Template-based dynamic time warping credit cards’ fraud prediction model

Abstract

The use of credit cards for electronic commerce purposes has been on the increase in recent time. Credit card being the most acceptable and popular mode of payment, the number of fraud cases associated with it is also on the rise. As a result of the extensive nature of credit card transaction details, it is challenging to identify fraud in a credit card system in recent years, which implies that the identification of credit cards' fraud accurately, quickly, and effectively is a gray area of research. In this article, an automated template-based credit cards' fraud prediction (CCFP) model is developed using dynamic time warping (DTW) technique. This proposed CCFP technique is novel, as it has not been previously used to predict fraud in credit cards' systems.  The performance of this proposed DTW-CCFP model is verified using the dataset that contains the credit card transactions of European cardholders. In addition, the performance of this proposed DTW-CCFP model is compared with three different machine learning (ML) prediction models: logistic regression (LR), decision tree (DT) and random forest (RF).  The results were documented using two different performance metrics: sensitivity and false discovery rate (FDR). The proposed DTW-CCFP model outperforms the LR, DT and RF techniques. Of interest, the proposed DTW-CCFP model achieves this feat using a small portion (less than 5%) of the dataset to build the template in comparison to the LR, DT  and RF techniques, which uses between 20%-30% of the dataset for training to achieve a fairly reasonable result