Research On Network Traffic Classification Based On Machine Learning Method

With the development and popularization of internet, network traffic data has been growing at a remarkable rate, which brings the convenience to the people, yet, makes them to face how to carry on the analysis effectively to these mass data and improve the Internet Quality of Service as well as promoting Internet into a quicker and better development. This paper mainly does research in traffic classification based on machine learning method and related techniques, including sniffering the network traffic, generating the statistical features, assigning the flow example, feature selection, and classifying application type of network traffic.In network traffic classifications based on the machine learning method, gaining the network flows sample, including training example and test example, is very important, firstly, we obtain the network packets by sniffering, and classify the gathering network packet's into network flows according to five tuples, secondly, after integrating the information from Packet-Level and Flow-Level, and analyzing the packet's attributes (size, count, time, flag) and flow's attributes (time), 37 statistical features are generated and the feature vector is formed which represents each network flow. Finally, we integrate port-based, payload-based, protocol analysis to assign network sample, and obtain an auto sample assignment system with high precision.In network traffic feature selection, we propose a method of feature selection based on feature distance and genetic algorithm. This method can find the good initial community of genetic algorithm effectively, so we can find a good features set in little iterative number of times of the genetic algorithm. The experiment result indicates the method can reduce the features quantity in order to reduce the studies and classification time, in addition to remove non-correlated or the redundant features to increase classification accuracy.In classification based on machine learning aspect, we do research in classifying usual application types of network traffic by using six machine learning methods, and analyze the experimental result. The experimental result indicates that classifier based on machine learning method can avoid drawback of the traditional network traffic classification approach to classify the network flow well with protocol encryption or dynamic port.With the above research, Traffic Classification System Based on Machine Learning Traffic Classification System Based on Machine Learning (Traffic Classification System Based on Machine Learning, TCSBML) has been designed and realized, containing the network traffic data acquisition, network traffic data analysis as well as the classification functional modules and so on, which has certain use value.