Accurate traffic classification is vital to numerous network activities, such as security monitoring, quality of service provisioning and network planning. However, current P2P applications, which generate a substantial volume of Internet traffic, use dynamic port numbers, HTTP masquerading and payload encryption to avoid detection. In this paper, we present an accurate P2P identification method using Bayesian networks. Based on the abstracted attributes of flow statistics, the optimal attribute subset is selected using genetic algorithms and P2P traffic is identified using Bayesian networks. We evaluate the algorithms and compare them to the previously used Naive Bayesian model and BP perceptron. Experimental results show that the proposed algorithms achieve better overall accuracy up to 95% with less cost. Further, our result indicates that the approaches are capable of identifying unknown P2P traffic and applicable to the real-time applications.
