Utilizing Machine Learning Algorithms and Regression Techniques to Develop Pavement Performance Indices for Dry Freeze Climate Regions

Pavement condition monitoring and road maintenance are crucial in ensuring optimal pavement performance. Accurate prediction of road conditions is essential when allocating funds to ensure better and safer roads that meet performance standards and provide a better driving experience. Two widely used measures for road conditions are the pavement condition index (PCI) and the international roughness index (IRI). Since the collection of IRI data is easier and more cost-effective than collecting pavement distress data, the objective of this study is to use the IRI of flexible pavements to develop PCI models by employing machine learning (ML) algorithms, namely k-nearest neighbors, random forest, decision trees, adaptive boosting and extreme gradient boosting, and ten conventional techniques, namely Linear, Logarithmic, Inverse, Quadratic, Cubic, Compound, Growth, Power, S-curve and Exponential Regression. The data used in this study is from the long-term pavement performance program in regions with dry climates in the United States and Canada. The ML models showed excellent predictive capabilities for the dataset, as evidenced by the high coefficients of determination of 96.5%, 97.5%, 98%, 98.4% and 98.5%. Root mean square errors of 4.261, 3.757, 3.156, 3.053, and 2.913, and mean absolute errors of 2.694, 2.375, 1.726, 1.680 and 1.657, considering the five ML algorithms, respectively. This proactive maintenance and planning approach demonstrates the value of M transportation infrastructure management since it improves efficiency and reduces costs while enhancing safety.