Real-time fraud detection in telecommunication network using call pattern analysis

Abstract

Telecommunication service providers are losing considerable percentage of their annual revenue due to fraudulent activities. Such activities also deteriorate customer experience. Therefore, real-time detection of such fraudulent activities is required to minimize the revenue loss and to preserve customer experience. Illegal termination of International calls (aka. SIMbox fraud) and extreme usage scenarios related to International revenue share fraud are two major fraudulent activities which make highest impact. While such activities can be detected by identifying behavioral and calling patterns of subscribers, they need to be detected in real time so that subscriber connections linked with an ongoing fraud activity can be terminated to minimize the impact of threat or revenue loss. Call Detail Records (CDRs) produced by telecommunication equipment contains attributes that are specific to a phone call or other communication transactions handled by the device could be used to detect behavioral and calling patterns of subscribers. However, traditional CDR analysis techniques do not facilitate time-sensitive monitoring and analytical requirements. Therefore, we propose a Complex Event Processing (CEP) based solution for the real-time identification of fraudulent and extreme usage subscriber patterns. We identified a rich set of features and set of call patterns, and then combined batch analytics with real-time analytics to increase the detection accuracy. We demonstrated the utility of the proposed solution using a real dataset from a service provider. The proposed solution achieved an accuracy of 99.9% with average latency of 16 call attempts per detection at input event rate of 230 events per second with modest hardware.