The Classification Research Of Network Games And Video Service Based On Feature Selection

In recent years, with the rapid development of the Internet, the network scale is growing and the different types of network multimedia applications are increasingly more and more rich. Among these, due to the large number of users, the degree of concern of network games and network video is also increased. How to identify the different types of network games and network video traffic accurately and efficiently is very important for improving the quality of service and user experience.This paper selects several current mainstream online games and the common types of network video as the research object, the network games selected include DOTA, DOTA2, LOL, HearthStone, XYQ and AssaultFire, the online video types selected are online standart definition video, online high definition video, online live video and video download. In the research of network traffic classification, this paper use the feature selection method to remove irrelevant features or redundant features in order to achieve the purpose of improving the accuracy of classification.This paper proposes a new SVM(Support Vector Machine) cascade classifier: At each level of the SVM classifier, by considering the information gain ratio and Pearson correlation coefficient to select feature, select the best combination of features for a certain type of data which can effectively distinguish this type of data from other types of data and use the selected best conbination to identify the type of data. The method proposed in this paper can be used in the classification of network games and network video traffic, and through many experiments, it can be proved that the proposed method can effectively improve the accuracy of classification.In order to verify the possibility of real-time classification of short sequence flow of network game and network video, the paper extracts the short sequence flow of network games and network video data with different packet number. The classification experiment find that a higher rate of correct classification can be obtained when the number of packets is small, the experimental results preliminary verify the possibility of real time classification of short sequences flow of network game and network video.This paper also classifies the data of the 6 main types of online games which are obtained in different time periods. According to the classification experiment, it is proved that the stability of the different flow statistic characteristics of different network games is not consistent in the long time scale.