DOOMSDAY: AN AI-BASED FAKE WEBSITE DETECTION AND PAYMENT FRAUD PREVENTION SYSTEM

Online payment fraud has emerged as one of the most damaging consequences of the rapid
digitization of financial services. Phishing websites, which impersonate trustworthy platforms
to extract sensitive information and payment credentials from unsuspecting users, represent a
particularly persistent and evolving threat. Most existing detection solutions are reactive in
nature, identifying fraud only after a user has already been exposed. This paper presents
Doomsday, a real-time AI-based system that detects fraudulent websites and actively
prevents users from completing payment transactions on unsafe platforms. The system
employs an XGBoost classifier trained on a balanced dataset of 11,430 URLs characterized
by 89 features encompassing URL structure, domain metadata, behavioral attributes, and
web-level signals. Hyperparameter tuning improves model accuracy from a baseline of
96.28% to a final 96.81%. A custom suspicious offer detection module provides an additional
behavioral layer by scanning URL strings for manipulative discount and reward-related
keywords commonly associated with scam activity. Explainable AI through SHAP delivers
transparent, feature-level reasoning for every classification decision, allowing users to
understand precisely why a website has been flagged. A Streamlit-based web interface
displays real-time risk scores, presents human-readable explanations, and disables payment
functionality automatically when fraud is detected. Doomsday demonstrates that proactive
pre-payment prevention, rather than post-incident response, is both technically feasible and
practically impactful in protecting users from online financial fraud