Traffic Analysis of Web Browsers

Tor network is currently the most commonly used anonymity system with more than 300,000 users and almost 3000 relays. Attacks against Tor are typically confirmation attacks where the adversary injects easily discernible traffic pattern and observes which clients and/or relays exhibit such patterns. The main limitation of these attacks is that they require a “powerful” adversary. Website fingerprinting is a new breed of attacks that identifies which websites are visited by a Tor client by learning the traffic pattern for each suspected website. Recent works showed that some classifiers can successfully identify 80% of visited websites. In this paper we use a classic classifier, namely, decision trees (C4.5 algorithm) and we study to which extent popular web browsers can resist to website fingerprinting attacks. Among four studied web browsers, Google Chrome offers the best resistance to website fingerprinting (5 times better than the other web browsers). Since most of existing fingerprinting techniques have been evaluated using Firefox web browser, we expect the accuracy results of existing works to be reduced in case Chrome browser is used.