliance management risk. In [10], the author mentions 7 Types of Financial Risks including Asset-backed risk, financing risk, Foreign Investment risk, Currency risk, Liquidity, Stock Market risk, and Interest Rate risk. In the Islamic perspective, risk management is very important for the implementation process according to Islamic principles. But today, risk management is often associated with the use of derivatives where the original design of the derivative is to manage risk. According to most scholars, it is argued that when derivative instruments are used to manage risk, gambling activities will begin and risk management activities will end [11, 12]. Previous studies by [13–17] allow the use of derivatives as long as they are used only for risk management purposes. Holy Qur’an surah al-Lukman verse 34 expressly that Allah commands all humans to always try so that something unexpected happens, meaning that Islam teaches to manage risk. However, the risk management allowed in Islam must uphold the principle of justice so that everyone’s rights are protected as referred to in a hadith, Prophet Muhammad reported that Allah is with people who cooperate with each other (syirkah), including in business, as long as the parties do not mutually treason (HR. Abu Dawud, authenticated by al-lHakim, from Abu Hurairah). However, the risk management process in Islamic finance is generally the same as in conventional finance [18]. This is what distinguishes Islamic finance from an additional process, namely Sharia screening. Negative sharia screening refers to the prohibition of usury, gharar, and maysir. On the other hand, a positive Sharia screening refers to the achievement of maqasid al-sharia. Risk management with the principle of fairness is the foundation that must be applied. In Arabic, it gives a statement that “fair or adl” has the meaning of “equal”. Meanwhile, according to Indonesian Dictionary, the word “fair” is defined as: (1) impartial, (2) obey the truth, and (3) proper / not arbitrary. Therefore, the principle of fairness in risk management must be able to make someone behave impartially and only obey the truth that is not arbitrary.
Risk Management and Islamic Value: A Conceptual Development
367
Holy Qur’an also makes it clear that Allah commands us to be enforcers of justice AnNisa’ (4): 135, An-Nahl (16): 90, Al-Ma-idah (5): 8, Al-A’raf (7): 96). The series of verses above remind a Muslim to behave fairly which will lead to piety, and result in prosperity. In addition, humans as caliphs on earth must accept and obey all His provisions (qada’ and qadar). Thus, individuals must be able to face the test by remaining obedient and behaving according to Islam. Allah says in Al-Baqarah verse 282–283 that it is important to have debt contract documentation before entering into the contract as a form of carrying out justice. In Yusuf verse 55 adds an order for someone to be good at guarding and knowledgeable. This shows that in order to achieve fairness and organizational performance (banks/companies), a trustworthy and knowledgeable manager is needed. Meanwhile, as stated in Yusuf verse 67, managers must be able to minimize the risks from various directions according to Islam in order to obtain success. In line with this, Holy Qur’an at-Thaha verses 31–32 states that Musa became a prophet because he has power that comes from Allah and makes Allah a friend in all his affairs. Therefore, the strength or knowledge of a manager comes from Allah so it is expected that the manager of a bank/company always relies on Allah in all his affairs. This study designs a theoretical model using the dimensions of Al-Adl Financing Risk Management (AFRM) and Financing performance. Based on the substantive and strategic dimensions to form a new concept through the proposition of Al-Adl Financing Risk Management (AFRM). Thus, Al-Adl Financing risk management (AFRM) is defined as the ability to manage risk by assessing and treating equally, monitoring and communicating in a balanced, selective and agreeing to accept the assessment. 2.2 Repayment of Financing In [19], the author stated that achieving a high loan collection rate (return rate) is a very important condition for microfinance institutions to be independent in the long term. According to [20], repayment of financing occurs when the borrower cannot distinguish between low-risk and high-risk borrowers and the loan contract is subject to limited liability. Furthermore, there are several factors that determine loan repayments including macroeconomic determinants regarding (a) macroeconomic stability can be the cause of low levels of non-performing loans (b) credit rationing can explain the low level of loan maturity because banks will only lend money to high-quality borrowers. Thus, Repayment of Financing can be influenced by macroeconomic stability and allotment of financing [20]. According to [21], the rate of loan repayment is the most common measure of the performance of a microfinance program and it greatly influences whether a program is likely to be financially sustainable. The rate of return can refer to the ratio between three categories, the first of outstanding loans divided by the total outstanding loans (four categories), and the ratio of borrowers in arrears refers to the number of borrowers who do not repay their loans to maturity relative to the total borrowers [22]. In addition, [23] stated that loan repayments can be measured using the debt-service ratio to income. This study measures the rate of return with outstanding loans, total outstanding loans, the ratio of debtors in arrears.
368
Jumaizi et al.
In [24], the authors stated that effective lending procedures have far-reaching consequences on loan repayments. To ensure loan security, proper market analysis, business screening, and follow-up are carried out. Without effective mechanisms in place, non-performing loans are unavoidable and loan recovery may be a major challenge for microfinance institutions. In [25], the authors believed that repayment is a willingness to pay back in the financial market and as the key to the existence of a healthy financial system. In particular, institutional arrangements to ensure debtor payments should be one of the main concerns of policymakers overseeing the operation of the banking system, as market financing failures generally stem from imperfect information and/or limited enforcement. Therefore, it needs the willingness to pay and debtor payments. 2.3 Mudharabah Financing Performance In [26], the author defines performance as the result of many individual decisions made continuously by management. According to [27] performance is a determination of the effectiveness of the organization’s operations on a regular basis. Meanwhile, in the Indonesian dictionary, performance is a noun that means something achieved, demonstrated achievement, and workability. Based on the definition of performance, this study focuses on the performance of Islamic banks where the measurement of the results that have been achieved such as the comparison of the amount of mudharabah financing with all financing provided by Islamic banks, the rate of return and profit-sharing, and the growth of total assets and net income [28]. According to Law No. 10 of 1998, financing based on Islamic principles is defined as the provision of money or an equivalent claim based on an agreement between the bank and another party that requires the party being financed to return the money or bill after a certain period of time in exchange for or profit-sharing. One of the forms of financing with Islamic principles is mudharabah as an ideal instrument in dealing with high financing risks. Based on accounting principles, mudharabah financing is a business cooperation agreement between the bank as the owner of the funds (shahibul mal) and the customer as the manager (mudarib) to carry out business activities with a profit-sharing ratio (profit or loss) according to the previous agreement [29]. According to [30], mudharabah is a partnership contract, which involves two parties, one of which is called rabb al-Mal, providing capital, and mudharib providing labor. In this case, rabb al-Mal is responsible for money, while mudharib is required to manage money with his agreed skills and knowledge [31]. Mudharabah became a popular commercial association among Arabs before and after Islam, especially as a long-distance trade [32]. In addition, the application of mudharabah as a sharia instrument that reduces the high-risk [33]. Based on the description above, it has been obtained the definition of Mudarabah Financing Performance is a performance or achievement of Islamic banking which is measured based on the results achieved such as the comparison of the amount of mudharabah financing with all financing provided by Islamic banks, the rate of return and profit-sharing, growth in total assets and net income, as well as being able to place funds in accordance with the agreed proportions. Through an in-depth study, this research finally succeeded in synthesizing the Islamic values of Al-Adl, the concept of risk management, and the principles of Islamic risk management, the following propositions can be drawn up:
Risk Management and Islamic Value: A Conceptual Development
369
Al-Qur’an and Hadist Risk Management Concept
Islamic Risk Management Principle Enterprise Risk Management
Principle of Justice Strategic M Risk
Operational M Risk
Asset Backed Risk
Financial M Risk
Foreign Invesment Risk
Knowledge M Risk
Compliance M Risk
Liquidity Risk
Currency Risk
Transparency
The Forbitddance of Usury The Forbiddance of Activities Which Have Forbiddance Elements
Stock Market Risk
Interest Rate Risk
Financing/ Credit Risk
The Forbiddance of Gharar
Al’Adl Financing Risk Management
Fig. 1. Integration of Islamic risk management principle and risk management concept
In general, banks are faced with various types of risks that affect their performance, namely strategic risk, credit risk, liquidity risk, market risk, and operational risk. However, Islamic banking encounters unique types of risk, namely, Sharia non-compliance risk, return risk, commodity, and inventory risk. This study is focused on credit risk or financing risk. Credit risk is considered the most important risk for banks because of its strong relationship with bank profitability and the economic growth [34]. Financing risk is a loss due to the obligor’s failure to pay debts. Credit or financing risk is the main variable that affects net income and market value of equity arising from credit or non-performing loans. It is known that the more banks obtain profitable assets, the greater the risk that the lender will fail to pay principal and interest by the specified date. In [8], the author also divides the risk management process into four dimensions: risk identification, qualitative risk assessment, quantitative risk assessment, risk mitigation, and treatment. Al’Adl Financing Risk Management combines financial or credit risk with the Islamic risk management principle, in which risk management is important in getting an offering solution from an Islamic perspective. Risk management is stated in the Holy Qur’an Surah Luqman verse 34, stating that Islam rigorously teaches to manage risk (mitigate
370
Jumaizi et al.
risk). In this case, the Islamic Risk Management Principle used the principle of justice, which is expected to create a balance between the parties bound by the contract. Loans given to people who are ineligible and unable to repay reflect oppression and injustice as this group is subject to higher rates due to their high-risk profile. Therefore, justice can provide equality, balance, choosing partners selectively, and agreements to produce Taqwa and prosperity. This study believes that Repayment of Financing and Mudharabah Financing Performance can be realized through the Al-Adl Financing Risk Management approach. Al-Adl Financing Risk Management is an integration of dimensions from previous research [8] with the inclusion of Al-Adl Islamic values. In this case, Al-Adl Financing Risk Management is the ability to identify, assess and monitor risks in a balanced, selective, and agreed on manner. This professional ability of Al-Adl Financing Risk Management has the potential to realize Repayment of Financing and Mudharabah Financing Performance. Thus, the realization of Mudharabah Financing Performance is influenced by the increase in Repayment of Financing built by Al-Adl Financing Risk Management. From the description above, the following propositions can be drawn up: Al-Adl Financing Risk Management Al-Adl Financing Risk Management is the ability to identify, assess and monitor risks in a balanced, selective and agreed manner. The professional ability of Al-Adl Financing Risk Management has the potential to realize Repayment of Financing and Mudharabah Financing Performance.
3 Methodology This research is explanatory research that emphasizes the relationship between research variables by testing hypotheses. The data sources of this research are (a) primary data in the form of respondents’ responses to the variables of Al-Adl Financial Risk Management, Rate of Repayment of Financing, and Financing Performance (b) secondary data in the form of lists of Islamic Rural Banks or Bank Pembiayaan Rakyat Syariah (BPRS) in Indonesia. The method of data collection was done by distributing questionnaires including open and closed questions. The population of this research is the Manager of Islamic BPR in Indonesia. The population of this research is the Manager of Islamic BPR in Indonesia. While the sampling used a purposive sampling technique with sample characteristics, namely BPR Syariah in Indonesia which uses financing (investment) contract from a mudharabah contract. The number of samples (sample size) refers to the opinion [35] which says that the number of samples is an indicator multiplied by 5 to 10 or at least 100 respondents. In this study, there are 24 indicators, while the basis for calculating the sample is 24 multiplied by 5 or equal to 120 samples. Technical analysis of the data using The Structural Equation Modeling (SEM) from the AMOS 20.0 package. The steps in SEM, according to [36] are (a) theory-based model development, (b) path diagram development, (c) flow chart conversion into equations, (d) input matrix and model estimation, (e) assess the possibility of identification problems, (f) evaluation of goodness-of-fit criteria, (g) interpretation and modification of the model.
Risk Management and Islamic Value: A Conceptual Development
4
371
Conclusion and Future Research Agenda
Al-Adl Financing Risk Management (AFRM) is fair risk management of financing, which is considered capable of improving the performance of Mudharabah financing. Managers who have AFRM will always carry out risk management by assessing, treating equally, monitoring and communicating in a balanced, selective manner, and agreeing to accept the assessment. Such behavior will form management that always prioritizes justice in every decision-making. If every manager has AFRM capability, then they are able to increase financing repayments and improve mudharabah financing performance. Future research should involve validating the measurement of each variable and testing the concept of Al-adl financing. Risk management at the empirical level and this model is very appropriate to be tested in the Islamic banking industry, with the manager as the unit of analysis.
References 1. Markowitz, H.: Portfolio selection. J. Finance, 7(1), 77–91 (1952). http://links.jstor.org/sici? sici=0022-1082%28195203%297%3A1%3C77%3APS%3E2.0.CO%3B2-1 2. Sapuan, N.M.: An evolution of mudarabah contract: a viewpoint from classical and contemporary islamic scholars. Procedia Econ. Financ. 35(February), 349–358 (2016). https://doi. org/10.1016/s2212-5671(16)00043-5 3. Ashraf, L., Hussain, H.: Mudarib in a Mud.a¯ rabah contract. Muzakarah of Archipelago Sharia Scholars 2011. Phuket, Thailand: Islamic Bank. (nd).Restricted Investment Account (RIA) through the Investment Account Platform (IAP) (2011) 4. Yustiardhi, A.F., Diniyya, A.A., Amirah, F., Faiz, A., Subri, N.S., Kurnia, Z.N.: Issues and challenges of the application of Mudarabah and Musharakah in Islamic bank financing products. J. Islam. Financ. 9(2), 26–41 (2020). (ISSN 2289–2109 e-ISSN 2289–2117) 5. Arsalan, M., Khan, S.D., Ur Rahman, A.: The obstacle factors of Musharakah and Mudharabah application in Pakistan. J. Akunt. dan Keuang. Indones. 17(2) (2020). https://doi.org/10. 21002/jaki.2020.11 6. Rosman, R., Abdul, R.A., R.,: The practice of IFSB guiding principles of risk management by Islamic banks: international evidence. J. Islam. Account. Bus. Res. 6(2), 150–172 (2015). https://doi.org/10.1108/JIABR-09-2012-0058 7. Hassan Al-Tamimi, H.A., Mohammed Al-Mazrooei, F.: Banks’ risk management: a comparison study of UAE national and foreign banks. J. Risk Financ. 8(4), 394–409 (2007). https:// doi.org/10.1108/15265940710777333 8. Damnjanovic, I., Reinschmidt, K.: Project risk management fundamentals. In: Data Analytics for Engineering and Construction Project Risk Management, Springer, Cham, pp. 23–41 (2020). https://doi.org/10.1007/978-3-030-14251-3_2 9. UK Standard, “A Risk Management Standard,” (2002) 10. Nath, S.: 7 Types of Financial Risks with Real-World Examples, Investing and personal Finance for milinnials, Beginnersbuck (2020) 11. Al-Saati, N.H., Hameed, A.R.A.: Gharar (risk) is permissible in classical Islamic law. J. King Abdulaziz Univ. Islam. Econ. 16(2), 3–19 (2003) 12. Al-Suwailem, Hedging in Islamic finance. Islamic Development Bank, Islamic Research and Training Institute (2006) 13. Al-Amine, M., Al-B. Muhammad, M.: Risk Management in Islamic Finance: Analysis of Derivative Instruments in Commodity Markets, First Edit. Al-Khin, Mostafa, Al-Bugha (2008)
372
Jumaizi et al.
14. Bacha, O.I.: Derivative instruments and islamic finance: some thoughts for a reconsideration. Int. J. Islam. Financ. Serv. 1(1), 1–33 (1999) 15. Kamali, “Pemikiran Hukum Muhammad Hasyim Kamali,” Mylawyers (2005) 16. Mohamad, S., Tabatabaei, A.: Islamic Hedging: Gambling or Risk Management? SSRN Electron. J. (2011). https://doi.org/10.2139/ssrn.1260110 17. Securities Commission, Resolutions of the Securities Commission Shariah Advisory Council, vol. Second Ed. (2007) 18. Mohamad Akram Laldin, M.: Risk management in Islamic finance. Harvard-LSE Workshop on Risk Management. Islamic Economics and the Perspective of Islamic Law and Ethics in the Current Financial Crisis. London School of Economics (2009) 19. Yaron, J.: What makes rural finance institutions successful? World Bank Res. Obs. 9(1), 49–70 (1994). https://doi.org/10.1093/wbro/9.1.49 20. Stiglitz, J.E., Weiss, A.: Credit rationing in markets with imperfect information. SSRN Electron. J. 71(3), 393–410 (1981). https://doi.org/10.2139/ssrn.1010965 21. Woolcock, M.J.V.: Learning from failures in microfinance. Am. J. Econ. Sociol. 58(1), 17–42 (1999). https://doi.org/10.1111/j.1536-7150.1999.tb03281.x 22. Arsyad, L.: Assessing factors affecting the repayment rate of microfinance institutions: a case study of village credit institutions of Gianyar, Bali. Gadjah Mada Int. J. Bus. 8(2), 247 (2006). https://doi.org/10.22146/gamaijb.5618 23. O’Toole, C., Slaymaker, R.: Repayment capacity, debt service ratios and mortgage default: an exploration in crisis and non-crisis periods. J. Bank. Financ. 133, 106271 (2021). https:// doi.org/10.1016/j.jbankfin.2021.106271 24. Florence, A., Waari, D.N.: Factors influencing loan repayment in micro-finance institutions in Kenya. IOSR J. Bus. Manag. 16(9), 66–72 (2014). https://doi.org/10.9790/487x-16936672 25. Besley, T., Coate, S.: Group lending, repayment incentives and social collateral. J. Dev. Econ. 46(1), 1–18 (1995). https://doi.org/10.1016/0304-3878(94)00045-E 26. Helfert, Technique of financial analysis, Seventh Ed. Homewood Illionis (1991) 27. Mulyadi, Balance Scorecard: Contemporary Management Tools for Multipliers of Company Financial Performance. Jakarta: Salemba Empat (2001) 28. Roziq, A., Sukarno, H.: The effect of Islamic financing schemes on risk and financing performance in Islamic banks in Indonesia. IQTISHODUNA J. Ekon. Islam, 10(1), 17 (2021). https://doi.org/10.36835/iqtishoduna.v10i1.729 29. Karim, A.A.: Bank Islam: Analisis Fiqih dan Keuangan. Jakarta: PT. Raja Grafindo Persada (2016) 30. ISRA, Islamic Financial System: Principles and Operations. Kuala Lumpur: ISRA (2010) 31. Hamid, O.H., Allaymoun, M.: E-Mudaraba suggested system for Islamic investments. J. Internet Bank. Commer. 24(1), 1–21 (2019). http://www.icommercecentral.com 32. Rahman, M.: Mudarabah and its applications in Islamic finance: an analysis. Asian J. Res. Bank. Financ. 8(6), 33 (2018). https://doi.org/10.5958/2249-7323.2018.00042.1 33. Ishak, M.S.I., Rahman, M.H.: Equity-based Islamic crowdfunding in Malaysia: a potential application for mudharabah. Qual. Res. Financ. Mark. 13(2), 183–198 (2021). https://doi.org/ 10.1108/QRFM-03-2020-0024 34. Ekinci, A.: The effect of credit and market risk on bank performance: evidence from Turkey. Int. J. Econ. Financ. Issues 6(2), 427–434 (2016) 35. Hair, J.F.: Multivariate data analysis: with readings. New York: Macmillan, Toronto: Maxwell Macmillan Canada, New York: Maxwell Macmillan International (1992) 36. Augusty, F.: Structural Equation Modelling dalam Penelitian Manajemen. Universitas Diponegoro, Semarang (2000)
The Consumption Value and Value Congruity: A Conceptual Development of Hasanah Value Congruity Ari Pranaditya1(B) , Ken Sudarti2 , and Hendar2 1 Faculty of Economics, Universitas Islam Sultan Agung (UNISSULA), Semarang, Indonesia
[email protected]
2 Lecturer of Management, Faculty of Economics, Universitas Islam Sultan Agung
(UNISSULA), Semarang, Indonesia {kensudarti,hendar}@unissula.ac.id
Abstract. This paper tries to propose a new concept that combines value congruity and religiosity called Hasanah Value Congruity (HVC). The concept describes as a unique capability of frontliner together as a marketing team, create a culture which congruence with the customer consumption value as a strategic asset, as customers prefer to use products and services that in line with their personal values. The values adhered in HVC is Islamic Values as the frontliner and the team’s culture capable to increase the level of one’s iman (ziadatul iman capability) and knowledge (the love of knowledge capability). HVC lies within the scope of internal marketing as enabling promises activity expected to increase marketing performance. Testing the validity of each dimension is needed, which can then be empirically tested with addition of service quality, internal communication and interpersonal relationship as variables. Keywords: The Theory of Consumption Value · Value congruity · Marketing culture · Hasanah value congruity
1 Introduction Internal marketing as the focus of this study based on a marketing perspective belief that satisfied and motivated employees are essential to satisfy external customers [1]. Internal marketing is an activity carried out by a company that provides excellent services [2] to internal customers, namely employees, which is an activity of “enabling promises” which is one of the three types of marketing in “The Service Marketing Triangle” [3]. “The Service Marketing Triangle” describes the interrelationship of three important elements of service organizations, namely: company, employee and customer which makes the process of marketing services divided, namely internal marketing, external marketing and interactive marketing. The interaction that the company perform to external customers, namely customers, is external marketing, in the form of “giving the promises” activities that offer quality products at competitive prices, attractive promotions and ease of obtaining products. While interactive marketing is an activity carried out by employees © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 373–381, 2022. https://doi.org/10.1007/978-3-031-08812-4_36
374
A. Pranaditya et al.
to customers, namely “keeping promises” activities by providing total quality service to tangible elements, empathy, responsiveness, reliability and assurance (Fig. 1).
Fig. 1. The service marketing triangle [4]
Frontliners as employees in the vanguard who are directly related to customers have a very important role in creating service value in order to satisfy customers [5] because when its strategic, value can be a source of competitive advantage, whether temporary or even sustainable [6]. Values play a central role in the cognitive structure of customers, therefore values that have good congruity with customers will certainly be a superior marketing strategic asset because it explains the similarity of values between customers and organizations where customers prefer to use products and services that represent their personal values [7]. However, Bao [8] stated that the effect of value congruence on performance is complicated. Schleicher, Watt, & Greguras [9] saw the relevance of the value dimension and the context of research having a significant impact on results, which is associated with sales performance [10] but nevertheless the effect of value congruence on performance, whether in role or contextual is less often [11], while Lee, S. & Jeong, M. [7] stated that it is necessary to investigate different types of congruity. The existence of value formed by frontliners creates a marketing culture that has been proven empirically have a considerable impact on organizational performance and profitability [12, 13] although some researchers say the relationship is short-term [14], highly problematic [15] and damaging economic performance [16]. Each dimension of value is a valuable information for marketers for their strategic initiatives [16] because it has an influence on purchasing motivation [18]. This view is supported by Yeonsoo et al. [19], who assert that the Theory of Consumption Values (TCV) allows for a deeper explanation since it examines fundamental reasons in the consumer decision-making process. Sheth et al. [20] identified five values or dimensions of value that influence consumer choices, including functional value, social value, epistemic value, emotional value and conditional value. TCV can be a strong base for marketers to build strategies for their products and services can be well accepted by consumers. The values consider to be the driver of consumer consumption in TCV however, is only measured by worldly measurement. It does not extend to the afterlife, that is
The Consumption Value and Value Congruity
375
the basic needs of Islamic values injection in the theory. As moslem’s consumption motivation is not only to fulfill their worldly needs, but also as a religious procession or ibadah. Allah Subhanahu wa Ta’ala explains in QS. Az Zariyat 56: “And I (Allah) did not create jinn and mankind except worship Me.” 5 values of TCV and value congruity does not explain the consideration of the value of worship. The presence of hasanah in value congruity becomes a ukhrowi dimension that perfects the value that can be felt by customers. Marketing culture that integrates hasanah with value congruity is expected to create hasanah value congruity which has the potential to improve marketing performance based on congruity theory [21] as customers prefer to use products and services that represent their personal values [7].
2 Literature Review 2.1 Congruity Theory Congruity theory emerge when a person is more likely to have a positive attitude towards an object when a person perceives an object or phenomenon consistent with what he is holding [21]. Congruity Theory developed into a variety of concepts, one of which is value congruity, which is an interactive marketing perspective because it explains the similarity of values between customers and organizations where customers prefer to use products and services that represent their personal values [7, 22]. Value congruity first emerge as the perceived similarities between values held by individuals and organizations [23] which explain the relationship within an organization. Value congruity is supported by the similarity-attraction theory [24]. Similarityattraction theory people prefer to maintain relationships with others similar to them. It’s Zhang and Bloemer [25], the one connects value congruity in marketing between sellers and consumers. However, congruity theory still grounded its concept on the relationship between individuals or groups with other individuals or groups and the relationship of individuals or groups with products or services in a particular context. None of them include the dimensions of the afterlife or the similarity of values nuanced to worship Allah. 2.2 The Theory of Consumption Value TCV was introduced by Sheth et al. [20] in an article titled “Why we buy what we buy: A theory of consumption values” published in the Journal of Business Research. Sheth et al. [20] identified five values (values or dimensions of value) that influence consumer choice. Together functional value, social value, epistemic value, emotional value and conditional value represent TCV. Functional value comes from the intrinsic capacity of a product for functional, utilitarian or physical performance, ie. its ability to fulfill the function of the product is created. Functional value is based on the assumption that customers are driven by alternatives with the best performance on their physical and utilitarian attributes or alternatives that have most of those attributes [27]. Its influence on customer choice is focused on traditional economic utility theory [26].
376
A. Pranaditya et al.
Social value is defined as the perceived and acquired utility of a product’s relationship with a particular demographic, cultural or social group [28]. Social values are values that allow individuals to develop close relationships with a particular community or group. It is often a specific reference group or community of which the individual wants to be a part of it [26]. Emotional value is related to the aspect of extrinsic consumption in terms of the product’s ability to evoke affective feelings or states. The more positive emotions that can come from a product or service, the more likely it is that customers will continue to use it [26]. Epistemic values relate to the perceived utility or acquired from an alternative capacity to arouse curiosity, provide novelty and/or satisfy the desire for special consumption knowledge or experience [29]. Then the conditional value comes from the product’s ability to provide functional or social value while in a particular situation or context and consequently depends on the specific circumstances the customer faces when making conditional values choices to promote or withhold a decision [29]. TCV in its development has experience addition of dimensions, including religious influence [30], monetary value [31], fashion value [32], expositional value [33] and religious value [34]. The addition of religious value based on Faith as a key Islamic value [34] is the latest dimension in the development of TCV. However, the addition of these dimensions comes from qualitative research with 21 respondents. This opens up opportunities for the integration of Islamic values in the development of TCV which is the focus of this research. 2.3 Hasanah Value Congruity In language/etymology, the Arabic word hasanah means good, beautiful, whose form of masdar hasanatan means goodness, grace, kindness, good deeds, privilege, virtue. Viewed in the terms of terminology, the meaning of the word hasanah is an act of virtue (charity) that will symbolically be placed on the scales to decide one’s salvation on the day of the afterlife trial. The meaning of hasanah is close to the word mahmudah meaning is commendable whose meaning is good, noble and so on. Mahmudah can be juxtaposed with the word muamalah to be Muamalah Mahmudah which means commendable muamalah. Muamalah itself is a law of syara related to world affairs and human life, such as buying and selling, trading and so on. Muamalah is any regulation that contains all matters relating to world affairs and every matter of matter concerning the material, marriage, and talaq established by following the general and detailed basics to be used as instructions for human beings in exchanging benefits. Two sources of the law on muamalah in the teaching of Islam are the Qur’an and the Sunnah or Hadith, which means the words, actions, and approval of the Prophet Muhammad. According to Abdullah of Ibn Abbas, there are seven indicator for hasanah in the world, that is qolbun syaakirun means his heart is always grateful., Al azwaju shalihah, have godly wives, al auladun abrar, have children who are devoted to parents, albiatu sholihah means the neighborhood, friends and good neighbors, arzaaqun mabruukah, that is, having sufficient halal sustenance, hubbul ilmi, ulamai wa majaalisihi, means loving the knowledge of the Quran, As-Sunnah, scholars and its functions (at tafaqquh fiddiin) and ‘umrun mabrukun, his age is a blessing in obedience.
The Consumption Value and Value Congruity
377
Related to the research context of higher education, albiatu sholihah which means the neighborhood, friends and good neighbors and hubbul ilmi, ulamai wa majaalisihi which means loving the knowledge of the Quran, As-Sunnah, scholars and its functions (at tafaqquh fiddiin) will be the appropriate indicator of Hasanah Value Congruity which defined as the frontliner’s ability to strengthen Ziadatul Iman Capability and the love of knowledge capability through improving customer understanding and confidence in the similarity of hasanah values. Hasanah Value Congruity has the potential to improve marketing performance. Ziadatul Iman Capability is the ability of frontliners in strengthening faith in their environment through the values of hasanah. Albiatu sholihah formed from colleagues and leaders who have congruity of social values, emotional values and islamic values hasanah. The congruity of social value, namely the importance to always exist in the environment of people who are with similar aqidah and sholeh as the Prophet shalallahu alaihi wa salam explained that “one’s follows din (religion; character; Morals) of one’s close friend. Therefore, one’s should pay attention to who he makes a close friend” HR. Abu Dâwud, no.4833; Tirmidhi, no.2378. narrated by Shaykh al-Albani in genealogy of ash-Shahîhah no. 927. The congruity of emotional value with customers is characterized by the desire to always be close to Allah subhanahu wa ta’ala, which is comfortable in the middle of friends who are sholeh. Allah subhanahu wa ta’ala explains who is among the people who are sholeh in QS At-Taubah 112, namely “ They are those who repent, who worship, who praise, who visit, who ruku´, who bow down, who command to do ma´ruf and prevent the act of munkar and who keep the laws of God. Ziadatul iman capability has the spirit of fastabiqul khoirot which is competing in goodness and reminding each other in kindness and patience. The next dimension, The Love of Knowledge Capability is the frontliner’s ability to strengthen the love of religious science formed from the congruence and integration of functional values, epistemic value and conditional value with Islamic values. In the context of higher education purpose as stated in UU No. 12 of 2012, namely in article 5 mentioned 4 (four) goals of higher education where the first is “The development of the potential of students to become human beings who believe and fear God Almighty and be noble, healthy, knowledgeable, capable, creative, independent, skilled, competent, and cultured for the benefit of the nation.” The congruence of functional values and integration of Islamic values with the Tridharma becomes an embodiment of the purpose of higher education in accordance with Law No. 12 of 2012 above. The congruence and integration of Islamic values with the Tridharma is expected to arouse tafaqquh fid diin, the spirit of understanding religious science which becomes an epistemic value that is to arouse curiosity, give new things or satisfy the desire for knowledge according to the order of the Prophet shalallahu alaihi wa salam “be you a knowledgeable person, or a person who studies, or a person who wants to listen to knowledge, or a person who likes knowledge” (Hr. Baihaqi). In addition to epistemic value, the integration of Islamic values with the Tridharma is also congruence with conditional value, providing functional or social value in certain situations or contexts, namely knowledge or ‘ilm in arabic is the only path to hasanah, the happiness of the afterlife.
378
A. Pranaditya et al.
2.4 Marketing Performance Performance is a multidimensional concept that does not only show the measurement of results but also the process in achieving results and conditions that enable the achievement of results [35]. Marketing performance in the context of higher education is the increasing number of new students who were previously marked by the increasing number of applicants. Patronage intention is also an indicator of whether customers will continue to use a service or switch to another service [36] which is the intention of students to resume their next level of education, either master or doctoral degree in the same institution. Another marketing performance indicator is word of mouth (WOM). According to Kumar, Petersen, and Leone [37], valuable customers are customers who through WOM can bring in the most profitable customers. What customers feel and what they tell others about the company/brand can affect revenue and profit [38]. Any activity involving WOM will basically be related to satisfaction and profit research. Positive WOM produces more results than negative ones. Nevertheless both have significant impact on potential customer behavior towards the company and purchase intentions [39]. Implications for higher education, excellent WOM not only has an impact on improving financial returns but also improves its image. Individuals who are satisfied and believe in higher education institutions will be willing to spread positive WOM and can easily provide recommendations to others, especially those closest to them. Through an in-depth study, the study finally succeeded in synthesizing Islamic values, value congruity and marketing culture. Hasanah Value Congruity proposition can be defined as follows:
Fig. 2. The integration of Congruity Theory and The Theory of Consumption Value with Hasanah as Islamic Value; developed by author for the study.
The Consumption Value and Value Congruity
379
Hasanah Value Congruity is the frontliner’s ability to strengthen Ziadatul Iman Capability and the love of knowledge capability through improving customer understanding and confidence in the similarity of hasanah values. Hasanah Value Congruity has the potential to improve marketing performance (Fig. 2).
3 Conclusion Hasanah Value Congruity (HVC) is an Islamic character that is thought to be able to increase marketing performance. Someone who has HVC will be able to strengthening faith in their environment through the values of hasanah, maintaining the team to always surround themselves in the environment of sholeh people with similar aqidah and ability to strengthen the love of religious science by arousing tafaqquh fid diin, the spirit of understanding religious science. In the context of higher education culture, shall each individual in a team has HVC, then the will helps everyone to understand and ‘feel’ the function of marketing”. As the feel of marketing aroused, hopefully it will attract potential customers with similar value and affect its marketing performance.
4 Future Research For future research agenda, it is better to involve validating the measurement of each variable and testing Hasanah Value Congruity concept at the empirical level. The type of this study is explanatory research, a research method that intends to explain the position of the variables studied and the influence between variables with each other. Antecedent variables offered are service quality, internal communication and interpersonal relationship referring to Webster’s [40] opinion as a widely accepted and implemented instrument for measuring marketing culture [41]. These variables are expected to be able to increase marketing performance. This model is very precisely tested in Islamic higher education institution in Central Java area of Indonesia, which expected to have already integrate Islamic values in its curriculum and management. Random samples collected from academic member of the Islamic higher education institutions responsible for marketing activities. The simple random sampling technique chosen due to the members of the population are considered homogeneous, without seeing and paying attention to the similarities or strata that exist in the population. Data collection is carried out by the dissemination of questionnaires, direct data collection carried out by submitting a list of statements to respondents. The questionnaire is submitted directly through online media in the form of google form. Statements include open and closed, an open statement is a statement that gives freedom to respondents to answer statements in accordance with their way of thinking. While a closed statement is a statement where the answers have been limited by researchers so that it is possible for respondents to answer at length according to their way of thinking. To process the data in this study used the Structural Equation Modelling (SEM) of the AMOS 20.0 software package. SEM is a set of statistical techniques that allow simultaneous testing of a series of relatively complex relationships. SEM is a combination of separate statistical methods, namely factor analysis and simultaneous equation modeling.
380
A. Pranaditya et al.
References 1. Lings, I.N., Greenley, G.E.: The impact of internal and external market orientations on firm performance. J. Strateg. Mark. 17(1), 41–53 (2009) 2. Shiu, Y.M., Yu, T.W.: Internal marketing, organisational culture, job satisfaction, and organisational performance in non-life insurance. Serv. Ind. J. 30(6), 793–809 (2010) 3. Gronroos, C.: Service Management and Marketing: A Customer Relationship Management Approach. John Wiley and Sons, Ltd., Hoboken (2000) 4. Zeithaml, V.A., Bitner, M.J., Gremler, D.D.: Services Marketing Strategy, Wiley International Encyclopedia of Marketing, December 2017 5. Heskett, J.L., Jones, T.O., Loveman, G.W., et al.: Putting the service-profit chain to work. Harv. Bus. Rev. 72, 164–170 (1994) 6. Barney, J.B.: Firm Resources and Sustained Competitive Advantage. J. Manage. 17, 99–120 (1991) 7. Lee, S., Jeong, M.: Enhancing online brand experiences: an application of congruity theory. Int. J. Hosp. Manag. 40, 49–58 (2014) 8. Bao, Y., Dolan, S.L., Tzafrir, S.S.: Value Congruence in Organizations: Literature Review, Theoretical Perspectives, and Future Directions (2012). ESADE Business School Research Paper 239 (2012) 9. Schleicher, D.J., Watt, J.D., Greguras, G.J.: Reexamining the job satisfaction-performance relationship: the complexity of attitudes. J. Appl. Psychol. 89(1), 165–177 (2004) 10. Weeks, W.A., Chonko, L.B., Kahle, L.R.: Performance congruence and value congruence impact on sales force annual sales. JAMS 17, 345–351 (1989) 11. Arthur, W., Jr., Bell, S.T., Villado, A.J., Doverspike, D.: The use of person-organization fit in employment decision making: an assessment of its criterion-related validity. J. Appl. Psychol. 91(4), 786–801 (2006) 12. Biloslavo, R., Trnavcevic, A.: Marketing culture as a tool for development of a sustainable organization. In: International Conference on Business Management and Economics (ICBME 2011), Yasar University, Izmir, Cesme, 6–8 October 2011 13. Postruznik, N., Moretti, M.: Innovation and communication as dimensions of the marketing culture: their influence on financial performance in Slovenia’s insurance and construction industries. Our Econ. 58(1/2), 35–47 (2012) 14. Kotter, J.P., Heskett, J.L.: Corporate Culture and Performance. Free Press, New York (1992) 15. Alvesson, M., Berg, P.O.: Corporate Culture and Organizational Symbolism: An Overview. Walter de Gruyter, Berlin (1992) 16. Kotter, J.P., Schlesinger, L.A.: Choosing Strategies for Change. Harvard Bus. Rev. 86(7–8) (2008) 17. Whalen, K.A.: Three circular business models that extend product value and their contribution to resource efficiency. J. Clean. Prod. 226, 1128–1137 (2019) 18. Ramkissoon, H., Nunkoo, R., Gursoy, G.: How consumption values affect destination image formation. In: Woodside, A.G., Megehee, C.M., Ogle, A. (eds.) Perspectives on CrossCultural, Ethnographic, Brand Image, Storytelling, Unconscious Needs, and Hospitality Guest Research (Advances in Culture, Tourism and Hospitality Research Volume 3), pp. 143–168. Emerald Group Publishing Limited, Bingley (2009) 19. Yeonsoo, L., Jinwoo, K., Inseong, L., Hoyong, K.: A cross-cultural study on the value structure of mobile internet usages: comparison between Korea and Japan. J. Electron. Commer. Res. 3(4), 227–239 (2002) 20. Sheth, J.N., Newman, B.I., Gross, B.L.: Why we buy what we buy: a theory of consumption values. J. Bus. Res. 22(2), 159–170 (1991)
The Consumption Value and Value Congruity
381
21. Osgood, C., Tannenbaum, P.: The principle of congruity in the prediction of attitude change. Psychol. Rev. 62, 42–55 (1955) 22. Wang, J.J., Zhang, C.: The impact of value congruence on marketing channel relationship. Ind. Mark. Manage. 62, 118–127 (2017) 23. Chatman, J.A.: Matching people and organizations: selection and socialization in public accounting firms. In: Academy of Management Proceedings. Academy of Management, pp. 199–203 (1989) 24. Byrne, D.: The attraction paradigm. Academic Press, New York (1971) 25. Zhang, J., Bloemer, J.M.: The impact of value congruence on consumer-service brand relationship. J. Serv. Res. 11(2), 161–178 (2008) 26. Goh, T.T., Suki, N.M., Fam, K.: Exploring a consumption value model for Islamic mobile banking adoption. J. Islam. Mark. 5(3), 344–365 (2014) 27. Tapachai, N., Waryszak, R.: An examination of the role of beneficial image in tourist destination selection. J. Travel Res. 39(1), 37–44 (2000) 28. Phau, I., Quintal, V., Shanka, T.: Examining a consumption values theory approach of young tourists toward destination choice intentions. Int. J. Culture, Tour. Hospital. Res. 8(2), 125–139 (2014). https://doi.org/10.1108/ijcthr-12-2012-0090 29. Hung, C., Hsieh, C.: Searching the fit pattern between cultural dimensions and consumption values of mobile commerce in Taiwan. Asia Pac. Manag. Rev. 15(2), 147–165 (2010) 30. Delener, N.: Religious contrasts in consumer decision behaviour patterns: their dimensions and marketing implications. Eur. J. Mark. 28(5), 36–53 (1994) 31. Park, B.-W., Lee, K.C.: An empirical analysis of online gamers’ perceptions of game items: modified theory of consumption values approach. Cyberpsychol. Behav. Soc. Netw. 14(7–8), 453–459 (2011) 32. Yang, H.-L., Lin, R.-X.: Determinants of the intention to continue use of SoLoMo services: Consumption values and the moderating effects of overloads. Comput. Hum. Behav. 73, 583–595 (2017) 33. Hedman, J., Bødker, M., Gimpel, G., Damsgaard, J.: Translating evolving technology use into user stories: Technology life narratives of consumer technology use. Inf. Syst. J. 29(6), 1178–1200 (2019) 34. Rodrigo, P., Turnbull, S.: Halal holidays: How is value perceived by Muslim tourists? Int. J. Tour. Res. 21(5), 675–692 (2019) 35. da Gama, A.P.: An expanded model of marketing performance. Mark. Intell. Plan. 29(7), 643–661 (2011) 36. Zeithaml, V., Berry, L., Parasuraman, A.: The behavioral consequences of service quality. J. Mark. 60, 31–46 (1996) 37. Kumar, V., Petersen, J.A., Leone, R.P.: How valuable is word of mouth? Harvard Bus. Rev. 85(10), 139–146+166 (2007) 38. Reichheld, F.F.: The one number you need to grow. Harv. Bus. Rev. 81, 46–54 (2003) 39. Söderlund, M., Rosengren, S.: Receiving word-of-mouth from the service customer: an emotion-based effectiveness assessment. J. Retail. Consum. Serv. 14, 123–136 (2007) 40. Webster, C.: Toward the measurement of the marketing culture of a service firm. J. Bus. Res. 21(4), 345–362 (1990) 41. Akroush, M.N., Al-Mohammad, S.M., Odetallah, A.L.: A multidimensional model of marketing culture and performance: a different approach to the use of Webster’s marketing culture measurement scale. Int. J. Contemp. Hosp. Manag. 27(7), 1442–1478 (2015)
Resilience of Companies Listed in Jakarta Islamic Index (JII) During the Pandemic COVID-19 Mutoharoh(B) and Naila Najihah Undergraduate Program of Accounting, Universitas Islam Sultan Agung, Semarang, Indonesia [email protected]
Abstract. This study aims to test the impact of pandemic outbreak towards the financial performance of companies listed in Jakarta Islamic Index (JII). In addition, the profits of these companies will also be compared to test the difference between the period before the pandemic and the period during the pandemic. The observation period was taken from 2017–2020 to represent the performance before and during pandemic. There are 24 data were selected to be observed and compared. The test relies on trend analysis and paired difference test to test whether there is a significant difference between the profits of companies listed on JII during pre-pandemic and during the pandemic period. The results show that there is no significant difference between the two conditions that occurs even though the average tendency of the company’s OPM and ROE is decreasing. Keywords: JII · Performance · Pandemic
1 Introduction The COVID-19 outbreak that has spread since the beginning of 2020 has had a huge impact on various aspects of life in the world, including the economic sector [1, 2]. As with the effects of the pandemic, a number of large retail companies in Indonesia have suffered losses and may not even be able to survive until the second year, 2021. Among these companies, the site industri.kontan.co.id, (31/5/2021) announced the closure of all Giant hypermarket outlets. In Indonesia as of July 2021 from PT Hero Supermarket Tbk or Hero Group. A similar thing was reported in Kompas.com (27/4/2021) that PT Matahari Department Store Tbk (LPFF) will close 13 Matahari outlets in various regions, due to operational costs which is a financial problem in general. Based on a survey conducted by the Ministry of Labor in 2020, it showed that around 88% of companies were affected by the pandemic which resulted in losses to the company’s operations. Therefore, research on the company’s profit performance is very necessary to be analyzed to see how bad the impact of the pandemic is on companies. This study aims to examine differences in the financial performance of companies represented by OPM and ROE at selected companies that listed in JII during the pre-pandemic and during the pandemic. The research period was taken in 2017–2019 as a reflection of © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 L. Barolli (Ed.): CISIS 2022, LNNS 497, pp. 382–392, 2022. https://doi.org/10.1007/978-3-031-08812-4_37
Resilience of Companies Listed in Jakarta Islamic Index (JII)
383
performance before the pandemic compared to 2020 which was the year the COVID-19 virus was increasingly spreading. Previous research by Ramelli and Wagner [3], found that stock returns in the food and staple retail sector were less affected than other sectors in the early days of the pandemic. In a similar industrial sector, Höhler and Lansink [4] analyze more thoroughly the companies in the food supply chain. In his study, it was found that companies with profitable stock returns produced higher cumulative returns during the pandemic. Research by Rababah et al. [5] revealed that small and medium-sized companies registered in China were the sectors worst affected by COVID-19 because they experienced a sharper decline in financial performance compared to other industries. The Vietnamese stock market has also not been spared the impact of COVID-19. This condition had a negative effect in the period before the lockdown and vice versa during the COVID-19 lockdown had a positive effect on stock performance in all markets and various business sectors in Vietnam [6]. In Indonesia, various studies have been carried out regarding the effects of the pandemic. Existing research relies on companies without specifying a specific level with the aim of analyzing which sectors are most affected [7]. The tendency of other research is to see the effect of the pandemic on businesses at a very vulnerable level such as MSMEs [1, 8]. This study tries to take from a different point of view by relying on the type of company with undoubted performance. Companies listed in the JII index must go through several stages of screening that are more stringent than other stock indexes. This becomes an interesting focus in order to test the resilience of these best companies from the threat of a pandemic. The COVID-19 pandemic is a unique phenomenon because of its very fast spread and ability to disrupt life [3]. In fact, some countries still have to struggle a lot to overcome this epidemic. No one knows for sure when and how things will return to normal. Many researchers have used phenomenon studies to measure the impact of the phenomenon or to see how the market reacts to a phenomenon [4, 9]. Related to the design of this study, phenomena studies are used to separate the effects of the pandemic from the effects of the general market. The test is carried out on the pattern of changes in company performance represented by operating profit margins (OPM) and return on equity (ROE) of companies with the best liquidity listed in the Jakarta Islamic Index (JII) to test the average ratio in the three years before the pandemic period (2017–2019) and during the period during the pandemic (2020). This research is very important to provide motivation in economic activity. As existing findings show the adverse effects of the pandemic in various sectors, in fact there is still hope that investment activity, especially in the Sharia stock index, is possible to provide more stable results. Investors and managers use this to assess company performance and develop strategic plans in dealing with crises caused by extreme phenomena. The selection of a sample from one of the Islamic stock indexes in Indonesia is also expected to motivate investors and potential investors to increase their investment in Islamic financial products. The implications of this research can be seen through the increase in company performance and public confidence in the resilience of companies in the Sharia stock index, especially in the JII list, so that investment in companies on the Sharia index list is also increasing. This research is structured within six section
384
Mutoharoh and N. Najihah
that consist of 1) Background, 2) Literature review, 3) Research methods, 4) Results, 5) Discussions, and 6) Conclusions.
2 Literature Review 2.1 COVID-19 Pandemic Impact on the Company’s Financial Performance The COVID-19 pandemic has become a new phenomenon that has forced a third of the world’s population to take lockdown measures [10]. Baig et al. [11] assesses the impact of the lockdown on the US stock market due to a decline in market stability and liquidity. Other things caused by this pandemic outbreak such as an increase in confirmed cases and death rates [12–15] have also had a negative impact on market performance of international stock. In addition, the impact of the COVID-19 pandemic around the world is severe on the financial side due to the cessation of economic activities due to the social restrictions imposed [16]. Bagnera, Steinberg and Edition [17] shows that nearly 90% of airline industry employees have been laid off. In addition, hotel operations are only around 20% of their normal capacity. The tourism and travel industry also suffered financial losses in 2020. All social agendas such as sports, culture, music concerts, had to be postponed indefinitely. The Indonesian government has analyzed that the economic crisis due to the COVID19 pandemic will have an impact on declining profits and financial performance in various types of businesses. The economic crisis can have an impact on the decline in sales of products produced by the company [18]. The decrease in total sales will certainly have an impact on the company’s financial performance. Some companies have even been liquidated due to financial difficulties [19]. The pharmaceutical sub-sector only experienced a slight decline in the average value of the liquidity ratio. The same thing also happened to the cigarette sub-sector, which did not significantly reduce the average current ratio, considering that cigarettes had become addictive. Their sales were not affected by the crisis. Meanwhile, the financial performance of the consumer goods industry sector was affected by the monetary crisis when viewed from the value of the debt to equity ratio (DER), which showed a decline in the ability to pay debts. The work from home (WFH) policy during the COVID-19 pandemic resulted in a decline in a number of sectors, but also resulted in an increase in other sectors. Tourism, transportation and aviation among others experienced a decline, but sales of household appliances included in the manufacturing sector continued to increase from February to April 2020 (Ministry of Finance, 2020). Sales of food and beverage products also continued to increase during the COVID-19 pandemic. 2.2 Resource-Based Theory This theory explains that the company’s performance will be optimal if the company has a competitive advantage that is difficult to imitate and is strongly attached to its characteristics. According to [20], to help the financial performance of mining companies in China from various risky economic situations, it is necessary to create new competitive advantages for long-term development. Competitive advantage is obtained by utilizing,
Resilience of Companies Listed in Jakarta Islamic Index (JII)
385
managing, and controlling owned resources, such as organizational processes and strategies companies in the face of various conditions, including when facing an economic crisis. In this study, the selection of superior companies is believed to have abundant resources that can be used as weapons in adapting to the worst conditions such as the COVID-19 outbreak. Thus, it is possible that these companies will not be affected and shaken like the industry in general and still have good performance during the pandemic. 2.3 Trends in the Performance of Companies Listed on the Sharia Stock Index The Islamic stock index measures the performance of a certain group of securities based on sharia principles [21]. The Sharia index filters the criteria for listed companies using two types of criteria, namely business activities and financial ratios [22]. With regard to business activities, the main business involvement is in Riba-based financial services; gambling; manufacture or sale of non-halal products; conventional insurance; entertainment activities that are not allowed according to sharia; manufacture or sale of tobacco-based products or related products; stock brokering or stock trading in approved non-sharia securities is not allowed in Islamic indexes [23]. Islamic indices also use financial limits to screen stocks. The Sharia Compliance Index does not allow investment in companies that earn significant income from interest or companies that have leverage with a certain ratio. Different types of Islamic indices determine specific criteria for different financial ratios in selecting stocks. The Dow Jones Islamic Market Index (DJIM) requires an upper limit of 33% for the debt ratio, 45% for receivables to total assets and 5% for interest income to income. In Indonesia, the Jakarta Islamic Index (JII) uses 45%, 55%, and 10% respectively for the ratio of debt, receivables to total assets and interest income from income [24]. Hanafi and Hanafi [25] found that JII provides better performance compared to conventional stock indices. Powell and DeLong [21] also find that the performance of the DJIM index has outperformed compared to comparable conventional indices. Albaity, Md Noman and Mallek [26] found that stock market returns between the three Islamic stock market indices, the Kuala Lumpur Syariah Index (Malaysia), the Dow Jones Islamic Market Index (US), and the Financial Times Stock Exchange Global Islamic Index (UK) were not significantly different from the stock index in general. Besides that, Habib et al. (2014) found that the Islamic index in India has outperformed the conventional index based on the measurement of returns and risk adjustment. The company rules that are included in JII in terms of financial terms include total interest-based debt compared to total equity 10% of total income. By looking at the criteria and performance produced by companies listed in JII, there is confidence that these companies have good resilience. Strict selection until finally the 30 best companies were selected that met the criteria leading to the perception that they had good performance. It is hoped that the company’s strong basic performance will not experience a significant decline due to extreme events, including those caused by the COVID-19 pandemic. Referring to the framework above, the hypothesis in this study is concluded that: Ho: There is no significant difference between the profits of companies registered with JII in the pre-pandemic period and during the pandemic.
386
Mutoharoh and N. Najihah
Ha: There is a significant difference between the profits of companies registered with JII during the pre-pandemic period and during the pandemic period.
3 Method This study aims to analyze performance of companies listed in JII during the prepandemic period and during the pandemic. To achieve this goal, several stages of research will be carried out. The first step is to do a sampling mapping by relying on a purposive sampling model from the research population which is all shares listed in JII. The research sample criteria must meet the requirements to be registered with JII during the research year. Data will be eliminated if new companies enter or leave in the middle of the research period. Secondary data in this study comes from financial reports published on the official website https://www.idx.co.id/. In addition, additional references to strengthen the theories, statements, and data in this study use scientific articles and academic journals, relevant textbooks, and other sources. The next stage is data processing. Financial performance in this study uses OPM and ROE ratios to represent the company’s resilience during a pandemic with its ability to generate profits during the assessment period. The value of OPM is calculated based on the value of operating profit (loss) divided by the company’s total revenue while ROE is obtained from the total return including Non-controlling Interest with total equity. The company’s financial performance, which is indicated by the two ratios, means that the higher the ratio, the better the performance, whereas on the contrary, when the OPM and ROE ratios decrease, it is a signal that performance is experiencing problems. After the ratio data is obtained and grouped, the next step is to test the data. This research is a quantitative research by utilizing trend analysis and Paired t-Test different test which was previously preceded by descriptive statistical test, KS normality test and Runs Test autocorrelation test with SPSS tool (Table 1). Table 1. K-S Test Unstandardized residual N Normal parameters
Mean Standard Deviation
OPM
ROE
24
24
0.000
0.000
6.154
5.124
Kolmogorov-Smirnov Z
0.631
0.412
Asymp. Sig. (2-tailed)
0.821
0.996
Resilience of Companies Listed in Jakarta Islamic Index (JII)
387
4 Findings 4.1 Operating Profit Management (OPM) 4.1.1 Normality Test The normality test in this study used the KS tool. The KS Test shows the OPM significance value is 0.821, which means that the data is normally distributed because the significance is >0.05. on the ROE variable, KS is worth 0.412 so that the ROE data is also normally distributed. Parametric test requires normal data distribution to be processed using paired t-test. As the results of the normality test have confirmed the main type of test that will be used in this study, the test can be continued to the next assumption (Table 2). 4.1.2 Autocorrelation Test Table 2. Runs Test Unstandardized residual (OPM)
(ROE)
Test value (median)
−1.62751
−0.198
Total case
24
24
Number of runs
5
7
Z
−0.908
0.000
Asymp. Sig. (2-tailed)
0.364
1,000
Autocorrelation should not occur in this data. This test is conducted to assess whether there is a correlation between the nuisance error in period t and the error in period t-1. The significance value of the Run Test on the OPM is 0.364 and the ROE is 1000, meaning that there is no autocorrelation because the Asymp. Sig (2-tailed) value is >0.05. Thus, the test can be continued on the Paired Test Paired t-test (Table 3). 4.1.3 Paired T-Test Table 3. Paired samples statistics
Pair 1
Mean
Standard deviation
Standard error mean
(OPM) Before
17.101
8,442
2.437
(OPM) During
15,468
6.573
1,897 (continued)
388
Mutoharoh and N. Najihah Table 3. (continued)
Pair 2
Mean
Standard deviation
Standard error mean
(ROE) Before
25,940
34,839
10,057
(ROE) During
22,219
37,709
10,886
Based on the table above, the average OPM value before the pandemic was 17.10 while during the pandemic it was 15.46. This means that descriptively there are differences in the average OPM before the pandemic before and during the pandemic. The average OPM before the pandemic tends to be higher than during the pandemic. The statistical results for ROE show similar conditions, namely the average ROE in the period before the pandemic was 25,940 while during the pandemic it was 22,219 so that a downward trend also occurred in ROE (Table 4). Table 4. Paired sample correlations Correlation
Sig.
Pair 1
(OPM) Before & During
0.685
0.014
Pair 2
(ROE) Before and During
0.989
0.000
Results show the correlation coefficient value is 0.685 (68%) with a significance value of 0.014 (0.05 with a t value of 0.912 indicating a positive and insignificant relationship between variables. Then, Sig. (2-tailed) which is the probability level of the test results is 0.381 or greater than 0.05. This indicates that H0 is accepted, i.e. there is no difference in OPM between before the pandemic and during the pandemic. The Mean value of 1.633 is Positive, which means that although there is no significant difference in OPM in the two tested conditions, there is still a tendency to decrease the ratio in during the pandemic as much as 1.633. The same results were also found in ROE. That is, at a significance level of 0.057 and a t-value of 2.124, it indicates that there is an insignificant positive relationship on the
Resilience of Companies Listed in Jakarta Islamic Index (JII)
389
ROE variable. The probability level of 0.57 indicates that H0 is accepted or there is no difference in ROE before the pandemic and during the pandemic but the means of 3,721 has a positive value indicating the tendency of the ROE variable during the pandemic to be lower than before.
5 Discussion The resilience of companies listed on JII has been proven based on the test results in this study. When various parties felt a tremendous impact on their condition, both in the profit and non-profit sectors, it turned out that JII only experienced a significant difference. This is in accordance with the resource-based theory where companies registered with JII with strict selection of liquidity have sufficient resources that are able to support the smooth running of business activities during the pandemic. JII’s constituents consist of 30 shares originating from companies that were screened with several criteria. The directives of the National Sharia Council (DSN) and Financial Services Authority (Bapepam-LK) Regulation Number IX.A.13 concerning the Issuance of Sharia Securities stipulate that entities that are included in the JII selection may not conduct business that is not in accordance with Islamic sharia, such as: 1. Gambling and gaming businesses that are classified as gambling or prohibited trade 2. Financial services with the concept of usury, buying and selling risks that contain gharar and maysir 3. Production, distribution, and trade and or supply of goods and or services that are haram because of the substance, not because the substance is based on the DSN-MUI or goods and or services that damage morals and are harmful 4. Invest in a company when the investee company’s debt ratio is dominantly obtained from usurious financial institutions on its capital. In addition, it is also determined that the financial ratios of the companies selected in JII are: 1. Total Interest-based Debt compared to total equity