Authors: Associate Professor M. Prasanna Kuamr, Assistant Professor M.Sivaparvathi, Assistant Professor D. Sravya
Abstract: The UPI's ascent to prominence has been nothing short of spectacular, and it has completely altered the landscape of online payment systems by facilitating simple, fast, and secure money transactions. Since the quantity of UPI transactions has grown in tandem with illegal activity, effective fraud detection is critical for maintaining user trust and ensuring the security of these transactions. This paper proposes a machine learning-based solution for real-time detection of fraudulent UPI transactions. The proposed system makes heavy use of data preparation techniques such exploratory data analysis (EDA), feature selection, data cleaning, and missing value management to improve data quality and model performance. The procedure involves the training and evaluation of several machine learning algorithms, such as XGBoost, Naïve Bayes, K-Nearest Neighbours (KNN), Decision Tree, Support Vector Machine (SVM), and Logistic Regression. Ensemble learning methods, such as Stacking Classifier and Voting Classifier, are used to bolster the prediction accuracy and resilience even more. A Flask-based web app with SQLite authentication incorporates the trained model to provide secure, real-time fraud prediction. It turns out that the Stacking Classifier achieves the highest detection accuracy of 99.4% of all the machine learning models according to the trial results. With its scalability, efficiency, and reliability, the proposed technique can detect fraudulent UPI transactions, which contributes to the improvement of digital payment system security.
