A SHAP-based framework for visualizing regional heterogeneity in pavement deterioration

Abstract This study focuses on regional heterogeneity in road-pavement deterioration. Accurate prediction of deterioration is essential for efficient maintenance planning. However, traditional statistical deterioration prediction approaches often have difficulty in capturing the complex nonlinear interactions among deterioration factors and the regional heterogeneity of deterioration, particularly when how regional conditions around road pavement affect the deterioration is uncertain. This study developed a SHAP-based framework for visualizing regional heterogeneity in pavement deterioration. Regional dummy variables are included in the model, and their SHAP values are interpreted as distributions to quantify the influence of regional characteristics on pavement deterioration. A case study using approximately 430,000 inspection records from Japan’s national highways validates the framework. The results reveal that regional location is less influential than primary factors such as time in service and traffic volume, yet its impact is comparable to structural characteristics such as pavement material. Western Japan exhibited overall low deterioration speed, while eastern Japan, especially the Sea of Japan coastal regions, showed high deterioration speed. Overall, the approach visualizes regional heterogeneity in detail and may support region-adaptive maintenance planning.