Graph Neural Network-Based Edge Classification Approach for Detection of Cybersecurity Attacks

Cybersecurity attacks, such as denial-of-service and infiltration attempts, can disrupt networks, compromise sensitive data, and result in financial losses. The growing prevalence of IoT devices exacerbates these risks, demanding advanced intrusion detection systems (NIDS). This study proposes a Graph Neural Network (GNN)-based framework for detecting and classifying multiple cybersecurity attacks. Unlike traditional NIDS approaches relying on node classification, our method introduces edge classification using network flow data, embedding features for source-destination pairs to improve detection accuracy. The GNN model achieves high performance even with limited training samples, making it suitable for rare attack scenarios. Evaluated against benchmark models such as decision trees (DT), artificial neural networks (ANN), and support vector machines (SVM), our GNN outperforms these methods, achieving 97% accuracy on the CIC-IDS2017 dataset. These results highlight the potential of edge-based GNN classification to enhance network security frameworks significantly.