A Novel SPITters Detection Approach with Unsupervised Density-Based Clustering

With the rapid popularity of VoIP, SPIT (Spam over Internet Telephony) based VoIP has become a security problem that cannot be ignored and SPITters (SPIT callers) detection turns into an urgent issue. Data mining is a practical method of SPITters detection. This paper considers three commonly used characteristics of VoIP users and presents the fact that the characteristic data distribution of SPITters in real data space is non-globular and irregular. Moreover, a novel approach is introduced to identify SPITters employing density-based clustering algorithm DBSCAN. The results on real dataset are superior to other commonly used unsupervised clustering algorithm in terms of the recall and precision of SPITter cluster.