Analysis of features selection for P2P traffic detection using support vector machine

Network traffic classification plays a vital role in various network activities. Network traffic data include a large number of relevant and redundant features, which increase the flow classifier computational complexity and affect the classification results. This paper focuses on the analysis of different type of features selection algorithms in order to propose a set of flow features that are robust and stable to classify Peer-to-Peer (P2P) traffic. The process of validation and evaluation were done through experimentation on the traffic traces from special shared resources. The classification of P2P traffic is using Support Vector Machine (SVM) measurable in terms of its accuracy and speed. The experimental results indicate that P2P SVM classifier with reduced feature sets not only results in a higher computing performance (0.14 second for testing time), but also achieves high accuracy (92.6%).