Machine-Learning Algorithms for Estimating Rutting Potential and Cracking Tolerance Index of Bituminous Mixtures

The present study aims to estimate Marshall-designed bituminous mixtures’ rutting and cracking resistance using different machine learning algorithms. The BS EN 12597-24 (dynamic creep test) and ASTM D8225-19 (Indirect Tensile Asphalt Cracking Test) standards define the test method for determining bituminous mixture rutting and cracking tolerance, respectively. Traditional tests methods provide fundamental details on the properties of bituminous mixtures, although they can be time-intensive and may require costly repeats. It has prompted interest in ML for more efficient assessments in the field of pavement engineering. These advanced methods show significant potential for predicting critical pavement properties in the paving industry. However, past studies have primarily focused on Superpave-designed bituminous mixtures and often overlooked the performance parameters of Marshall mixes. Also, many previous studies relied on limited data sets that compromised the robustness and applicability of their findings. Furthermore, the quantification of design variables (input data) has not been adequately addressed earlier. To address these concerns, five different ML models were developed for predicting permanent strain (%) and Cracking Tolerance Index (CTIndex) based on the considered design factors. This study employed a distinctive approach for quantification. The results show that the random forest model achieved greater accuracy in predicting permanent strain (%) and CTIndex. This is evidenced by low values for root mean square error, mean absolute percentage error, and mean absolute error, along with a high coefficient of determination, indicating a precise fit and dependable predictions. Moreover, a graphical user interface was developed to increase the model’s usability and facilitate its widespread application.