Improved estimation of flexible pavement alligator fatigue cracks using hybrid gradient boosted models

Flexible pavements are prone to deterioration under repeated traffic loading and environmental influences. This study explores the use of ensemble machine learning for estimating the load associated alligator fatigue cracks in flexible pavements. A gradient boosted model (GBM) algorithm is employed and integrated with sparrow search algorithm (SSA), particle swarm optimization (PSO), and differential evolution (DE) optimization techniques to enhance the performance of the proposed framework. The current study utilized 323 data points from 86 sections across 19 U.S. states representing diverse climatic zones from the Long-Term Pavement Performance (LTPP) database. A total of 28 input variables, involving pavement structure, binder properties, asphalt layer properties, volumetric parameters, traffic, and climatic conditions, were evaluated and used to model the fatigue cracks. The Shapley Additive Explanations (SHAP) method was applied to assess the input variables in estimating the fatigue cracks. The results indicate that the GBM-DE model outperformed the other proposed models in both training and testing stages, achieving a coefficient of determination (R²) of 0.96 and root mean square error (RMSE) of 2.11 in the testing stage. The sensitivity evaluation shows that traffic level, pavement age, and climatic variables have a worthy impact on the modeling of fatigue cracks in flexible pavements.