Published Paper Details:

Digital payments have been transformed by the quick expansion of Unified Payment Interface (UPI) transactions, but this has also made them more susceptible to fraud. Due to the dataset's extreme imbalance and the fact that fraudulent transactions make up a very small percentage of all transactions, fraud detection in UPI transactions poses difficulties. This study suggests a strong framework for detecting fraud that makes use of ensemble learning techniques, such as XGBoost, LightGBM, and neural networks, to increase prediction accuracy and manage data imbalance. The use of sophisticated ensemble techniques to handle class imbalances such as oversampling, under sampling, and synthetic data generation--as well as the integration of feature engineering to detect anomalous patterns are among the major advances. Numerous tests show that the suggested models can minimize false positives while achieving high detection rates. The findings imply that ensemble learning considerably improves UPI fraud detection systems' effectiveness, offering a scalable and practical solution for practical uses.