Using Graph Neural Networks to Improve Network Traffic Anomaly Detection Performance

Network traffic anomaly detection is an important part of ensuring the safe and stable operation of the network, but many deep learning-based methods have high computational costs when processing large-scale real-time traffic and are difficult to deploy in high-throughput network environments. To this end, this paper studies the use of graph neural networks to improve network traffic anomaly detection and other performance. This paper first defines and sets the topology and feature matrix based on the network traffic anomaly detection problem. Then, the study embeds the timing features, protocol features, and connection features of the traffic into the graph, aggregates the influence features and relationship features of different nodes in the neural network through the read-in module, and applies hidden features to establish the link relationship between the area of known target information and the unknown target information to achieve network traffic anomaly detection. Finally, this paper conducts experiments by converting real network data traffic into graph structure data. The experimental results show that the accuracy of the GNN model reaches 0.91, the precision is 0.89, the recall is 0.94, and the F1-score is 0.91, showing its efficiency and robustness in the task of network traffic anomaly detection. The accuracy results show that the two graph structures GCN and GAT perform similarly in the anomaly detection of node 1 and edge 1-2, with a precision of 1.00, indicating that the two models can effectively identify abnormal traffic of these nodes and edges.