Research on accident risk prediction and grade classification of horizontal curve sections of expressways based on SHAP-SVR model

Abstract

Conventional prediction models used in expressway risk-assessment systems often provide insufficient support for identifying and classifying crash-prone road segments. This study develops an interpretable framework integrating SHAP with support vector regression (SHAP–SVR) to predict crash risk on expressway horizontal-curve segments and establishes a four-level risk classification scheme based on the predicted values. The proposed framework was evaluated using data collected from nine expressways in Anhui Province, China, and was further applied to segment-level roadway characteristics to identify high-risk locations. The results demonstrate that the framework achieves strong predictive performance while improving the interpretability of the machine-learning model by quantifying the contribution of individual risk factors. The proposed approach therefore provides a practical and interpretable basis for identifying, assessing, and managing crash risk on expressway horizontal-curve segments.