Research And Implementation Of Traffic Identification Based On Support Vector Machine

With the development and popularization of internet, the amount of the application on the Internet has been growing quickly,and network traffic data has been growing quickly.the flow generated by which, is more and more complicated.Categorizing the internet flow,in case to improve the level of the internet service and make internet more secure, is really urgent right now.In this paper,we discuss about the SVM algorithm, then the way to capture data packets from the Internet and how we genarate network features from the packets,then the way to filter the features in able to optimize the effect of the categorize.finally accomplish a Internet data flow categorizing system based on the SVM algorithm, which has the accuracy of91percentMy work is mainly as following:I focused on the method by which we can categorize internet flow of data without the help of port and payload.Most of the Internet applications are using dynamic port to transfer data, the data itself are mostly encoded.So-it is very hard for traditional ways to categorize Internet data flow, including by port and payload.I use the SVM algorithm, accomplish a system which can categorize internet flow without the help of port and payload.I found more than20features used by SVM, and we chose some of them by the recursion algorithm. With the help of the chosen features we can categorize internet flow of data fast and accurate.For instance,the variance of the capacity of the data packets is highly related to whether the flow is P2P data flow.I adjusted the parameter and the kernel function of SVM, according to the method called cross validatation and the method called grid.By using the method, we guarantee the stability of SVM.I chose the most reasonable parameter,by using which the accuracy of SVM machine is generally well, not good at one case and bad at another.I accomplished the system, which can categorize flow of internet data by the accuracy of91percents.It overcome the weak point of the traditional Internet flow categorizing way,and achieve a very convincing accuracy.