Research On P2P Traffic Classification Based On FCM Algorithm

Peer-to-peer(P2P) applications have become one of the most important network applications with the fast development of network. It provides users with a wealth of resources and better quality of service. P2P technology brings so many benefits, then however, it also brings so many disadvantages, such as, network congestion, viruses, network security, copyright issues and so on. therefore, it has become critical to effectively identify P2P application.The identification methods which based on port number and payload characteristics are traditional methods for identifing P2P application. P2P applications based on dynamic random port number and P2P applications using encryption technology are gradually increased with the fast development of network, so these traditional identification methods are no longer apply to new P2P applications. In such situation, a method based on machine learning algorithms for dealing with P2P traffic classification was emerged. characteristics of P2P traffic as input stream are used in this method, and the characteristics are separated from ports number, protocols. Through continuous training classifiers to obtain high classification rate of classifier. Currently, there are so many research on machine learning algorithm, and machine learing algorithm can be divided into supervised and unsupervised. My paper have in-depth studied these algorithms, and based on previous research experience, I have mastered some rules.Supervised machine learning algorithm requires identified sample data, it can not recognize the new P2P applications. Then, more and more new applications are emerged with the fast development of network technology. So my paper use unsupervised classification algorithm to indentify P2P application. K-means algorithm is a typical unsupervised machine learning algorithm, it is a hard clustering algorithm, to some extent, it does not meet the objective reality. Therefore, based on full study of machine learning algorithms, my paper presents the fuzzy classification method to deal with P2P traffic classification. And ultimately I determine to use fuzzy C means(FCM) clustering algorithm to accomplish this task. During the study, I have in-depth studied FCM algorithm, packet capture, feature extraction and classifier's evaluation. I propose a effective method for determining parameter m of FCM. I selected matlab fuzzy logic toolbox to train the classifier, meanwhile, in order to contrast my experiment, I also selected K-means algorithm to be experimented based on Weka data mining software. Experiments indicate that the FCM algorithm has good classification results.