Feature Mining for Internet Video Traffic Classification

Multimedia data especially video traffic becomes increasingly popular in Internet traffic. How to extract effective features from video streams for fine-grained classification is a huge challenge. This paper collected 6 kinds of typical online video streams from the real network, analyzes and proposes a new set of features, e.g. the statistics of valid main protocol values in the flow to remove useless information in getting valid values. Experimental results show that these new features perform better in fine grained video traffic classification in comparison with an existing method.