Towards Designing Robust and Efficient Classifiers for Encrypted Traffic in the Modern Internet

Xi Jiang, Shinan Liu, Saloua Naama, Francesco Bronzino, Paul Schmitt, Nick Feamster

Abstract: Over the past several decades, the Internet infrastructure has evolved in many ways and one notable trend is encrypted transport which renders conventional traffic classification methods increasingly less effective. In this position paper, we argue that existing classifiers for encrypted network traffic are suffering from crucial problems associated with low robustness against model drifts and inadequate efficiency for real-life deployment. We propose potential solutions to these challenges by reducing the feature space required for such classifiers and exploiting robust network-level features across multiple datasets across time and space.