Abstract
The purpose of this study is to examine the effectiveness of three distinct artificial intelligence models in identifying credit card fraud: the Support Vector Machine (SVM), Multilayered Perceptron Classifier (MLPC) and the Random Forest. To increase accuracy and robustness against adversarial assaults, the models were trained on an array of credit card payments. The results of the study showed that all three models had their own strengths and weaknesses, with the MLPC and Random Forest achieving highest accuracy, 99% and SVM model achieving the lowest accuracy with 95%. Furthermore, of the three models, Random Forest model was the most robust against adversarial attacks, a Generative Adversarial Networks (GAN) model was introduced for data augmentation, being the most successful in synthetic data generation and model validation. In addition to comparing the models, we also proposed a framework for institutions dealing with money and transactions to effectively detect and prevent credit card fraud using adversarial Machine Learning. Overall, the findings of this research highlight the worth of carefully weighing accuracy against robustness trade-offs when choosing a credit card fraud detective model. The integration of artificial data through GAN-based data augmentation further strengthens the models' performance, making them more reliable in real-world applications, and subsequently empowering institutions to safeguard their financial systems effectively.