Comparative Analysis of UPI Fraud Detection Using Ensemble Learning
Dontha Madhusudhana Rao, T. Varun, Telagamsetty Viswajith Gupta, Dokala Manoj Kumar · 2025
The rapid rise of the Unified Payments Interface has made digital transactions faster and more convenient, but it has also led to a significant increase in fraudulent activities. This study explores two separate transactional data sets with extensive transactional data, followed by intensive preprocessing steps with advanced feature engineering and class balancing to fix internal quality concerns of data. A thorough evaluation of Machine Learning models is done including traditional algorithms like decision tree, Logistic Regression, and Naive Bayes, advanced Ensemble methods such as Random Forest, Gradient Boosting, AdaBoost, Bagging, Extra Trees, XGBoost, and CatBoost. Performance is assessed using metrics such as precision, recall, F1 score, ROC-AUC, specificity, and accuracy. Ensemble methods like XGBoost and CatBoost, frequently achieved above 99% accuracy with very few misclassifications. The results indicate that Ensemble-based approaches could effectively detect fraudulent activities and, most importantly, reduce financial losses, build confidence within customers, and pave the way for managing risk proactively with financial institutions and digital payment platforms.