Improved SVM method for internet traffic classification based on feature weight learning

Abstract

Network traffic classification plays an extremely important role in network management and service. Support vector machine (SVM) is widely adopted to classify traffic flows for its high accuracy. All features selected are treated equally in traditional SVM network traffic classification, which take little consideration of that each feature exerts a different influence on classification. Therefore, we adopt feature weight learning to assign individual feature a corresponding weight according to its importance on classification. Moreover, SVM is a method to solve binary classes problem and multi-classification problem is usually decomposed into a series of binary classification problems. Considering the differences of sample distribution between those binary classifiers, the improved SVM network traffic classification method proposed in this paper computes its own feature weights and parameter values for each individual SVM binary classifier. Experimental results show that the improved method proposed has a higher and more stable performance.