Predictive modeling of longitudinal cracking in CRCP using PSO-tuned gradient boosting machines

Abstract Longitudinal cracking poses a serious threat to the longevity and functionality of continuously reinforced concrete pavement (CRCP). Using structural, traffic, and climatic data taken from the Long-Term Pavement Performance (LTPP) database, this study presents a machine learning system based on a gradient boosting machine (GBM) optimized using particle swarm optimization (PSO) to forecast longitudinal cracking. The proposed PSO-GBM model achieved the lowest mean RMSE (2.661) and highest R 2 (0.984) across fivefold cross-validation, outperforming baseline GBM, linear regression, random forest, artificial neural networks (ANN), and support vector regression (SVR). Compared to traditional and untuned models, the PSO-GBM offers improved generalization and a stronger ability to capture nonlinear interactions among variables. Feature importance and sensitivity analyses identified L3 thickness, age, and AADTT as key predictors. Despite the model’s exceptional predictive accuracy, computational demands and data availability may limit its practical application. However, the results offer useful information for transportation organizations looking to improve maintenance planning techniques and incorporate intelligent predictive tools into pavement management systems.