Detecting Encrypted Traffic: A Machine Learning Approach

Detecting encrypted traffic is increasingly important for deep packet inspection (DPI) to improve the performance of intrusion detection systems. We propose a machine learning approach with several randomness tests to achieve high accuracy detection of encrypted traffic while requiring low overhead incurred by the detection procedure. To demonstrate how effective the proposed approach is, the performance of four classification methods (Naive Bayesian, Support Vector Machine, CART and AdaBoost) are explored. Our recommendation is to use CART which is not only capable of achieving an accuracy of 99.9% but also up to about 2.9 times more efficient than the second best candidate (Naive Bayesian).