This paper examines the robustness of machine learning (ML) models for credit card fraud detection (CCFD). Specifically, we investigate the impact of adversarial attacks on tabular credit card transaction data. While previous research has explored adversarial attacks on image data, research on tabular data in CCFD has been limited. In this paper, we employ a gradient-based adversarial attack method to attack tabular data in both black-box and white-box settings and analyze the results. Experimental results demonstrate that tabular data is vulnerable to even subtle perturbations, and that adversarial examples generated using gradient-based attacks are effective against non-gradient-based models as well. This highlights the need for developing robust defense mechanisms for CCFD algorithms.
