Machine learning modeling of pavement performance and IRI prediction in flexible pavement

Accurate forecasting models for the International Roughness Index (IRI) are essential for the efficient management and maintenance of road infrastructure. Because they enable transportation authorities to more effectively distribute resources and improve both road safety and comfort, they are crucial to strategic planning. One of the major challenges in IRI prediction is the variability in pavement conditions and the complexity of influencing factors. This work uses extensive data from the Long-Term Pavement Performance (LTPP) program to forecast the IRI for flexible pavements using machine learning techniques. The majority of pavements have lower IRI values, indicating good quality, according to the first descriptive statistical study, which included histogram visualization. The selection of features for machine learning modeling was then guided by a heatmap correlation matrix, which showed important correlations between several pavement properties and IRI. Key variables including beginning IRI, pavement age, and effective asphalt composition were found using Random Forest feature importance analysis. The practical significance of the initial IRI being the most crucial predictor suggests that early-life pavement quality substantially influences future performance, emphasizing the need for high construction standards and preventive maintenance. Predictive accuracy was assessed using RMSE, MSE, R-squared, MAE, and computational efficiency using a range of machine learning models, such as Linear Regression, Support Vector Machines (SVM), Ensemble Trees, Gaussian Process Regression (GPR), and Artificial Neural Networks (ANNs). With an R-squared value of 0.92 and an RMSE of 0.15, the Rational Quadratic GPR model stood out as having performed better in our investigation. A key achievement of this study is the development of a highly accurate IRI prediction model that surpasses traditional empirical approaches, demonstrating the effectiveness of advanced machine learning techniques. The least successful regression kernel, however, was the Least Squares Regression Kernel. Sensitivity studies were conducted to evaluate the impact of crucial elements, including pavement age, temperature, AADTT, KESAL, Freeze Index, precipitation, and starting IRI. The U-shaped relationship between temperature and IRI revealed that moderate temperatures are optimal for pavement performance, while extreme temperatures lead to increased roughness. The results indicated that these factors had an effect on IRI, with the initial condition having a significant impact on future pavement smoothness.