A Real-Time Fraud Detection Algorithm Based on Intelligent Scoring for the Telecom Industry
Kun Niu, Haizhen Jiao, Nanjie Deng, Zhipeng Gao · 2016
Fraud detection is one of the biggest challenges in the telecom industry. Commonly used approaches, such as rule sets, outlier detection, and classification, have high computational cost, so they don't work well on mass data in terms of accuracy and speed. Besides, those algorithms are not good at detecting new fraud patterns. In this paper, we propose an UIS (United Intelligent Scoring) algorithm for fraud detection which has three merits. First, it has lower computational complexity. We use Manhattan distance instead of Euclidean distance to measure similarity between fraud samples and ordinaries. Second, new fraud patterns can be detected effectively by joint fraud probability. Finally, UIS is able to generate and update real-time scores, which detects early-time fraud and minimizes economic losses. Integrated experiments on real datasets of the telecom industry demonstrate that UIS is real-time, effective, and robust in different situations.