Artificial Intelligence-Based Intrusion Detection Using Graph Neural Networks for Internet of Things Networks

Abstract

Purpose: This work introduces a new network security paradigm called AI based intrusion detection system for IoT networks using Graph Neural Networks (GNNs) to improve their security. It solves the issue of many current IDS's that can be perceived as network traffic as discrete events, and lack the ability to detect relational, dynamic and hidden patterns of attack among IoT devices, flows and services.
Design/methodology/approach: The design of the study is quantitative research design, type of research is explanatory research and study design is cross sectional study. 510 respondents who know about IoT networks, cybersecurity, artificial intelligence, intrusion detection systems and IT security were included. The measurement model, the structural model, direct effects and mediating effects were assessed using PLS-SEM.
Findings:: The results show that the quality of the constructed graph of IoT networks, the ability of the GNN, the quality of the representation of the relation between the traffic items, and the optimization strategy for training the AI model have a positive and significant impact on the performance of IoT intrusion detection. The most significant direct effect was from the GNN model capability. The validity of the representation quality of relational traffic as a partial mediator was tested with regard to the three independent variables as well as in terms of the intrusion detection performance and as a complementary partial mediator.
Implications: It extends the graph theory, information processing theory and representation learning theory in the field of IoT cybersecurity theoretically. At a practical level, it assists in improving the design of graphs, the ability to model, the approach to optimization and the quality of the representation for cybersecurity managers, IoT engineers, AI developers, and network administrators.
Originality/value: This work is original because it emphasizes and clarifies how and why the quality of the representation of traffic in a relational model improves the safety of IoT systems while considering the relational representation of traffic as a key mediating mechanism.