Development of Gene Expression Programming–Based Rutting Prediction Model for Smart Pavement Management Using LTPP Data

Rutting is a critical distress that severely compromises the performance of the road, especially in severe climate conditions and heavy traffic loads. Accurate rutting prediction is key to improving pavement maintenance and management. This study proposed a rutting prediction model based on gene expression programming (GEP) using the long‐term pavement performance (LTPP) database. The GEP model was trained using asphalt pavement data from the real world, extracted from the LTPP database containing important pavement performance variables such as pavement thickness (PT), dynamic modulus of AC layer, precipitation, temperature values, types of bound and unbound bases located under dry and wet (no‐freeze) regions, and traffic loads (annual average daily truck traffic [AADTT], gross vehicle weight [GVW], and equivalent single axle load [ESAL]), to comprehensively reflect actual pavement conditions. The dataset contains 166 data points from 39 different pavement sections located across various climatic zones with sufficient diversity in traffic and environmental conditions. The model achieved R 2 values of 0.87 (training) and 0.79 (validation), and root mean square error (RMSE) values of 1.1731 and 1.6264, respectively. The results show that the model has excellent predictive power for the rutting prediction. The GEP model outputs such symbolic regression equations, which can be expressed with explicit and interpretable mathematical forms, leading to a better‐equipped model framework for PMS application (transparency). SHapley Additive exPlanations (SHAP) analysis was also used to assess the contribution of input variables, significantly increasing the interpretability and reliability of the model.