Research On Network Traffic Classification Based On Decision Tree

As the foundation of network management, traffic engineering and network security detection, the research to network traffic classification technology has significant important practical value. The traditional port-based and DPI-based classification method of network traffic becomes failure because of the popular of p2p technologies and encryption technologies. Classification method based on Machine Learning and traffic flow statistics becomes to be the new direction of this field as it can effectively address these issues above.Currently, most traffic classification based on machine learning app-roach aim to get great flow classification accuracy, without considering the bytes classification accuracy. But, with the imbalance of flow become more and more, bytes classification accuracy has more power to express the effectiveness of traffic classification. Therefore, how to improve the bytes accuracy on the condition that keep the flow accuracy is becoming a valuable subject in traffic classification.In this paper, C4.5_cs algorithm is applied to the classification of network traffic. What's more, a method is provided to calculate the cost matrix of C4.5_cs according the application background that classi-fication network traffic. Experimental results show that the method we use can improve the bytes accuracy of network traffic classification much, suitable for unbalanced traffic classification.As there are some invalid characters and redundant features in flow features of the network, feature selection algorithm is introduced into classification of network traffic in this paper. Because we use C4.5_cs algorithm for traffic classification, so a feature selection algorithm combine with chi-square statistics,C4.5_cs algorithm and the genetic search algorithm is proposed. Experimental results show that after the introduction of the feature selection algorithm to C4.5_cs in this paper, the speed of classifying newwork flow improved by nearly 2 times faster.Finally, on the basis of NetMate and Weka, we designed and implemented a network traffic classification prototype system based on C4.5_cs algorithm:NTFCS, and the main module and the key implemental techniques of traffic acquisition, classification, and data storage are briefly introduced.