A GNN-based Novel Approach to Detect Malicious Traffic in Intrusion Detection System

Rapid acceleration in Internet of Things (IoT) has tremendously increased the amount of digital information shared over the network. This increase in information escalates the number of network intrusions to gain access to valuable information flowing in the network. This work presents a Graph Neural Network (GNN) based IDS using benchmark dataset (NSL-KDD). The proposed model detects network invasion, assures security of IoT devices and their associated traffic. Initially, the work applies filter-based feature selection techniques to select significant features from NSL-KDD dataset that can easily differentiate malicious traffic from normal network traffic. Thereafter, GNN is applied on these selected features and experimental results demonstrate the prominence of GNN to recognize intrusion by attaining 93.10% accuracy for binary and 89.64% for multi-class intrusion classification. This work also conducts experiments on various machine learning and ensemble algorithms and the result of the presented study highlights the importance of choosing the most efficient algorithm for the development of advanced IDS. The authors have further compared their results with existing solutions for intrusion detection in terms of accuracy, precision and recall and outperformed them.