Abstract This study presents a machine learning approach to predict the structural number of flexible pavements using subgrade soil properties and environmental conditions. Four algorithms were evaluated, including random forest, extreme gradient boosting, gradient boosting, and K nearest neighbors. The dataset was prepared by converting resilient modulus values into structural numbers using the bisection method applied to the American Association of State Highway and Transportation Officials 1993 design equation. Input variables included moisture content, dry unit weight, weighted plasticity index, and the number of freeze and thaw cycles. Each model was trained and tested using standard performance metrics. Gradient boosting achieved the highest accuracy with a determination coefficient of 0.917. Moisture content was identified as the most significant predictor in most models. The findings demonstrate that machine learning models can accurately predict pavement thickness requirements based on readily available soil and environmental data. This approach reduces reliance on expensive and time-consuming laboratory tests and provides a practical and efficient tool for pavement design. This study highlights the potential of machine learning models in enhancing pavement design by accurately predicting structural performance parameters based on soil and environmental factors.
Asadullah Ziar (Tue,) studied this question.