Effective defense against fingerprinting attack based on autocorrelation property minimization approach

The website fingerprinting attack is one of the most important traffic analysis attacks that is able to identify a visited website in an anonymizing network such as Tor. It is shown that the existing defense methods against website fingerprinting attacks are inappropriate. In addition, they use large bandwidth and time overhead. In this study, we show that the autocorrelation property is the most important success factor of the website fingerprinting attack. We offer a new effective defense model to resolve this security vulnerability of the Tor anonymity network. The proposed defense model prevents information leakage from the passing traffic. In this regard, a novel mechanism is developed to make the traffic analysis a hard task. This mechanism is based on decreasing the entropy of instances by minimizing the autocorrelation property of them. By applying the proposed defense model, the accuracy of the most effective website fingerprinting attack reduces from 98% to the lowest success rate of the website fingerprinting attack, while the maximum bandwidth overhead of the network traffic remains on about 8%. Recall that the current best defense mechanisms reduce the accuracy of the attack to 23% with a minimum bandwidth overhead of more than 44%. Hence, the proposed defense model significantly reduces the accuracy of the website fingerprinting attack, while the bandwidth overhead increases very slightly (i.e., up to 8%).