AI and machine learning for the analysis of data flow characteristics in industrial network communication security. (29th July 2021)

• Record Type: Journal Article
• Authors: Xu, Zhi; Lu, Jun; Wang, Xin; Zhang, JiaHai; Alazab, Mamoun; García-Díaz, Vicente
• Is Part Of: International journal of ad hoc and ubiquitous computing. Volume 37:Number 3(2021)
• Page Start: 125
• Page End: 136
• Publication Date: 2021-07-29
• Subjects: artificial intelligence -- machine learning -- data flow characteristics -- industrial network communication security; Ubiquitous computing -- Periodicals; Embedded computer systems -- Periodicals; Electronic data processing -- Distributed processing -- Periodicals; Wireless communication systems -- Periodicals; Computer architecture -- Periodicals; 004.2
• Languages: English
• ISSNs: 1743-8225

Abstract

AI and machine learning are revolutionary technologies being explored by the communication industry to integrate them into communication networks, provide modern services, improve network efficiency and user experience. The intrusion detection system is important for ensuring security of the industrial control system. Hence, in this paper, a machine learning assisted intrusion detection system (MLAIDS) has been proposed to analyse data flow characteristics in industrial network communication security. The progressive use of proposed ML algorithms will improve IDS functionality, especially in industrial control systems. Analysis of data flow characteristics given in this article involves the method of ensuring an adequate degree of security for a dispersed industrial network concerning some main elements, including system features, the present state of requirements, the implementation of suitable countermeasures that may lead to reducing the security risk under a predefined, acceptable threshold. The numerical results show that proposed MLAIDS method achieves high detection accuracy of 98.2%, a performance ratio of 97.5%, a prediction ratio of 96.7%, F1-score of 95.8%, and less root mean square error of 10.5% than other existing methods.