Predicting and Analyzing Urban Road Accident Severity under Low-Visibility Conditions

Low visibility is a significant risk to driving safety. In this paper, three algorithms are utilized to predict the severity of traffic accidents under low-visibility conditions. These algorithms include multiple logistic regression, random forest, and XGBoost. The XGBoost model has the highest prediction accuracy compared to others. To further interpret the results of the XGBoost model, this paper uses the SHAP values to identify the ranked impact of various feature factors on accident severity. A comparison with high-visibility conditions reveals that certain factors, such as direct property damage and central isolation infrastructure, have a greater impact on accident severity in low visibility.