DETECTING A SPIT CALLER USING UN-SUPERVISED MACHINE LEARNING

Spam over Internet Telephony (SPIT) is a form of un-solicited communication whereby an attacker sends automatically pre-recorded phone calls to several subscribers for purposes such as advertisements of product and services etc. In the near future, SPIT is expected to become a threat inhibiting the delivery of voice services over INTERNET because of its technical and economic characteristics. In this report, we purpose a detection mechanism to identify SPIT callers. This mechanism transforms the problem of identification of SPIT callers to a data clustering one by mapping callers to data points based on some characteristics. These characteristics (features) are gathered based on signaling characteristics, caller reputation and call metrics. These features include modified SymRank, termination coefficient, strong tie coefficient and average call duration. These features distinguish between a normal caller and SPIT caller but cannot stand alone as shown in the Results section. The data clustering was done by an un-supervised learning algorithm known as k-means which clusters closely related data points. To test our model, we simulated a testbed since we could not obtain real traffic. The testbed was generated in java, while the actual clustering algorithms were written Octave. The results gathered clearly showcased the high efficacy of our model, as it had accuracy (>95%) in all tests. After comparing our model with other models proposed by fellow researchers, we realized that our model either surpassed or was comparable to the other models considered.M.S. in Electrical Engineering, May 201