Application Of SVM In Detection Of P2P Network Traffic

In recent years, as a new network application, P2P technology is leading the direction of development of the Internet. At the same time, the management of P2P network has also become the biggest problem in the internet. Therefore, effective implementation of identification and control of the P2P traffic becomes the current problem to be settled urgently. Traditional control methods of P2P traffic are based on fixed-port detection technology, with the development of technology, applications of P2P tend to employ dynamic random port, port hopping, HTTP masquerading and payload encryption techniques to escape from detection. In addition, other methods are proposed, based on deep packet inspection technology and the flow characteristics of the detection technology, but none performs perfectly, if simply adopt one methods.Support vector machine has the advantage of good generalization ability and highly accuracy classification for the small sample set. The dissertation studies the application of support vector machine in P2P detection, on basis of the most sophisticated algorithm in statistical learning theory, which supports the related theory of support vector machine.We analyzed the P2P traffic feature and the problem in the control of P2P traffic, by consulting a number of literatures, and compared the current P2P traffic detection technology and the advantages and disadvantages respectively, proposed to identify P2P traffic using support vector machine. The main research and innovation of this dissertation: Firstly, we studied on support vector machines theory and least squares support vector machine, proposed an improved least squares support vector machine algorithm. Secondly, according the thought of applying support vector machine to P2P, we proposed a detection model of P2P traffic based on support vector machine, which using least squares support vector machine as learning machine model, designed P2P traffic detection experiments. By training classifier with cross-combination method, rational kernel parameter is obtained, and discussed affect of adjusting the parameters. We evaluate the classification in term of false positive, false negative and detection rate. Thirdly, we analyzed several kinds of multiple classifiers SVM algorithm, proposed the improved strategy of two categories support vector machine based on multi-classification algorithm, and designed an classification experiment based on two categories support vector machine, and achieved good results.In the end, we make a conclusion. Especially brings forward some suggestion for the dissertation, and points out the major creativeness for future work.