Web Site Fingerprint Attack Generation Technology Combined with Genetic Algorithm

Abstract

An anonymous network can be used to protect privacy and conceal the identities of both communication parties. A website fingerprinting attack identifies the target website for the data access by matching the pattern of the monitored data traffic, rendering the anonymous network ineffective. To defend against fingerprint attacks on anonymous networks, we propose a novel adversarial sample generation method based on genetic algorithms. We can generate effective adversarial samples with minimal cost by constructing an appropriate fitness function to select samples, allowing us to defend against several mainstream attack methods. The technique reduces the accuracy of a cutting-edge attack hardened with adversarial training from 90% to 20–30%. It also outperforms other defense methods of the same type in terms of information leakage rate.