Network Intrusion Detection System Using Graph Attention Network and Particle Swarm Optimization

The number of people who access the Internet increases rapidly, which also leads to some problems besides the advantages. One of the most dangerous challenges is related to se-curity, which is not tackled efficiently by traditional mechanisms. Hence, an Intrusion Detection System using a Graph Attention Network (GAT) is proposed in this work. Firstly, a high-quality feature set that can improve the performance and decline data dimension is extracted by Particle Swarm Optimization-based feature selection. Afterward, the optimal models are selected by the 5-fold cross-validation (CV) and the chosen feature set for multiple Machine Learning and Deep Learning classifiers. Finally, reliable validation results are also achieved through the 5-fold CV technique. According to the performance, the optimal feature set can improve the classification outcome, and GAT is the best classifier for IDS design. Moreover, the achieved results also outperform previous research, which shows the prospects of tackling security problems in practical environments.