Real time VoIP traffic classification

This paper presents our research on using machine learning to classify Skype, VoIP and other traffic. We are interested in using a short sliding window rather than the entire flow. We show that the most effective features for classification are packet length and packet interval arrival time. Classifiers constructed to use these features are able to identify traffic into the categories of Skype, VoIP and Other with better than 99% recall when presented with a ten second sample of the flow.