The Effect of Feature Selection on Credit Card Fraud Detection Success

Abstract

In this study, a deep learning based credit card fraud detection system has been designed and implemented. The information provided by credit card transactions is not sufficient in fraud detection systems. In order to increase the success of the model, new features were created by grouping the previous transactions according to features such as merchant category, transaction type, and payment type. Then these features were selected with feature selection methods and the most impact ones were determined. Thus, thanks to the newly added features, the payment habits of the cardholder were better learned by the model and the success of the model was increased.