UPI Guard – AI-Powered UPI Fraud Detection System

UPI Guard is an AI-powered fraud detection and analysis system designed to help identify suspicious patterns in digital payment transactions. It demonstrates how machine learning and graph-based analysis can work together to investigate potentially fraudulent financial activity. The project is designed for students who want to understand how artificial intelligence can be applied to cybersecurity, financial technology, and fraud analytics.

Traditional rule-based fraud detection systems rely on predefined conditions, such as unusual transaction amounts or specific transaction patterns. However, suspicious activity can sometimes be difficult to identify by examining individual payments alone. For example, several accounts might send money to a common account, which then transfers the funds to another account within a short period. Although individual transactions may appear ordinary, the relationships between accounts can reveal a potentially suspicious pattern.

UPI Guard addresses this challenge by analyzing transaction information and incorporating graph-based features that describe relationships between accounts. It compares three approaches: traditional rule-based detection, machine learning using LightGBM with transaction-level features, and LightGBM enhanced with graph-based features. This comparison helps students understand how different fraud detection approaches can be evaluated using the same experimental dataset.

The system uses LightGBM, a machine learning algorithm based on gradient-boosted decision trees, to estimate the fraud risk associated with transactions. Graph-based features provide additional information about account relationships and money-flow patterns. The system can present suspicious activity through an alert queue and provide explanations to help users understand why a transaction or account was flagged for investigation.

UPI Guard also includes an interactive fraud simulation and investigation workflow. Students can explore simulated transaction activity, examine connected accounts, review alerts, and understand how different detection approaches behave. These features make the project suitable for learning about machine learning, graph analytics, fraud detection, model evaluation, and explainable artificial intelligence.

The application uses a Python-based backend, a React frontend, FastAPI for API services, LightGBM for machine learning, and SQLite for data storage. It is designed to run locally on a computer using simulated transaction data, making it suitable for demonstrations, experimentation, and academic project presentations without requiring a connection to live UPI infrastructure.

UPI Guard is an educational and research-oriented project for final-year students interested in AI, machine learning, cybersecurity, and financial technology. It provides a practical opportunity to explore how transaction analysis, account relationships, and machine learning can be combined to investigate suspicious payment activity.