Decision Tree Based Classification Of Internet Flows

With the development and changes of Internet technology, the emergence of a large number of new applications and technologies, it put forward new requirements and challenges to the management of network. the accurate and efficient identification network traffic contributes to slove these problems.Method based on statistical features is mainstream method for identification and classification of network flow now, the key point of this method is the choice of characteristics of network flow. This thesis analyzes HTTP,PPStream, QQ, Skype, Dota, XunleiKanKan,which are the six commonly used services on Internet, based on the statistical features such as characteristics of average packet size, the maximum packet size, entropy of packet size, stream length, flow rate, packet number, time interval and ratio of downstream and upstream packet bytes etc..Analysis found that the characteristics of various Internet traffic is different: Internet traffic can be divided into Xunlei KanKan and PPStream; HTTP,QQ and Skype,Dota by the characteristic of flow rate.The feature entropy of packet size can distinguish Xunlei KanKan and PPStream, QQ and HTTP, Skype; also HTTP and Skype can be easily recognition by the feature ratio of downstream and upstream packet bytes. Thus you can use flow rate, entropy of packet size, and the ratio of downstream and upstream packet bytes these features recognition and classification of Inteenet traffic. The study also found that after a period of time, characteristics of Xunlei KanKan with small changes and relatively stable, namely the collected data are valid. Finally, using the C4.5 algorithm of decision tree to do the classification experiments, verifying that the features we selected have more effective result for the identification and classification of specific network traffic.