Learning of neural networks for fraud detection based on a partial area under curve

Abstract

This paper addresses an effective approach of training a neural network (NN) classifier for real-world credit card fraud detection. In the proposed approach, an evolutionary search algorithm is used to directly improve the performance of a NN classifier in a local operating range in terms of the detection rate of fraudulent usages by optimizing a partial area under a domain-specific curve. The experimental results on 'real' credit card transactions data demonstrate that the proposed approach produces classifiers of a higher detection rate in a desired range of false detection rates. © Springer-Verlag Berlin Heidelberg 2005.