Generic attributes for Skype identification using machine learning

Abstract

Identification of Skype traffic using machine learning is an area of current research interest. Previous Skype classifiers have usually been reliant on version specific features. Consequently, a classifier that works for a specific version of Skype is unlikely to work for successive versions. Classification of Skype has been successful with previous research showing 98% precision for Skype version 3. But classification using Skype version 3 attributes were not successful in identifying Skype version 4. In our experiments, we use machine learning to classify Skype version 3 and 4 for a subflow size of 100 with characteristics common to both versions. We discuss attributes that are generic to Skype and show that Skype can be identified with 97% precision and 86% recall.