Research On Unsupervised Classification Method Of Internet Video Traffic

With the popularization of Internet and the development of streaming media technical, there are various types of network applications emerging in the Internet. At the same time, the growth of network video traffic is attaching more and more attention. As we all know, video services have more strict requirement on bandwidth and delay and different network video services have different QoS requirements, which have a greatly influence on the effective network management. Accurate identification of video traffic is critical for efficient network management and better utilization of network resources. As a result, the issue of network traffic classification has attached the attention of the research community in recent years.This paper aims at six kinds of typical network video services, including online standard & high definition video play, HTTP-download video data, QQ video chat, Xunlei-download video and Sopcast live TV. This paper mainly studies two aspects of network video traffic classification, 1) the selection of statistical features; 2) the improvement of classification scheme. The major contributions of this thesis are list as follows:Different network video traffic has different features reflecting its essential attribute. How to find the effective features for video traffic classification remains a huge challenge. Therefore, this paper is dedicated to finding effective statistical features for video streaming classification. In this paper, the raw data streams are processed the raw data streams with thoroughly analysis and data mining. Then we find some new and effective features for video traffic classification by extensive study on statistical features of typical Internet video applications. Finally, we further validate their effectiveness in video traffic classification.According to the diversity of the network video applications and their different QoS requirement on demand, it is necessary to classify network video streams more fine-grained. Furthermore, the specific features are only suitable to distinguish among considered classes. With the number of they have to choose among increasing, the performance of all classification algorithms will decrease. That is to say, it is different to classify all kinds of traffic for one time. So a hierarchical clustering network video classification scheme is adopted in this paper. In this scheme, network video streams are classified with different features(or feature combinations) in each clustering layer. The experimental results show that the proposed method achieves significantly higher classification accuracy with comparison to existing methods.