Predictive Web Application for Road Accident Risk in the Peruvian Highlands Using Random Forest, GBM and Extra Trees

Emilio H.; Cabana O.; Carlos J.; Medina A.; Alberto C.; Quintana A. · Proceedings of the International Conference on Soft Computing and Machine Intelligence, ISCMI (2025) · DOI: 10.1109/iscmi67495.2025.11358482

Resumen

Road accidents remain one of the leading causes of injuries and fatalities worldwide, and the challenge is especially severe in rural and mountainous regions of Peru. The Peruvian highlands, with their narrow roads, steep slopes, and unpredictable weather conditions, concentrate more than half of the country's fatal accidents. Traditional prevention strategies, however, are often reactive and fail to anticipate where new accidents are most likely to occur. This paper presents a predictive web application that uses machine learning to identify high-risk road zones and supports accident prevention efforts. The system is powered by three decision-tree ensemble algorithms: Random Forest, Gradient Boosting Machine (GBM), and Extra Trees; chosen for their proven accuracy and robustness in structured and heterogeneous datasets. A dataset of more than 6,000 historical accidents from the National Road Safety Observatory (ONSV) was analyzed, including geographic, temporal, and road-related features. After data cleaning and preprocessing, the models were trained and compared using standard evaluation metrics. The best-performing model, Random Forest, achieved F1 = 92% and ROC-AUC = 0.94, meeting and surpassing our quality threshold on the 2021-2023 ONSV data. The predictive insights were integrated into a user-friendly web platform that displays interactive heat maps and filtering options by region and time. Unlike previous studies that focus on urban environments, our work emphasizes rural high-altitude contexts, where accident prevention tools are scarce. By providing accessible, data-driven visualizations, this system can support drivers, companies, and local authorities in making informed decisions that contribute to safer roads in vulnerable communities.

Instituciones: Universidad Peruana de Ciencias Aplicadas