Encrypted Network Traffic Analysis: To Website Fingerprinting and Beyond

Maintaining one’s privacy online is often thought of as using strong passwords and using encrypted network communications. However, website fingerprinting has been proven to expose even highly encrypted, salted, and padded communications. Using just packet size, direction, and a known IP address an adversary can observe and accurately predict websites a user visits online. In this paper, we go beyond just website fingerprinting and show an adversary is able to identify specific articles a user visits, specific Google searches they conduct, and specific actions they take in Virtual Reality. We analyzed network traffic by developing a set of features that yields better performance than previous work, and predicted user behavior using a Random Forest classifier. Ultimately we show that it is possible for an adversary to predict user’s online activity for various applications.