Detecting fraud in cellular telephone networks

[摘要] ENGLISH ABSTRACT: Cellular network operators globally loose between 3% and 5% of their annual revenue totelecommunications fraud. Hence it is of great importance that fraud management systemsare implemented to detect, alarm, and shut down fraud within minutes, minimisingrevenue loss. Modern proprietary fraud management systems employ (i) classificationmethods, most often artificial neural networks learning from classified call data records toclassify new call data records as fraudulent or legitimate, (ii) statistical methods buildingsubscriber behaviour profiles based on the subscriber's usage in the cellular network anddetecting sudden changes in behaviour, and (iii) rules and threshold values defined byfraud analysts, utilising their knowledge of valid fraud cases and the false alarm rate asguidance. The purpose of this thesis is to establish a context for and evaluate the performanceof well-known data mining techniques that may be incorporated in the frauddetection process.Firstly, a theoretical background of various well-known data mining techniques isprovided and a number of seminal articles on fraud detection, which influenced this thesis,are summarised. The cellular telecommunications industry is introduced, including a briefdiscussion of the types of fraud experienced by South African cellular network operators.Secondly, the data collection process and the characteristics of the collected data arediscussed. Different data mining techniques are applied to the collected data, demonstratinghow user behaviour profiles may be built and how fraud may be predicted. Anappraisal of the performances and appropriateness of the different data mining techniquesis given in the context of the fraud detection process.Finally, an indication of further work is provided in the conclusion to this thesis, inthe form of a number of recommendations for possible adaptations of the fraud detectionmethods, and improvements thereof. A combination of data mining techniques that maybe used to build a comprehensive fraud detection model is also suggested.

[发布机构] Stellenbosch University