FraudShield AI: A Hybrid Multi-Modal Framework for Phishing, Message, and UPI/QR Fraud Detection

FraudShield AI: A Hybrid Multi-Modal Framework for Phishing, Message, and UPI/QR Fraud Detection presents a unified security framework designed to detect multiple forms of digital fraud, including phishing URLs, fraudulent text messages, UPI/QR payment scams, and suspicious documents. The proposed framework combines machine learning techniques and rule-based security mechanisms to analyze potential threats and generate explainable risk assessments. The system incorporates a 28-feature Random Forest classifier for URL detection, a TF-IDF-based Logistic Regression model for message classification, OpenCV-based QR code detection with deterministic UPI fraud detection rules, and OCR-based document analysis. FraudShield AI is implemented using a FastAPI backend and a React with TypeScript frontend. It also includes a Chromium Manifest V3 browser extension, an authenticated user dashboard, social safety alerts, and downloadable PDF incident reports. In controlled experimental evaluations, the URL detection model achieved 100% accuracy on 560 held-out synthetic URLs, while the message classification model achieved 95.83% accuracy on 24 held-out SMS samples. Both models reported an ROC-AUC score of 1.000. These results are based on controlled datasets and should not be interpreted as proof of equivalent performance in real-world fraud scenarios. This research discusses the system architecture, methodology, implementation, experimental evaluation, limitations, and potential future enhancements of the proposed multi-modal fraud detection framework.