Fraud Detection in UPI Transactions Using Machine Learning

Abstract

In this application, UPI (Unified Payments Interface) transaction data is used for fraudanalysis, with advanced machine learning models and explainability techniques applied. The system takes numerical and categorical UPI transaction records as input and preprocesses them through cleaning, encoding, and standardisation. Multiple regression-based classification algorithms, including Logistic Regression, Ridge, Lasso, and ElasticNet, are trained to identify fraudulent patterns. A final ensemble model is developed to improve robustness and prediction accuracy. Feature importance is derived using SHAP(Shapley Additive exPlanations) and surrogate regression models for interpretability. Model performanceisevaluatedusingaccuracy,precision, recall, F1-score, AUC, and confusion matrix. The proposed method predicts fraudulent transactions more accurately than traditional rule-based systems and provides transparent explanations for each decision, enabling secure real-time fraud detection.

Introduction

Digital payments in India have grown rapidly, with UPI becoming a widely used platform for instant and cashless transactions. However, this growth has also increased fraud risks such as phishing, unauthorized transactions, and scams. Traditional rule-based systems struggle to detect new and evolving fraud patterns, creating a need for advanced solutions.

Machine learning (ML) and artificial intelligence (AI) are used to address these challenges by analyzing large-scale transaction data and identifying patterns of fraudulent behavior. Techniques include supervised learning (e.g., logistic regression, decision trees), unsupervised learning (e.g., anomaly detection), and deep learning for capturing complex transaction patterns. However, issues like data imbalance, evolving fraud tactics, and the need for real-time detection remain challenges.

To improve performance and reliability, modern systems use hybrid and ensemble models, along with explainable AI tools like SHAP, which help interpret model decisions and increase transparency.

The proposed system is a web-based UPI fraud detection platform that allows secure data upload, preprocessing, feature engineering, and model training using multiple regression algorithms. It combines these models using ensemble learning to improve accuracy and reduce errors. The system also uses SHAP-based explainability to show why transactions are flagged as fraudulent.

Finally, the platform provides results through a user-friendly web interface and evaluates performance using metrics like accuracy, precision, recall, and AUC. The results show that the ensemble model performs better than individual models, offering a reliable and scalable solution for real-time fraud detection in digital payments.

Conclusion

The proposed machine learning–based UPI fraud detection system effectively analyzes transactional behavior by leveraging a combination of advanced regression-based algorithms and a robust ensemble learning framework. Through meticulous preprocessing, the raw UPI transaction data was cleaned, standardized, and transformed into a structured format, enabling the models to capture subtle behavioral cues and patterns linked to fraudulentactivity.Comprehensivefeature engineering further strengthened the system by extracting critical indicators such as unusual transaction timings, rapid payment sequences, device inconsistencies, beneficiary risk profiles,and deviations from normal spending behavior. Individual models—including Logistic Regression, Ridge, Lasso, and ElasticNet—were trained to evaluate their predictive strengths and interpretability, while the ensemble model demonstrated superior accuracy, improved recall of fraud cases, and reduced false positives by integrating the diverse learning capabilities of all base classifiers. To ensure transparency and regulatory compliance, SHAP-based explainability was incorporated, providing clear transaction-level insights into why specific payments were flaggedas fraudulent. The complete platform, implemented in Python and accessible through a secure web- based interface,supportsefficient data upload,real- time fraud scoring, and interpretability-driven analysis. Overall, the system delivers a reliable, transparent, and scalable solution for UPI fraud detection, significantly enhancing the ability of financial institutions to identify and mitigate fraudulent activities within India’s rapidly expanding digital payment ecosystem.

References

[1] Sharma, A., Gupta, P., & Ramesh, S. “Machine LearningApproaches for Digital Payment Fraud Detection.” IEEEAccess, vol. 11, pp. 45123–45136, 2023. [2] Rao, K., & Singh, R. “Sequential Pattern Analysis forMobile Payment Fraud Detection Using LSTM Networks.”Journal of Financial Crime Analytics, Elsevier, 2022. [3] NPCI.“UPIFraudTrendsandRiskIntelligenceReport.”NationalPayments CorporationofIndia(NPCI),2024. [4] Kumar, V., & Bansal, S. “Hybrid Predictive Models forFinancial Fraud Detection.” Springer Journal of BankingAnalytics,2023. [5] Gupta, P., Roy, M., & Bhattacharya, D. “Explainable AIUsing SHAP forFinancial Fraud Detection.” ACM Transactionson Intelligent Systems, 2024. [6] Verma, D., & Chatterjee, S. “A Review on Digital PaymentFraud and Machine Learning Countermeasures.” IEEETransactions on Information Security, vol. 18, no. 4, pp. 1129–1143,2022. [7] Patel, A., & Mehra, R. “Anomaly-Based Detection of UPITransaction Fraud Using Isolation Forest and One-Class SVM.”International Journal of Cyber Forensics and Security, 2023. [8] Kaur, J., & Sandhu, H. “Feature Engineering Techniques forEnhancing Fraud Detection in Digital Transactions.” ElsevierFinancial Data Mining Journal, 2021. [9] Zhou, Z., & Lin, Y. “Ensemble Learning Methodsfor CreditCard and Online Payment Fraud Detection.” Expert SystemsWith Applications, vol. 204, 2022. [10] Ibrahim, M., & Ali, F. “Deep Learning Models for Real-Time Financial Fraud Prediction.” Neural Computing andApplications, Springer, 2023. [11] Jain, S., & Das, A. “Behavioral Biometrics for SecuringUPI and Mobile Wallet Transactions.” Journal of InformationAssurance and Security, 2022. [12] Pal, R., & Tiwari, A. “A Comparative Study of LogisticRegression and Decision Trees for Online Fraud Detection.”International Conference on Machine Intelligence, IEEE, 2021. [13] Ahmed, S. et al. “Graph Neural Networks for Fraud RingDetectioninFinancialNetworks.”IEEETransactionsonNeuralNetworks and Learning Systems, 2023. [14] Reddy, K., & Mohan, P. “A Survey on AI-Based FraudDetection Systems in Digital Banking.” Journal of DigitalEconomy and Security, 2023. [15] Chakraborty, T., & Bose, P. “SHAP and LIME BasedInterpretability for Fraud Detection Models.” Wiley ExpertSystems,2024. [16] European Payments Council. “Instant Payment FraudDetection Using Machine Learning.” EPC Fraud PreventionReport,2021. [17] Sahu,A.,&Dwivedi,R.“Real-TimeFraudDetectioninUPI Transactions Using Ensemble GradientBoosting.”InternationalJournal ofFinancialTechnologyandSecurity,vol. 9, no. 2, pp. 88–101, 2023. [18] Bhattacharjee, S., & Khan, M. “Adaptive MachineLearning Models for E-Payment Fraud Detection UnderConceptDrift. ”IEEETransactionson EmergingTopicsinComputational Intelligence, 2024. [19] Narang, P., & Soni, K. “UPIScam Trends and MachineLearning–Based Countermeasures in India.” Journal ofCybersecurityand DigitalForensics,vol.6,no.1,pp.34–49,2023. [20] Ahmed,M.,&Ribeiro,R.“AnomalyDetectionTechniquesfor Financial Transactions Using Autoencoders.” ProcediaComputer Science, Elsevier, vol. 218, pp. 502–510, 2023.

Copyright

Copyright © 2026 Dr. D. Sreenivasulu, Shabad Chakri, Ujgiri Karthik, Ullengula Aravind . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.