Accurate prediction of pavement performance is essential for sustainable pavement management, especially in flood-prone regions where environmental stressors accelerate deterioration. This study develops a machine learning-based comparative framework to evaluate the contributions of baseline pavement condition, maintenance and rehabilitation (M&R) activities, and environmental exposure to predicting changes in Pavement Condition Index (ΔPCI) across 11,214 matched pavement segments in Southeast Texas from 2019 to 2023. Three nested modeling scenarios were evaluated using Linear Regression, Random Forest, and XGBoost, with performance evaluated using R2, MAE, and RMSE. Baseline variables alone showed limited predictive capability, whereas adding M&R history produced the largest improvement. Environmental and flood-related variables provided further gains, particularly for nonlinear ensemble models. XGBoost achieved the highest predictive performance in the fully integrated scenario (R2 = 0.65, MAE = 10.63, RMSE = 14.02). SHAP analysis identified SDI2019 and PCI2019 as the strongest predictors, while selected M&R and environmental variables also contributed meaningfully. The findings demonstrate that integrating treatment history and environmental exposure substantially improves pavement performance prediction and supports more sustainable, climate-resilient pavement management and helps agencies prioritize maintenance and allocate resources more effectively.
Bhusal et al. (Fri,) studied this question.