Data-driven modeling of punchouts in CRCP using GA-optimized gradient boosting machine

Abstract

Punchouts represent a severe form of structural distress in Continuously Reinforced Concrete Pavement (CRCP), leading to reduced pavement integrity, increased maintenance costs, and shortened service life. Addressing this challenge, the present study investigates the use of advanced machine learning to improve the prediction of punchout occurrences. A hybrid model combining Gradient Boosting Machine (GBM) with Genetic Algorithm (GA) for hyperparameter optimization was developed and evaluated using data from the Long-Term Pavement Performance (LTPP) database. The dataset comprises 33 CRCP sections with 20 variables encompassing structural, climatic, traffic, and performance-related factors. The proposed GA-GBM model achieved outstanding predictive accuracy, with a mean RMSE of 0.693 and an R
2
of 0.990, significantly outperforming benchmark models including standalone GBM, Linear Regression, Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Networks (ANN). The study revealed that traffic variables, particularly Annual Average Daily Traffic (AADT) and Kilo Equivalent Single Axle Load (KESAL), were the most influential predictors of punchouts, followed by structural parameters such as the thickness of the third pavement layer (L3). Climatic variables were found to have comparatively lower impact. These findings offer practical engineering implications, highlighting the importance of traffic loading and structural design in minimizing punchout risk. The developed GA-GBM model not only enhances prediction accuracy but also provides a data-driven foundation for optimizing pavement design and maintenance strategies. This research contributes a robust and interpretable tool to support infrastructure sustainability and extend the service life of CRCP systems.