Research on encrypted traffic analysis based on federated learning
Cui Youwen 1, 2, Feng Qianye 1, He Yunhua 1, Gao Jiantong 1, 2, Shan Boyu 1, 2, Liu Xinyan 1
1. School of Information Science and Technology, Northern Polytechnic University; 2. Wenmai Lianfang (Beijing) Technology Co., Ltd.
Abstract： In the era of informatization, the encrypted traffic is exploding. While ensuring the security of information transmission, it also gives criminals plenty of opportunities, and poses unprecedented challenges to the classification and identification of traffic. Although traditional rule-based identification methods and flow-level behavior characteristics can achieve higher accuracy classification and identification, it still needs to be improved in data privacy and security.This paper focuses on the network encryption traffic identification system based on federated learning. Aiming at the traffic characteristics encrypted by SSL / TLS, an efficient encryption traffic identification model is proposed. The model mainly realizes the accurate classification of encrypted traffic through feature extraction and model training. The scheme can carry out information sharing and model training without touching the original data. The accurate encrypted traffic analysis model is obtained by weighted average strategy, and the high-risk traffic mixed in massive data is effectively monitored. Experiments on encrypted data sets effectively verify the feasibility of the method.
