Personalized federated learning framework for network traffic anomaly detection

With the widespread use of real-time sensors in various fields, such as IoT systems, it is important to improve the performance of most network traffic anomaly detection methods, which have low accuracy and high false alarm rates. However, there are two key challenges to address. In this study, we proposed a personalized federated anomaly detection framework for network traffic anomaly detection, in which data are aggregated under the premise of privacy protection and relatively personalized models are constructed by fine-tuning. Subsequently, a network traffic anomaly detection method based on the self-coding of long-and short-term memory networks was proposed. Real network traffic was tested to analyze the effects of the model structure and external noise on the detection performance, and the experimental results verified the correctness of the proposed method. Compared with other data-reconstruction-based detection methods, the proposed method has higher detection accuracy and better detection performance.

Automated summary

In this study, a personalized federated anomaly detection framework for network traffic anomaly detection was proposed, and a new network traffic anomaly detection method based on the self-coding of long-and short-term memory networks was introduced. The proposed method was validated through testing on real network traffic.