Predicting Errors in Accident Hotspots and Investigating Spatiotemporal, Weather, and Behavioral Factors Using Interpretable Machine Learning: an Analysis of Telematics Big Data

Abstract

The utilization of explainable machine learning models has emerged as a key technique for predicting and interpreting various aspects of road traffic accidents (RTAs) in recent years. This study aimed to predict the occurrence of errors in road accident hotspots and interpret the most influential predictors using telematics data. Data from 1673 intercity buses across Iran in 2020, merged with weather data, formed a comprehensive dataset. After preprocessing, 619,988 records were used to build and compare six machine learning models. and the best model was selected for interpretation using SHAP (SHapley Additive exPlanation). Six models including logistic regression, K-nearest neighbors, random forest, Extreme Gradient Boosting (XGBoost), Naïve Bayes, and support vector machine were developed and XGBoost demonstrated the best performance with an area under the curve (AUC) of 91.70% (95% uncertainty interval: 91.33% − 92.09%). SHAP values identified spatial variables, especially province and road type, as the most critical features for error prediction in hotspots. Fatigue emerged as an important predictor, alongside certain weather variables like dew points. Temporal variables had a limited impact. Incorporating various spatiotemporal, behavioral, and weather-related variables collected by telematics, our analysis underscored the significance of spatial variables in predicting errors in accident hotspots in Iran. Policymakers are advised to prioritize decisions strengthening road infrastructures to mitigate the burden of RTAs.