What question did this study set out to answer?

The aim is to develop a more accurate fraud detection system for digital payments, especially for UPI transactions.

April 11, 2026Open Access

Fraud Shield-UPI: The Secure UPI Fraud Detection System

Key Points

The aim is to develop a more accurate fraud detection system for digital payments, especially for UPI transactions.
Implemented a machine learning framework using SMOTE for handling class imbalance.
Utilized PCA for reducing dimensions of data to improve model performance.
Employed XGBoost for high-performance classification of fraudulent transactions.
Developed a comparative evaluation platform using traditional algorithms like Decision Tree and SVM.
Deployed as a web-based application for real-time fraud prediction and interaction.
The SMOTE-PCA-XGBoost model significantly outperformed baseline models in accuracy, precision, and recall.
The framework effectively reduced false positives and false negatives in fraud detection.
Demonstrated capability to adapt to evolving fraud patterns and complex transaction behaviors.
Practical feasibility shown for integration into real-world financial platforms.

Abstract

The rapid expansion of digital payment platforms has significantly transformed financial transactions worldwide. In India, the Unified Payments Interface (UPI) has emerged as one of the most widely adopted real-time payment systems due to its speed, convenience, and low transaction cost. However, the increasing popularity of UPI has also led to a substantial rise in fraudulent activities, including phishing attacks, unauthorized fund transfers, identity theft, and account takeover incidents. Traditional rule-based fraud detection systems rely on static thresholds and predefined heuristics, which are often unable to adapt to evolving fraud patterns and complex transaction behaviors. Furthermore, fraud detection datasets are typically highly imbalanced, where fraudulent transactions represent only a small fraction of the total data, making accurate detection more challenging. To address these limitations, this study proposes FraudShield-UPI, a machine learning-based fraud detection framework designed to improve the accuracy and reliability of fraud identification in digital payment systems. The proposed framework integrates Synthetic Minority Oversampling Technique (SMOTE) to handle class imbalance, Principal Component Analysis (PCA) for dimensionality reduction, and Extreme Gradient Boosting (XGBoost) for high-performance classification of fraudulent transactions. The system is implemented as a web- based application using the Flask framework, enabling real-time fraud prediction and interactive transaction analysis. In addition to the proposed model, a comparative evaluation platform is developed to benchmark traditional machine learning algorithms including Decision Tree, Support Vector Machine (SVM), and Random Forest using the same dataset and evaluation metrics. Experimental evaluation on a simulated UPI transaction dataset demonstrates that the proposed SMOTE-PCA-XGBoost model significantly outperforms baseline models in terms of accuracy, precision, recall, and F1-score, while effectively reducing both false positives and false negatives. The results highlight the capability of the proposed framework to detect fraudulent transaction patterns with improved reliability. The modular architecture and web-based deployment further demonstrate the practical feasibility of integrating the system into real- world financial platforms for enhanced digital payment security.

Ask AI

Helpful

Bookmark

View Full Paper