Analysis of Cardholder Spending Behavior and Transaction Authentication to Enhance Credit Card Fraud Detection

As a result of the popularity and flexibility of using credit cards, the increment of credit card fraud trans-actions is growing and becoming a significant security issue that threatens cardholders, merchants, and financial institutions. Therefore, this paper proposes a predictive framework to improve credit card fraud detection by deeply analyzing cardholders' behavior, including spending, traveling, and shopping peak days behavior. Both profile approaches are used, the owner approach (cardholder spending behavior analysis) and the operation approach (transaction authentication and location). Moreover, our novelty method is applied to estimate the transaction's authentication score, which helps to distinguish between original and fraudulent transactions more accurately. In addition, traditional machine learning, ensemble learning, and deep neural network algorithms are applied to our predictive framework. The performance metrics used to evaluate our classifier models are accuracy, precision, recall, and f1-score. The recall metric got our attention because misidentifying fraudulent transactions will cause critical financial losses for all parties. Hence, the results show the effectiveness of our proposed framework and the improvement of the classification performance and prediction accuracy for all our classifier models due to its efficiency in understanding cardholder history and her/his transactions historical.