Research on Intrusion Detection Based on Self-Attention and Residual Network

With the network technology's rapid advancement and application scenarios' continuous expansion, network attacks' frequency and complexity have steadily increased, posing increasingly severe threats to network security. To address the challenges in traditional intrusion detection models, including insufficient feature extraction capabilities, the risk of gradient vanishing during training, and low detection accuracy, this paper proposes a deep learning model that integrates self-attention mechanisms and residual networks for intrusion detection. By incorporating the self-attention mechanism, it captures the dependencies between different features in the input data, enhancing the model's ability to recognize complex patterns. Additionally, the introduction of residual blocks from the residual network addresses the gradient vanishing issue during model training, improving the model's feature extraction capability. The model was verified on the KDDCup99, UNSW-NB15, and CIC-IDS2018 datasets, the accuracy rates were 98.33%, 98.69%, and 98.48%, and the precision rates were 98.75%, 97.99%, and 96.32%, and the recall rates were 96.42%, 98.33%, and 97.84% respectively. Compared to other models, the proposed model demonstrates better performance in intrusion detection.