A Novel P2P Traffic Identification Model Based on Ensemble Learning

Peer-to-peer (P2P) traffic has occupied major fraction of all internet traffic. Hence, P2P flow identification becomes an important problem for network management. In our work, we propose an ensemble classification approach for P2P traffic identification, which integrates six DTNB(combination of naive Bayes and decision tables) algorithm and dynamic weighted integration method. The proposed P2P identification scheme can be divided into three stages. In the first stage, we use feature selection algorithm to extract P2P flow characteristics. In the second stage, we use DTNB algorithm to learning the pattern of P2P traffic characteristics. In the third stage, we use dynamic weighted integration method to increase the detection accuracy and reduce false positive in classification. To verify the performance of the proposed P2P identification based on ensemble classification, we collect network traffic traces from NJUPT campus using NETMATE, and run WEKA experiments. The experimental results show that the ensemble classification approach for P2P flow identification can achieve at an average of 97% accuracy rate and 4% false positive rate. Through experiment and giving comparisons of precision, true positive, false positive and ROC curve between the proposed ensemble method and traditional methods such as naive Bayes(NB) , decision trees(DT) and single DTNB algorithm, we find that the proposed method has a better P2P traffic identification accuracy and stability.