An investigation on experimental issues in financial fraud mining

Financial fraud has shown itself to be a fundamental issue throughout history, and due to its substantial impact on society consideration into the best method of solving it is highly important. There are a several key experimental issues that are relevant to computational intelligence-based financial fraud detection, and in this paper we will investigate three of them: choice of detection algorithm, performance metrics, and feature selection. The characteristics of these three issues has been described as they are understood in existing literature, but we have identified that there are large gaps in the research that need further attention. We will reduce this deficit by conducting an in-depth investigation of detection algorithms, performance metrics, and feature selection using a series of controlled simulations for a credit card fraud problem and analysing the results.