GRAPH NEURAL NETWORKS FOR REAL-TIME CYBERSECURITY THREAT DETECTION AND INTRUSION PREVENTION

Abstract

Digital networking and connecting systems are developing rapidly, leading to more complex and frequent cybersecurity issues. Intrusion detection and prevention systems rely on a set of rules. Many of these systems cannot detect advanced attacks, as these systems also cannot understand complex behaviors of networks. This research inspects the potential of Graph Neural Networks (GNNs) to assist, in the modeling of network behavior as dynamic graphs, and to provide real time detection of cybersecurity threats and intrusion prevention. GNNs are able to describe the relationships and behaviors of elements of a network. GNNs are also able to detect abnormal communication and develop advanced features to improve the detection of threats. The research in this paper presents a solution framework that employs GNNs to analyze and detect threats against the integrity of network traffic and the interactions of network users and the security events in a network. Graph based learning techniques, and their potential to support GNN intrusion detection frameworks, are assessed. In this research, GNNs are considered as the advanced solution of intelligent cybersecurity systems. GNNs have the ability to deepen the understanding of behaviors and relationships while providing real time detection of threats and intrusion prevention.