This paper presents RoadGuard AI, a full-stack intelligent road accident risk prediction system that combines machine learning with real-time data integration. The system uses three real-world Kaggle datasets comprising over 3. 5 million accident records from the United States, United Kingdom, and India to train and evaluate Logistic Regression, Random Forest, and Gradient Boosting classifiers. The Gradient Boosting model achieved 84. 72% accuracy on the UK Road Safety dataset with 2 million records. The system integrates live GPS location detection, real-time weather data from Open-Meteo API, and reverse geocoding from OpenStreetMap to provide context-aware risk predictions. A novel smooth risk scoring formula using predictₚroba output generates a continuous 0-100% risk score instead of discrete risk classes. The prototype is built using Flask (Python) backend with JWT authentication and SQLite database, and a React (Vite) frontend with an interactive risk dashboard. The system demonstrates that combining multiple global accident datasets with real-time environmental data can produce actionable, evidence-based road safety predictions.
Katiyar et al. (Tue,) studied this question.