Evaluation of GAN Applicability for Intrusion Detection in Self-Organizing Networks of Cyber Physical Systems

Abstract

Wireless self-organizing adhoc network of cyber physical systems, a kind of machine-to-machine (m2m) communication network connecting different cyber devices, e.g. networks of moving vehicles and mobile facilities (VANET, FANET, MARINET, MANET), industrial Internet of Things (IoT), sensor network of smart building, has become a concern for cyber security for last decade. Every device in this network is a cyber physical system that can independently get commands, do operations, move in the space, and it becomes possible to manage it remotely. For its security sensitivity, this ITC system is required to be protected with an adequate and effective intrusion detection mechanism. Traditional artificial neural networks (ANNs) applied to solve this issue become unhelpful for their computational constrains for big data processing and lack of comprehensive training datasets. The paper explores generative adversarial ANNs to detect security intrusions in large-scale networks of cyber devices. Based on the results of the experiments, an assessment is made of applicability of generative adversarial ANNs to detect security anomalies, and also practical recommendations are discussed for their utilization in the connected cyber physical networks.