Application Research Of Flow Statistical Features On Network Traffic Classification

Identify the type of network traffic accurately is a fundamental problem in networkmonitoring and management field. And it is also one of the hotspot and difficulty topicsof research. With more and more traffic transferred on the network, the flow types aremore and more complicated. The traditional classification technology using port numberor payload information is getting unable to meet the requirements and development ofnetwork traffic classification. So, the network traffic classification based on flowstatistical features attracts more and more attention. Based on the summary ofresearchers at domestic and abroad, this paper make a deeply study on network trafficclassification based on flow statistical features. The works of this paper are as follows.On the one hand, we put forward a network traffic classification method based onstatistical features. The method includes only the flow statistical features extracted fromthe raw traffic data. Then we use the clustering machine learning method based onK-means to obtain the flow clusters from the raw data. Finally, we combine the relevantflow and majority voting method to label the type of flow cluster. Experiments based onpublic datasets show that the method can reach high precision and F-measure value.On the other hand, we propose a general framework about the working procedureof the statistical feature based network traffic classification, and put forward aoptimization method for traffic classification on this framework. First of all, with theanalysis and summary of the previous research result, we put forward a general workingframework about the flow statistical based network traffic classification,. Secondly,based on this general framework, we use the filter strategy for feature selection in thefirst and the wrapper strategy of heuristic sequential forward searching in the next, findout an optimal feature subset at last. And we apply the optimal feature subset in trafficclassification. Public datasets indicate this method can always get a stable optimal set.The classification results indicate the optimal feature subset could obtain a better timeefficiency under the same classification precision.To sum up, this paper presents a network traffic classification method based onflow statistical features and constructs a general framework of traffic classification.Based on this framework, an optimization method for network traffic classification isproposed. By means of the optimization method, the classification speed can beimproved effectively.