Data-Driven Prediction of Punchout Occurrence in CRCP Using an Optimized Gradient Boosting Model

Punchouts distress represents a major structural deficiency in Continuously Reinforced Concrete Pavements (CRCPs), contributing to premature deterioration, reduced ride quality, and increased maintenance demands. To improve the prediction of punchout occurrence, this study develops a hybrid data-driven modeling approach that combines Gradient Boosting Machines (GBMs) with Particle Swarm Optimization (PSO). The proposed framework utilizes 395 observations obtained from 33 CRCP sections in the Long-Term Pavement Performance (LTPP) database, incorporating structural, climatic, traffic, and performance-related variables. PSO was applied to systematically tune key GBM hyperparameters, including the number of boosting iterations, learning rate, and tree complexity, in order to enhance predictive accuracy. Model performance was evaluated using five-fold cross-validation, where the optimized PSO-GBM model achieved an average RMSE of 1.09 and an R2 value of 0.947, outperforming conventional GBM as well as Random Forest, Support Vector Regression, Artificial Neural Networks, and Linear Regression models. Variable importance and sensitivity analyses revealed that Layer 3 thickness, pavement age, annual average daily traffic, and precipitation play dominant roles in punchout development. The consistency of residual distributions and the stability of hyperparameter sensitivity trends further confirm the robustness of the proposed framework. Overall, the results demonstrate that integrating evolutionary optimization with ensemble learning provides an effective tool for modeling complex pavement distresses and offers practical support for proactive maintenance planning and long-term management of CRCP infrastructure.