Network Traffic Classification Based On Message Statistics

Accurate and efficient network traffic classification is an important network management task. Two way messages in a session follow the underlying application protocol to exchange information. In this paper, we propose a novel application classification method based on message statistics, concisely representing the protocols' unique characteristics. We present algorithms using SVD-based and information gain based algorithms to select the proper message feature set. As shown by the evaluation experiments, using the selected message features, a simple decision tree is able to reach the classification accuracy over 99%, which is comparable to other more sophisticated machine learning results.