Securing Unified Payments Interface: A Deep Learning Approach for Fraudulent Transaction Detection

Abstract

Unified Payments Interface (UPI) is a system that integrates multiple bank accounts into a single mobile application, enabling seamless fund transfers and business payments. The proposed study presents a deep learning-based approach for detecting fraudulent UPI transactions using a large online payments fraud detection dataset containing over 6.3 million transaction records. The dataset is highly imbalanced, with fraud cases forming only a small fraction, making it a realistic yet challenging test for fraud detection methodologies. To address this, a novel model combining BiLSTM and Transformer-based encoders was developed to capture both temporal dependencies and contextual relationships in transaction sequences. The performance of the models was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC metrics. Experimental results demonstrated that the BiLSTM model significantly outperformed conventional machine learning methods such as Logistic Regression, Random Forest, and Decision Tree. The BiLSTM achieved an accuracy of 99.90%, precision of 99.99%, recall of 99.81%, and F1-score of 99.91%. Visualization through accuracy/loss curves, confusion matrix, and ROC analysis further validated the model’s robustness and stability. These findings confirm BiLSTM as a reliable and effective real-time fraud detection system for digital payments, enhancing the security and performance of financial transactions compared to traditional approaches.