Credit card fraud poses significant risks to financial institutions and cardholders, necessitating robust detection systems. This paper reviews advancements in machine learning and deep learning for credit card fraud detection, including traditional rule-based systems and advanced deep learning methods like ANNs, GNNs, LSTMs, and GANs. These techniques enhance detection accuracy but face challenges in interpretability, applicability, and data management. For instance, deep learning models often lack transparency due to their black-box nature, while class imbalance and dynamic fraud patterns complicate model applicability. Data issues, such as limited availability and privacy concerns, further hinder system development. To address these challenges, this study proposes theoretical solutions: integrating expert systems to improve model interpretability, employing hybrid techniques like GAN-SMOTE combinations and federated learning to enhance applicability, and leveraging advanced methods for data quality and privacy preservation. The findings highlight the need for future research to balance model complexity with transparency, ensuring scalability and addressing data limitations. These insights aim to guide the development of more efficient, accurate, and robust fraud detection frameworks to safeguard financial security and support the sustainable growth of the credit card market.
Cunxiang Bian (Wed,) studied this question.