SECUREPAY: An Explainable Cross-Modal AI Framework for UPI Payment Evidence Verification Using Screenshot Forensics, QR Intent Analysis, and Transaction Anomaly Detection
The widespread use of the Unified Payments Interface (UPI) has made digital payments in India faster and more convenient. At the same time, it has created new opportunities for fraud, including edited payment screenshots, malicious QR codes, phishing attempts, and false claims of successful payment. Many existing methods examine image manipulation, QR-code content, or unusual transactions separately. As a result, they may miss inconsistencies that become clear only when these sources are considered together. This paper presents SECUREPAY, an explainable cross-modal artificial intelligence framework that brings these forms of analysis into one process. It combines screenshot forensics, Optical Character Recognition (OCR), QR-code intent analysis, and machine-learning-based transaction anomaly detection. A cross-modal consistency mechanism compares details such as payment amounts, beneficiary information, transaction identifiers, payment status, and payment intent across the evidence that is available. A weighted risk model then classifies the evidence as low, medium, or high risk and provides reasons for the assessment. The framework is intended to be evaluated through module-wise and integrated experiments, comparative analysis, ablation studies, robustness testing, and cross-dataset validation. Since experimental validation has not yet been completed, no performance improvement is claimed. SECUREPAY is designed to help users and merchants assess suspicious payment evidence; it does not replace independent confirmation that a transaction has been settled.
Authors
- Yash Kantaram Hembade
- Mayur Dnyaneshwar Kapadi
- Nikhil Balu Kapuskar
- Aditya Ramdas Naikade
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24286179
- Primary Topic
- Digital Media Forensic Detection
- Type
- article
- Field-Weighted Citation Impact
- 0.00