Credit Card Fraud Using Adversarial Attacks

Banks lose billions to fraudulent activities every year, affecting their revenue and customers. The most common type of financial fraud is Credit Card Fraud. The key challenge in designing a model for credit card fraud detection is its maintenance. It is pivotal to note that fraudsters are constantly improving their tactics to bypass fraud detection checks. Several fraud detection methods for identifying fraudulent credit card transactions have been developed. However, in order to further improve on the existing strategies, this paper investigates the domain of adversarial attacks for credit card fraud. The goal of this work is to show that adversarial attacks can be implemented on tabular data and investigate if machine learning approaches can get affected by such attacks. We evaluate the performance of adversarial samples generated by the LowProfool algorithm in deceiving the classifier.