Effective Adversarial Examples Identification of Credit Card Transactions

Credit cards are currently a prevalent method of transactions. However, credit cards are susceptible to forgery, leading to numerous cases of fraud. Such actions result in financial losses for consumers, merchants, and banks. Detecting a large number of well-crafted counterfeit credit cards is often challenging through manual means. As a result, much research has been focused on employing Artificial Intelligence to achieve high detection performance. However, the accuracy of these AI-based methods may be challenged by attack techniques using Adversarial Examples. To address this issue, this paper utilizes neuron activation status distribution and Deep Neural Networks as detection tools. Furthermore, the experiments employ three methods to generate Adversarial Examples, showcasing the effectiveness of the proposed detection approach. This ultimately aims to safeguard the rights of credit card users.