Abstract: A P2P traffic identification model was constructed by the combination of K-means ensemble and support vector machine. It owns high accuracy, stability and overcomes complexity of cluster model. Firstly, the three base clusterer was formed by few labeled sample, and then the each cluster's label was assigned by MAP. The unlabeled sample's label is the same with the closest cluster. Identification model based on SVM was built by new sample set. hhe model makes the best of ensemble learning's stability and SVM's generalization ability, theoretical analysis and result demon-strate its feasibility.

Key words: Traffic identification, Support vector machines, K-means, Ensemble learning