Credit Card Fraud Detection Using Deep Learning

Abstract

In the light of using advanced technologies for the purpose of facilitating daily activities for people, using credit cards for online purchasing is highlighted. The increased level of using credit cards opens the door for malicious users to fraud, which in turn highlights the need to detect fraud for the purpose of protecting assets of users. In this context, this work presents a deep learning-based system for fraud detection in credit cards. The system is trained on a public dataset obtained from the common online repository (UCI) from the Kaggle website. Before training the system, pre-processing is conducted so that unbalancing data issue is solved using the SMOTE technique. Next, the data is cleaned from duplicated records and outliers. Three models (RF, SVM, and NB) are trained on the used dataset, where the cross-validation technique is utilized in the process of splitting original dataset into training and testing portions. In addition, correlation coefficient method is employed to select the most features that had the most contribution to the target class (fraud or normal). Experiments are conducted to (1) select the best K value for cross-validation firstly; and (2) secondly, to obtain the optimal value of correlation coefficient to be used as a threshold to elect the best features involved in the training process. Finally, a voting technique is used to generate the final class prediction. In terms of accuracy, the proposed system showed superiority when compared to similar works where the accuracy level that was achieved was 0.99985.