Predictive Model for Detecting Traffic Accident Risks Using Random Forest in the Transport Sector

Del Pino Mejia H.G.; Surichaqui M.A.E.; Alarcón J.S.D. · 35th International Conference on Computer Theory and Applications, ICCTA 2025 - Conference Proceedings (2025) · DOI: 10.1109/iccta68914.2025.11519955

Resumen

Accidents in the transportation sector are a critical public safety issue in Peru, especially in densely populated urban areas. Traditional response systems are often insufficient to prevent incidents due to their reactive nature and the lack of real-time risk analysis. To overcome this limitation, a predictive model based on the Random Forest algorithm is proposed to identify and monitor traffic accident risks with greater accuracy. The proposed solution analyzes large volumes of data from multiple sources, such as GPS coordinates, Geospatial maps, weather conditions, and historical records, to detect hazardous patterns and generate early warnings. The system processes data in real time using machine learning techniques, considering factors such as vehicle density, poor road conditions, and adverse weather. The proposal is structured into six main components aligned with the data pipeline methodology: data acquisition, storage, processing, analytics and predictive modeling, exploitation, and visualization. Together, these phases enable a complete flow from collecting and organizing heterogeneous data sources to generating predictive insights, deploying early warning mechanisms, and delivering visualization tools such as heat maps and automated alerts for decision-makers. The solution integrates mechanisms to ensure data quality and a continuous learning approach, enabling the model's accuracy to improve as new data is incorporated. The expected outcome is a measurable decrease in risk exposure, with accident-prone hotspots potentially showing up to a 25% reduction in critical incidents when compared to traditional monitoring methods.

Instituciones: Universidad Peruana de Ciencias Aplicadas