Abstract
Transverse cracking represents a significant level of distress in Continuously
Reinforced Concrete Pavement (CRCP), which is damaging to the pavement’s
functionality and durability. This study intends to construct a state-of-the-art hybrid
machine learning algorithm that accurately predicts transverse cracking in CRCP by
integrating a Gradient Boosting Machine (GBM) with a Genetic Algorithm (GA). The
analysis comprised 33 CRCP sections using the Long-term Pavement Performance
(LTPP) database with 20 dependent variables covering traffic, structural, climatic, and
performance aspects. The longitudinal and hybrid GA-GBM model was found superior
to baseline models, including standard GBM, Random Forest, Support Vector
Regression (SVR), Linear Regression, and Artificial Neural Network (ANN), with
RMSE of 0.034 and R² of 0.98926. Annual Average Daily Truck Traffic (AADTT),
Kilo Equivalent Single Axle Load (KESAL), and Temperature were found to be the
most impactable, sensitive structural variables, such as Concrete Layer Thickness (L4
Thickness), and Total Thickness. Precipitation and Freeze Index as climatic factors
were also identified to have moderate significance. These results demonstrated the
significant effects of traffic, structural, and climatic elements on transverse cracking.
The findings highlight the GA-GBM model's potential to direct data-driven pavement
management strategies and show how robust and dependable the model is for
predictive modeling. To further improve prediction accuracy and applicability, future
research should broaden the scope by examining sophisticated machine learning
techniques and adding more variables
