Traffic Classification Using Stream Data Mining Algorithm

The emerging of new classes of applications makes Internet congestion problem even worse. In order to monitor if network runs with security. high efficiency and stability and ensure enough bandwidth in local area network, traffic classification and application identi-fication become more important.The core content of this paper are included as follows:(1)We analyze characteristic of current traffic identification methods.And we make a brief introduction of the background of traffic classification, for example, current classifi-cation model, capturing network flow, network flow characteristic extracting. Secondly, we analyze the weakness of the existing traffic classification approach.(2) We present a traffic classification approach based on stream-data mining, namely CVFDT. Compared to the original methods of C4.5 and Naive Bayes, the nodes of CVFDT created incrementally and the accuracy rate will be increased as more training data flow into. In addition, the original classification model can't update dynamically, while CVFDT methods can be adapt to the changing network environment.(3)We explore the use of a stream data mining approach called STRKM for classifying traffic and design an on-line network application identification framework. STRKM, evolved by K-means algorithm, can update cluster model and detect new application. The experiment result shows this approach has lower error rate when the parameter is appropriate. It also can update the cluster center and can be well applied to the online traffic classification.We bring the stream data algorithm to the field of application identification, solve the problem of not identifying dynamic network.The research work in this paper not only extend the application area of the stream data algorithm, but also has certain practical significance for the online traffic classification problem.