Research And Application On Network Traffic Classification Based On Machine Learning Technology

With the development of information technology, lots of network traffic created by a wide variety of Internet applications takes up many network resources, which brings great challenges to quality of service and negative impact on Internet security. In order to use network resources efficiently and provide effective means of control for network managers, the technology of network traffic classification to identify the application layer protocol has become the hot issues in recent years. Under the background and put the double-way data packets sequence in transport layer that have the same five tuple as the research object, the paper research on the network traffic classification based on machine learning technology and apply it.Firstly, the paper introduces the network classification technology based on well-known port and character matching, but the classification accuracy can’t meet the needs of present because of some defects. So we lead to the machine learning technology. Then the paper study on the Naive Bayes, C4.5decision tree, support vector machine and ensemble learning algorithm of classification.Secondly, based on the character of machine learning algorithm, the paper put forward two improved algorithm:one is the algorithm based on the sample reduction strategy and SVM and another one is the algorithm based on result feedback. The first kind of improved algorithm based on the theory of information gain rate and sample mass center, it deletes the attributes which have little influence on the result of the classification and the sample close to the center of sample mass, only uses potential support vector samples for training to ensure accuracy and increase the efficiency of training. Another kind of improved algorithm based on the collection of misjudgment sample, feedback the correct classification in misjudgment sample to the test sample and increases the accuracy of classifier.At last, the paper describes the capture of data packets, create of traffic, determine of attribute, process of sample labeling, discrete the continuous data. Then use classic data set on the algorithm based on machine learning and its improved algorithm in application. The results show that machine learning classification algorithm can avoid the defects of the traditional classification algorithm and classify different application layer protocol in a high precision. The experimental results show that the improved algorithm based on SVM can keep a high classification accuracy with a greatly shorten the training time, and the improved algorithm based on feedback machine learning successfully putting the correct result in misjudgment set feedback to the test sample, increase the classification accuracy. The experimental proved the improved algorithm has increased the training efficiency and precision of the original algorithm.