Investigation Of Attribute Selection And Identification For Network Traffic Flow

Network traffic analysis and classification is a fundamental technology enabling network operators to monitor network usage and manage network effectively. It is of great significance to classify network traffic accurately. However, with the great development of Internet, network traffic becomes various, new traffic emerges in endlessly and private traffic is widely used. All these make traffic classification more difficult.Facing to this situation, in this thesis, the characteristics extraction and traffic identification of P2P streaming media based on deterministic characteristics was researched. Three identification methods were proposed, which were based on a certain string/bits, a port and a certain string/bits or some certain strings/bits. And a typical P2P streaming media application was used to verify the effectiveness of the characteristics extracted. Then, automatically mining signatures technology for protocol based on payload of traffic was proposed, and this technology was implemented as a module of a Network Traffic Classification and Analysis System. The correctness and validity of signatures generated by the automatically mining technology were verified by experiments. In addition, the classification of network traffic based on attribute selection was also researched in this thesis. The experiments for classification were based on the workbench of Weka, which is a data mining software. The experimental data was divided into two sets, UDP set and TCP set, based on their transmission layer protocol. The classification result of UDP set is satisfied, but TCP set’s still could be improved. All in all, removing irrelevant and redundant attributes from a training data set can simplify high-speed network traffic flow classification without much impact on the classification accuracy.