Combination of multiple detectors for credit card fraud detection

This paper presents a signal processing framework for the problem of automatic credit card fraud detection. This is a critical problem affecting large financial companies that has increased due to the rapid expansion of information and communication technologies. The framework establishes relationships between signal processing and pattern recognition issues around a detection problem with a very low ratio between fraudulent and legitimate transactions. Solutions are proposed using fusion of scores which are related to the familiar likelihood ratio statistic. Moreover, the classical detection problem analyzed by receiving operating characteristic curves is mapped to real-world business requirements based on key performance indicators. A strong practical case which combines real and surrogate data is approached, including comparison of the proposed methods with standard methods.