Network Traffic Classification Based On Multiple Classifier

With the rapid development of information technology, a variety of Internet applications generated a lot of network traffic, consumed a huge network resources, not only to the network quality of service has brought great challenges, but also lead to Internet security problems production. For rational and efficient use of network resources and provide effective management and control tools for network administrators, Network traffic classification is the foundation and prerequisite conditions. Among various kinds of network flow classification techniques, Based on machine learning technology with automaticity and adaptability, became hotspot issues in recent years.Machine learning field found that there exist a phenomenon which a different characterization, different classifiers on the classification performance are complementary to each other, at the same time, using the multiple classification may improve the classification accuracy. Therefore, the study of multiple classifiers attracted wide attention from scholars, and has achieved good results in a number of practical applications of handwriting recognition, time series prediction, gene sequence prediction. However, the application of multi-classifier approach in the field of network traffic classification has just begun. In this background, in this paper, the classification of network traffic based on multiple classifier technology were studied.The main work is as follows:1. introduce the basic concept of flow classification, the major traffic identification methods and their advantages and disadvantages, and focus on the identification method based on statistical features. This paper analyzed methods of machine learning algorithms for traffic classifications in detail, which provide a basis for further research to identify effective methods.2. Based on this understanding-in the actual network management, the managers want to be identified and may be controlled by just a few specific applications, this paper proposed to build applications for specialized applications concern classifier and method combinations work together. At the same time, by means of the experiment, to ensure that each application-specific classifier for applications has a good recognition results. This method requires less experimental work, but it get higher accuracy, the experimental results show that the classification performance was better than a single classification method involved in the experiment.3. we analyze a variety of concepts for the previous proposed multi-classifier method drift in-depth, on this basis, selected the integrated multi-classifier method based on the weight as a research starting point,improved the main technical aspects of the method.