Credit Card Fraud Detection using Reinforcement Learning

Abstract

Financial transactions are still plagued by credit card fraud, which poses a serious threat to both individuals and businesses. The evolution of fraud techniques frequently outpaces the ability of antiquated methods to detect them. This research uses reinforcement based learning, more especially Deep Q-Learning, to examine credit card fraud detection. The first steps in this approach involved processing the dataset to extract features that would help with data normalization and classification. Subsequently, a DQN architecture that was appropriate for detecting credit card fraud was created and included parameters that would self-adjust over the course of several training sessions. After receiving training, DQN was able to distinguish between real and fraudulent transactions with an accuracy score of 90.54% on the testing set. To sum up, the findings suggest that the application of reinforcement learning, especially Deep Q-Learning, appears to be a practical and trustworthy technique for identifying credit card fraud. The constant learning process built on transaction practices makes it easier to predict how wrongdoing will change over time while maintaining transaction security. The current study adds to the body of knowledge on fraud prediction techniques by offering financial institutions, and businesses targeted advice and insights to help them effectively combat fraudulent activity.