Innovative Machine Learning Approaches for Predicting the Asphalt Content During Marshall Design of Asphalt Mixtures
Mutahar Al-Ammari, Ruikun Dong, Mohammed Ali Nasser, Abdullah A. Almaswari
Abstract
A flexible pavement with a proper Marshall mix design is essential for ensuring driving longevity, safety, and comfort. The increasing labor demands, costs, and time consumption for evaluating the Marshall mix design properties are due to extensive sample preparation, testing procedures, and material requirements. Consequently, this study aims to compare the conventional method of calculating the optimum asphalt content in Marshall mix design with machine learning approaches. This study focused on identifying the optimal asphalt content through the use of advanced machine learning methods, aiming to improve the accuracy of predicting the performance of asphalt mixtures. Therefore, this research investigates the application of various machine learning-based regression techniques to predict the properties of asphalt mixtures, focusing on evaluating their effectiveness in modeling this complex relationship. The main properties of interest include the Marshall stability, flow, VMA, VFA, and unit weight, all of which adhere to the Marshall mix design. A substantial database comprising 60 datasets was curated to aid in the development of these predictive models. Two stages were carried out in this research. The first stage was focused on determining the ideal asphalt content through conventional techniques, while the second stage involved comparing various algorithms to improve the prediction capabilities for asphalt pavement performance. At the end of the study, the comparisons of the various algorithms for the asphalt mixture parameters revealed that the neural network model outperformed all the others, achieving the highest accuracy based on R2 and MSE values. This highlights the neural network’s effectiveness in capturing the complexities of asphalt mixtures and its superior predictive capabilities compared to conventional methods, emphasizing its advantages in enhancing accuracy and reliability in asphalt mixture analysis.