Intelligent Prediction of Pavement Structural Degradation Using a Multimodel Ensemble Learning Approach

Accurate prediction of the long‐term structural response of pavement systems is critical for performance evaluation and intelligent maintenance within the context of structural health monitoring. This study proposes an automated machine learning framework to predict the time‐dependent evolution of pavement rutting and deflection using full‐scale monitoring data collected from the RIOHTrack facility between 2016 and 2023. A high‐dimensional time‐series dataset was constructed to integrate loading, environmental, and material parameters. The AutoGluon platform was employed to perform multimodel comparison and develop a weighted stacked ensemble model incorporating XGBoost, random forest, and neural networks. The ensemble achieved superior generalization performance, with R 2 values of 0.9821 and 0.9817 for rutting and deflection prediction, respectively. SHAP analysis revealed that cumulative load, service time, and temperature were dominant factors influencing structural response. Long‐term forecasting indicated cyclic yet progressive degradation patterns, consistent with the measured temperature sensitivity and mechanical evolution of the pavement structure. The proposed approach demonstrates the potential of automated ensemble learning for structural health monitoring and long‐term performance assessment of pavement systems.