Abstract
Credit cards provide a convenient and efficient means for online transactions. However, the rise in credit card usage has also led to an increase in fraudulent activities, causing significant financial losses for both cardholders and financial institutions. This study aims to enhance credit card fraud detection by addressing challenges such as public data accessibility, high-class imbalance, evolving fraud patterns, and high false alarm rates. Existing literature explores various machine learning techniques for fraud detection, including Extreme Learning Methods, Decision Trees, Random Forests, Support Vector Machines, Logistic Regression, and XGBoost. However, due to their limited accuracy, there is a growing need to apply advanced deep learning techniques to improve fraud detection performance. In this research, we conducted a comparative analysis of machine learning and deep learning algorithms using the European card benchmark dataset. Initially, traditional machine learning methods were applied, achieving moderate accuracy improvements. Subsequently, three convolutional neural network (CNN)-based architectures were implemented, further enhancing fraud detection performance. By optimizing the number of hidden layers, training epochs, and leveraging the latest deep learning models, our approach achieved outstanding results: 99.9% accuracy, 85.71% F1-score, 93% precision, and 98% AUC. Our proposed deep learning model outperforms state-of-the-art machine learning and deep learning techniques in credit card fraud detection. Additionally, by balancing the dataset and refining deep learning algorithms, we effectively minimized the false negative rate, making our approach viable for real-world fraud detection applications.