ENHANCING UPI TRANSACTION SECURITY: A DEEP LEARNING APPROACH FOR FRAUD DETECTION

UPI fraud has become a major concern in digital payments, with cybercriminals using advancedtechniques to exploit security loopholes. Existing fraud detection systems often fail to accuratelypredict fraudulent transactions due to their evolving nature. Traditional models like ConvolutionalNeural Networks (CNN) struggle with large datasets, requiring significant computational power andtime, making them inefficient for real-time fraud detection.To address these limitations, a deeplearning-based ensemble model is proposed, combining Artificial Neural Networks (ANN), LongShort-Term Memory (LSTM), and Gated Recurrent Units (GRU). ANN detects complextransaction patterns, LSTM identifies sequential dependencies in financial data, and GRU optimizesefficiency by reducing parameters while maintaining accuracy. This integration enhances frauddetection by improving precision and minimizing overfitting.The ensemble model effectivelybalances computational efficiency and predictive accuracy. Unlike CNN, which faces challengeswith large-scale transactions, this approach processes vast amounts of data in real time. Moreover,by leveraging deep learning, the model continuously adapts to emerging fraud patterns, increasingits detection capability over time. This proactive fraud detection system strengthens security indigital payments, reducing financial losses for individuals and organizations while enhancing trustin online transactions.