A Hybrid Approach to Network Intrusion Detection Based On Graph Neural Networks and Transformer Architectures

In this paper, we propose a model of a Network Intrusion Detection System (NIDS) named E-T-GraphSAGE (ETG), which fuses Graph Neural Network (GNN) and Transformer techniques. With the widespread adoption of the Internet of Things (loT) and cloud computing, network structures have become complex and vulnerable. The efficacy of traditional intrusion detection systems is limited in the context of novel and unconventional cyber-attacks. This paper proposes a novel approach to address this challenge. GNN is used to capture the complex relationships between network nodes and edges, analyze network traffic graphs, and identify anomalous behaviors. By introducing the Transformer, the model enhances its ability to handle long-range dependencies in network streaming data and to understand network dynamics at a macro level. The GraphSAGE-Transformer (ETG) model is proposed to optimize the edge features through the self-attention mechanism to exploit the potential of network streaming data and improve the accuracy of intrusion detection. The experimental results show that the model outperforms the existing techniques in key performance metrics Tests on several standard datasets (BoT-loT, NF-BoT-loT, NF-ToN-loT) validate the broad applicability and robustness of the ETG model, especially in complex network environments.