关键词: P2P流量识别, 有监督机器学习, 双支持向量机, K均值集成, 时间代价

Abstract: Most of the P2P traffic identification methods have the problem of high time cost. Therefore, it was proposed to use TWin Support Vector Machine (TWSVM) whose time cost was a quarter of the common Support Vector Machine (SVM) to build classifier. Kmeans ensemble was used to create labeled sample set and labeled sample set was combined as the training sample of the TWSVM. At last, the constructed classification model was used to identify P2P traffic. The experimental results show that the method based on Kmeans and TWSVM can significantly decrease time cost of the P2P traffic identification, and has a higher accuracy rate and better stability than the standard SVM.

Key words: P2P traffic identification, supervised machine learning, TWin Support Vector Machine (TWSVM), K-means ensemble, time cost