Enhancing the robustness of credit card fraud detection systems against adversarial attacks using machine learning

Abstract

Credit card fraud detection systems are essential for protecting money transactions and stopping illegal use. The usefulness of these systems is seriously threatened by the increasing prevalence of adversarial assaults, as attackers try to alter input data in order to trick machine learning algorithms and evade detection. The problem of making credit card fraud detection systems more resilient to such hostile attacks is addressed in this study. It explores current defense tactics, analyses the status of adversarial attacks, and evaluates how they affect model performance. The study offers a comprehensive strategy that involves implementing several defense strategies, modelling hostile settings, and assessing their effectiveness using important performance indicators. The results show that the proposed approach greatly enhances the robustness of fraud detection models, effectively reducing the influence of adversarial manipulations Credit card fraud detection systems face a significant challenge as adversarial attacks grow more complex, enabling attackers to alter data and avoid detection. By methodically evaluating defense tactics against these threats, this study seeks to improve fraud detection models. Various adversarial attack types, including hybrid, white-box, and black-box attacks, are simulated to identify weaknesses in current systems. Four defense mechanisms, namely, Neural Cleanse, Random Noise, General Adversarial Training, and Defensive Distillation are implemented and assessed. The results demonstrate that none of these methods offer total protection, although they improve model resilience to some degree. It backs up the claim that financial institutions do not yet have infallible defenses against hostile attacks. The study provides insightful information about enhancing fraud prevention by outlining the advantages and disadvantages of each strategy. This study supports the development of more effective credit card fraud detection systems by offering actionable insights and a structured approach to designing and assessing protective strategies against adversarial threats. The findings have implications for financial institutions seeking to fortify their security posture and protect customers from fraudulent activities, fostering trust and confidence in digital financial transactions.