This paper presents a VoIP Fraud Detection Framework by exploiting VoIP and/or network-OSS/BSS vulnerabilities. This can be accomplished by analyzing the behavior of the VoIP user using an ontology model so that different types of fraud scenarios could be identified. Using this ontology, an unsupervised learning algorithm has been implemented that describes the user behavior and/or the correlation among various features by analyzing CDR data. The statistical model that has been used is a Bayesian Network. The performance of the proposed model is optimized (minimizing the percentage of false alarms) by configuring the parameters of the Bayesian Network properly.
展开▼
