Enhanced Non-Defended Website Fingerprinting Attack Model Using Deep Learning

Onion routing (Tor) is a routing protocol that uses multiple-layer encryption to protect privacy and security for authentic users during web browsing. The website fingerprinting (WF) attack, which may be used to identify which websites a user is browsing are one of the vulnerabilities present in the Tor browser, despite its security. There are many modern and advanced attack models to build a WF attack model but in this work, Convolutional Neural Network (CNN) is presented considering two different scenarios Close World (CW) and Open World (OW). In CW scenario, the proposed attack model attains 98.20% accuracy on traffic traces without utilizing any defense model and when the same attack model is used in the OW scenario, it attains 97.10% accuracy. In OW scenario, the precision comes out to be 0.99 and the recall is 0.90, even in the presence of imbalanced data without any defenses. Overall, in this work, the proposed non-defended attack model can identify monitored websites with a precision of 0.99, even in the real OW scenario.