Machine learning modeling of transverse cracking in flexible pavement

Transverse cracking in flexible pavements poses significant challenges to road infrastructure, impacting durability and increasing maintenance costs. This study addresses the lack of predictive models specifically for transverse cracking by employing machine learning techniques. Utilizing data from the long-term pavement performance (LTPP) program, we analyzed the influence of various factors including pavement structure, environmental conditions, and traffic loads. Descriptive statistical analysis revealed a positive skew in the frequency of transverse cracks, while a heatmap correlation matrix identified pavement age, effective asphalt content, and freeze indices as key contributors to cracking. Machine learning models were evaluated, with the Exponential Gaussian Process Regression (GPR) model demonstrating superior predictive accuracy, achieving an R-squared value of 0.70 and RMSE of 17.50, outperforming other models such as the Cubic SVM. Sensitivity analysis emphasized the linear relationship between pavement age and transverse cracking, highlighting the progressive impact of aging on pavement integrity. This research provides a comprehensive and innovative machine learning-based approach to predict transverse cracking, offering valuable insights for pavement management strategies that aim to optimize maintenance schedules and extend pavement service life, particularly in regions with varying environmental conditions.