User profiling and classification for fraud detection in mobile communications networks

摘要

The topic of this thesis is fraud detection in mobile communications networks by means of user profiling and classification techniques. The goal is to first identify relevant user groups based on call data and then to assign a user to a relevant group. Fraud may be defined as a dishonest or illegal use of services, with the intention to avoid service charges. Fraud detection is an important application, since network operators lose a relevant portion of their revenue to fraud. Whereas the intentions of the mobile phone users cannot be observed, it is assumed that the intentions are reflected in the call data. The call data is subsequently used in describing behavioral patterns of users. Neural networks and probabilistic models are employed in learning these usage patterns from call data. These models are used either to detect abrupt changes in established usage patterns or to recognize typical usage patterns of fraud. The methods are shown to be effective in detecting fraudulent behavior by empirically testing the methods with data from real mobile communications networks.