Unsupervised Learning Based P2P Traffic Identification Technology Research

With the development of network technology of P2P,P2P technology is become morewidely, and the identification of P2P traffic has been pursued by P2P technology researchers.As more and more applications, the identification of P2P traffic has become increasinglydifficult.This paper introduces the P2P technology and analyzes several typical P2P trafficidentification technologies. From the advantages and disadvantages of these technologies, thispaper presents an improved algorithm; the improved algorithm is based on unsupervisedlearning of a kind of clustering algorithm. This paper from the data packet level and data flowlevel analyzes the statistical characteristics of P2P traffic, which the stream is selected averagevariance of packet size, the duration of flow, the flow of the packet size conversion ratio, thedata flow of the average number of bytes in the packet, and download and upload speeds ratio.Secondly, the paper briefly describes the advantages and disadvantages of K-means algorithmand DBSCAN algorithm, and then to be improved (DBK algorithm). Besides the finding theprocess of initial point of the algorithm, this paper adds the BIC which obtain BIC core pointsas the initial node, and then through the K-means clustering algorithm.Finally, this algorithm comparing with K-means algorithm and DBSCAN algorithm forthe experiment. The concludes were carried out from the accuracy and false positive rate.From these concludes：DBK algorithm running time is long but it is relative to the other twoalgorithms of CRT visits and its average accuracy is better, and the misjudgment rate isrelatively low. From these,we can illustrate that the algorithm has a better accuracy rate andlow false positive rate, so as to arrive an improved algorithm of this paper, and this algorithmis effectively and feasibility.