P2P flow classification based on wavelet transform

Abstract

P2P (Peer-to-Peer) flow classification is very meaningful for network management, performance analysis, quality of service (QoS) melioration, and so on, since P2P applications occupy most traffic of current Internet. Machine learning classification methods have attracted wide attention because of high classification accuracy, and the capability of classifying unknown P2P traffic. Existing machine learning methods mainly use the time domain characters of flows to classify P2P traffic. Experiment results show that this kind of methods has high classification accuracy if the training data and test data are captured from the same network environment. Otherwise, the classification accuracy bears great instability. The main reason is that some time domain characters of flows are instable and sensitive with the change of network environment. To improve the stability of machine learning classification methods, in this paper we carry out a framework of time domain and frequency domain characters based machine learning classification method. In addition to the existing time domain characters, we adopt wavelet transform based frequency domain characters of flows to machine learning classification method. Experiment results show that the proposed framework is sufficiently stable no matter the training data and test data are captured from the same network environment or not.