Published Paper Details:

--The large-scale usage of digital payment systems, more so the Unified Payments Interface (UPI) in India, has led to the corresponding increase in fraudulent transactions, which poses a serious threat to financial security. Static rules or conventional machine learning models are used by traditional fraud detection systems, which are incapable of learning and adapting to evolving patterns of fraud. In this paper, we propose a novel approach using Generative Adversarial Networks (GANs) for enhancing UPI fraud detection. Our GAN model consists of a Generator, which learns to create synthetic fraudulent transactions, and a Discriminator, which learns to distinguish between valid and fraudulent transactions. This adversarial pro cess enables the system to detect unseen or rare patterns of fraud that traditional models may not detect. We trained our model on a data corpus of transaction features such as transaction amount, time, location, and device metadata. Experiments indicate that the GAN-based approach outperforms baseline classifiers such as Support Vector Machines and Random Forests, particularly in detecting sophisticated and novel instances of fraud. Our proposed framework can be implemented in real-time fraud monitoring systems, thus providing a dynamic and adaptive protection against UPI fraud. Scaling the model to multi modal datasets and implementing the system for live transaction monitoring are the directions for future work.