A new Unsupervised User Profiling Approach for Detecting Toll Fraud in VoIP Networks

摘要

Significant amounts of money are lost worldwide due to toll fraud attacks on telecom service providers or their customers. These attacks can be detected or prevented by a fraud detection system. Acquiring labeled data for the analysis of fraud cases is a major problem. This paper proposes an autonomous unsupervised user profiling approach for fraud detection using Call Detail Records (CDR) as data for the analysis and considers problems like random fluctuations in data. Two profiles for each user are used to measure user behavior in different time spans. The two profiles of every user are compared to each other, and changes in user behavior are measured. Describing the change in a numeric value allows checking for extreme changes and detecting fraud. For the detection of random events, a global profile is used. Two profiles are cumulating behavior information for all users, measuring global events in a reliable way. The approach provides low false positive rates. Also, recent fraud cases concerning Fritz!Box Voice over Internet Protocol (VoIP) hardware are analyzed and a detection approach based on this work is proposed.