TFHM:A Traffic Feature Hiding Scheme Based on Generative Adversarial Networks

The traffic encrypted by HTTPS on the Internet is increasing year by year. In most countries and regions, the proportion of encrypted traffic has exceeded 90%. The findings suggest that even encrypted traffic is at risk of privacy leakage. The anonymous communication network hides the communication relationship and transmission path between users through technologies such as mixnet and onion routing, but lacks a traffic feature hiding mechanism. Attackers can use the encrypted traffic analysis algorithm based on machine learning to de-anonymize the anonymous network. This paper proposes a traffic feature hiding scheme, called TFHM, which uses a generative adversarial network (GAN) to generate adversarial samples that are indistinguishable from target traffic, providing traffic feature hiding capabilities for anonymous networks and preventing listeners from classifying and analyzing anonymous traffic. According to the experimental results, the TFHM method can reduce the accuracy of encrypted traffic analysis methods based on Naive Bayes, Decision Tree, and SVM by 30%.