The Detection of Abnormal Behavior by Artificial Intelligence Algorithms Under Network Security

With the continuous evolution of network attack methods, traditional rule-based and signature-based security strategies are becoming increasingly hard to deal with increasingly complex network threats. The research focuses on the problem of network traffic anomaly detection in network security, and proposes an improved Transformer and Generative Adversarial Networks network traffic anomaly detection model. The innovation lies in utilizing the Patch segmentation in the Transformer module to reduce information loss, while introducing random masked data blocks to enhance the anti-interference ability of Generative Adversarial Networks, and proposing a class balance model. Therefore, a Transformer Multi Receive Field Fusion (Trans-M) model for network traffic anomaly detection is constructed. The performance test results showed that after category balancing, the accuracy, recall, and F1-score of each model were been significantly improved. The accuracy of the Trans-M model on the balanced dataset arrived 98.12%, an improvement of 8.59% compared to before balancing. The recall rate of the Trans-M model was improved by 8.62% to 97.86%. On Balanced F Score (F1-score), the highest score of the Trans-M model was 98.46%, which was 8.18% higher than before balancing. The experiment outcomes demonstrate that the raised network traffic anomaly detection system is superior to common anomaly traffic detection models and can meet the actual network security protection needs.