A System for Predicting Accident Risk on Highways of Pernambuco

Abstract

Traffic accident statistics are a worldwide concern and bring great damage to society, both economic andsentimental. Machine learning models applied to accident prediction have the potential to serve as a tool in decisionmaking and to improve the accuracy and impact of accident reduction measures. This paper aims to develop avisual and interactive accident prediction system to help the decision-making process of federal road agents inPernambuco. The system consists of a machine learning model trained with accident data provided by Brazil'sFederal Highway Police and an interactive tool for viewing the riskiest points on the map of Pernambuco. Regressionmodels were applied to predict the number of accidents given the identification of the highway, section, year,month, day of the week and weather conditions. The Random Forest model presented the best results accordingto the evaluation metrics considered in the study.