Online Network Traffic Classification Algorithm Based on RVM

Abstract

Since compared with the Support Vector Machine (SVM), the Relevance Vector Machine (RVM) not only has the advantage of avoiding the over- learn which is the characteristic of the SVM, but also greatly reduces the amount of computation of the kernel function and avoids the defects of the SVM that the scarcity is not strong, the large amount of calculation as well as the kernel function must satisfy the Mercer's condition and that human empirically determined parameters, so we proposed a new online traffic classification algorithm base on the RVM for this purpose. Through the analysis of the basic principles of RVM and the steps of the modeling, we made use of the training traffic classification model of the RVM to identify the network traffic in the real time through this model and the “port number+ DPI”. When the RVM predicts that the probability is in the query interval, we jointly used the port number and DPI. Finally, we made a detailed experimental validation which shows that: compared with the Support Vector Machine (SVM) network traffic classification algorithm, this algorithm can achieve the online network traffic classification, and the classification predication probability is greatly improved.