Network Traffic Classification Based on Personalized Federated Learning

Machine learning methods have been widely applied to network traffic classification in the past decade. Most studies focused on improving the accuracy of network traffic classification by refining algorithms, neglecting the issues of machine learning security and privacy protection. In order to protect user privacy during the training process and maintain the accurate classification for the network traffic, this paper proposes a network traffic classification model based on personalized federated learning. Within the federated learning framework, we consider the statistical heterogeneity of the training data for local network traffic classification models, and an adaptive local aggregation module is utilized to integrate the global and local network traffic classification models towards local objectives which enhances generalization performance of the global network traffic classification model in each client, and ensures effective convergence of the federated learning model during the network traffic training process and provides stable classification performance. Experiments conducted on public network traffic datasets demonstrate that the proposed method can accurately classify network traffic with an accuracy rate of ${99.36 \%}$ and protect data privacy and security.