Impacts of External Factors on Crash Injury Severity in Urbanised Areas: An Exploratory Analysis

Abstract

The safety of urban roads is closely intertwined with residents' daily travel and has consistently been an important research topic of concern. In addition to the subjective behaviour of drivers, understanding the impact of external environmental factors on the severity of crashes is critical to risk management. As a result, this study employed a Bayesian Optimisation-Light Gradient Boosting Machine (BO-LightGBM) to investigate the effects of land use, weather and road conditions on the severity of urban car crashes on workdays and holidays. Additionally, the Shapley Additive explanation (SHAP) was adopted to explore the non-linear effects of the variables. The dataset was records of car crashes in the main urban area of Hangzhou from 2011 to 2017, a period with rapid urbanisation. The results indicate that the LightGBM model achieves a significant performance boost and outperforms traditional regression models and XGBoost after Bayesian optimisation. Crashes that occur in the office and congested areas on workdays generally result in less severe injuries; the temperature, humidity and visibility show strong correlations with crash severity. The findings also highlight which areas are more likely to produce serious crash injuries and provide insights into urban crash prevention.