Network Traffic Classification Based On Message Statistics

With the development of network technique and the rapid increase of the demand of applications on the network, it arouses great consumption of the bandwidth of the network, network congestion and the enormous debasement of network performance. In order to ensure the network performance all-around, all the components of the network need to be supervised. Network traffic classification is one of the important stages to the network programming and management. The speed and performance of the approach of network traffic classification influence directly the performance of network management, which includes network planning, traffic engineering, accounting and billing, and anomaly detection and mitigation. Therefore, the research of the network traffic classification becomes the focus.The thesis proposes the messages of network connection as the basic element of statistics based on analysis of behavior characteristic of kinds of network application. A excellent feature set,which includes maximal and minimal payload size of the request messages in bytes and the number of the messages in a session, is selected from original message statistics based features by SVD-based and information gain based feature selection algorithms. Then the thesis applies C4.5 decision tree to the classification of network traffic after analyzing and comparing popular machine learning algorithms.The experiment of network traffic classification is tested with the DARPA data sets of MIT Lincoln Labs and the data sets collected in Software College lab of Huazhong University of Science and Technology. And the result shows that the total classification accuracy achieves above 99% with the message statistics based C4.5 Decision Tree classification algorithm. Comparing with other network traffic classification techniques, this method is proved more advantageous.