Tor anonymous traffic identification based on gravitational clustering

Abstract

The anonymous communication technology has brought new challenges to traffic analysis since it creates a private network pathway. Clustering analysis has been proved to be efficient in grouping Internet traffic. However, the cluster number of traditional clustering algorithms must be pointed, like K-means. In this paper, the gravitation is introduced into the process of clustering in order to develop an improved Tor anonymous traffic identifier called gravitational clustering algorithm (GCA). In the proposed method, we consider each sample in the dataset as an object in the feature space, and the new object moves into the corresponding cluster according to gravitational force and similarity. The GCA was applied to a data set consisting of 2366 Tor network flows and 20926 other network flows. Simulation test evaluated and compared the performance of the proposed classifier with three state-of-the-art clustering algorithms. The tests yielded that the average accuracy rate, R and FM coefficient of the proposed GCA algorithm exceed 0.8. However, among the other three clustering algorithms, K-means can achieve the highest detection rate (0.5).