Design And Implementation Feature Learning And Detection Of Mobile Smart Terminal Traffic

The traditional Internet services fall behind fast growing demand of daily life with the popularity of the network, the mobile smart terminal technology is developing rapidly in recent years. In the context of the encrypted protocols widely used in the mobile smart terminal, the diversity and complexity of the traffic pose great challenges for network management. The research on the networking protocol level accounts for a large percentage of the traditional data stream classification methods. However, there are short of research on different applications based on one same encrypted protocol. In consideration of these issues, a classification method based on the statistical feature with the trie structure was proposed, and a system to classify the encrypted traffic was designed and implemented.Firstly, a detailed description of the common network traffic classification methods was given and analyzed. The general design methods and the evaluation criteria of the classification were presented and compared.Secondly, the specific model of the encrypted traffic was described. The vector of four flow statistical was constructed through the classification of the four application traffic with the SSL protocol for the mobile smart terminal. The continuous attributes should be discrete before used.Then, a method based on the trie was proposed to detect the traffic, a rangesplitting algorithm was presented to resolve the problem of cross interval. The methods to construct the trie and search the traffic type in the trie were described. The experiment result showed a good real-time of classification, and the accuracy reached above 95%.Finally, a mobile smart terminal encrypted traffic classifier system was designed. The whole structure of the system consisted of three modules: the data collection module、the feature learning module and the traffic classification module. We detailed the specific process and tested the functionality and the performance of the modules by designing the cases, the experiment results were much as predicted.