The Design And Implementation Of Network Traffic Classification Based On Decision Tree

With the rapid growth in the number of network users and the expanding of network applications scale, the types of Internet applications based on TCP/IP protocol are becoming more and more. The traditional identification methods based on port or deep packet inspection are difficult to fulfill the demand of traffic identification in the future, because of the emergence of a variety of Internet applications with anti-monitoring capabilities. It becomes a highly challenging problem to identify network traffic efficiently, accurately, and intelligently on line. In this paper, we study the methods of network traffic classification based on decision tree, then we design and implementation the traffic classification system. The article mainly includes the following works:Firstly, we introduce the main technologies on Internet traffic identification, and analyze the advantage and disadvantage of identification methods based on port or deep packet inspection. Then we sum up the advantage of machine learning in Internet traffic classification. Furthermore, we compare Bayesian classification model with decision tree model and their basic ideas, and give the traffic classification basic idea and models based on decision tree, then we show the common way to select feature.Then, we design the flow and function modules of traffic classification system based on the approach mentioned before. The traffic classification system includes four modules which are traffic collecting module, features selecting module, traffic identifying module and visualization of the identification results module. Then we introduce the idea of constructing a decision tree, and design the C5.0decision tree in our traffic classification system.Finally, we show the way to collect the network traffic, and then extract the features which are proven high performance in collecting the Internet traffic.we classify the collected Internet traffic and show the result of classifying friendly. We launch several experiments on the traffic classification system, and analysis the traffic classification results which prove our system both accuracy and efficiency.We apply the machine learning to traffic classification. Then we design and build a build C5.0decision tree model and classify the selected traffic.