ScamShield AI – Intelligent UPI Scam Detection

About ScamShield AI

Inspiration

Digital payments have become an essential part of everyday life in India. However, users can encounter increasingly convincing scams involving impersonation, urgent payment requests, unfamiliar beneficiaries, and AI-generated messages. Many people struggle to recognise suspicious patterns before sending money.

We came up with ScamShield AI, a proposed intelligent fraud-awareness system designed to identify potentially suspicious payment patterns and communicate the associated risks in simple, understandable language.

What it does

ScamShield AI combines rule-based fraud indicators with a machine-learning model to assess transaction patterns. It aims to identify signals such as unusual transaction amounts, unfamiliar beneficiaries, and repeated payment attempts.

Rather than displaying only a risk score, the system explains the signals behind a warning and suggests safer next steps, such as independently verifying the recipient, reviewing the payment, or reporting a suspicious request.

How we plan to build it

The proposed prototype uses React.js for the user interface, Python FastAPI for backend services, and Scikit-learn for machine-learning-based classification. A Random Forest model can be trained and evaluated using synthetic or properly authorised transaction data.

The system will process permitted transaction features, combine model predictions with explainable rules, and display a low, medium, or high risk category with supporting reasons. The prototype will be evaluated using precision, recall, F1-score, false-positive rate, and processing latency.

Privacy and responsible AI

Privacy and user trust are central to our design. The prototype will use synthetic data during development and will not require UPI PINs, OTPs, passwords, or full payment credentials. A warning will indicate potential risk rather than prove fraud. The system will not automatically block transactions or accuse individuals based on a single weak signal.

Challenges and learning

Key challenges include distinguishing suspicious activity from legitimate unusual behaviour, reducing unnecessary alerts, explaining model predictions, and handling evolving scam patterns. We aim to address these challenges through careful feature selection, baseline comparisons, validation testing, and clear user-facing explanations.

Future scope

Future improvements could include authorised integration with banking or payment-provider systems, privacy-preserving behavioural analysis, improved detection of emerging scam patterns, and a human-review workflow for suspicious cases.

Our vision: Make digital payments safer by turning suspicious patterns into understandable warnings and actionable guidance.

Project status note: Describe only the features that have actually been implemented as completed. Any planned models, evaluations, integrations, or features should remain identified as proposed until demonstrated.