Use of Dempster-Shafer theory and bayesian inferencing for fraud detection in mobile communication networks

Abstract

This paper introduces a framework for fraud detection in mobile communication networks based on the current as well as past behavioral pattern of subscribers. The proposed fraud detection system (FDS) consists of four components, namely, rule-based deviation detector, Dempster-Shafer component, call history database and Bayesian learning. In the rule-based component, we determine the suspicion level of each incoming call based on the extent to which it deviates from expected call patterns. Dempster-Shafer's theory is used to combine multiple evidences from the rule-based component and an overall suspicion score is computed. A call is classified as normal, abnormal, or suspicious depending on this suspicion score. Once a call from a mobile phone is found to be suspicious, belief is further strengthened or weakened based on the similarity with fraudulent or genuine call history using Bayesian learning. Our experimental results show that the method is very promising in detecting fraudulent behavior without raising too many false alarms. © Springer-Verlag Berlin Heidelberg 2007.