An Overview of Machine Learning Approaches for Predicting Marshall Parameters in Modified Asphalt Mixtures (MAM)

Abstract

Asphalt is a basic material used for pavements and roads construction due to its durability and capacity to endure loads, pressure and stress. To minimize construction cost, asphalt is typically combined with readily available and economically recycled materials. Some of these materials includes glass furnace dross, ashes from municipal waste incineration, crushed bricks, plastic, glass, and crumb rubber, sourced from waste tires. Evaluating Marshall parameters after modification is crucial to maintaining the original capacity of asphalt to withstand enormous stress. There are several Marshall Parameters of modified asphalt mixture (MAM) but Marshall Stability (MS) and Marshall Flow (MF) are the most critical parameters in evaluating the performance of the MAM. Researchers have repeatedly relied on Marshall test in the laboratory to determine the Marshall properties of MAM, which have proved to be expensive, labor-intensive, and tedious. Numerous studies have proposed machine learning (ML) models as an alternative to the traditional evaluation method of Marshall parameters of MAM. The popularity of ML models is largely due to their ability to learn patterns and predictive characteristics from complex data. ML models have been successfully used in the prediction of Marshall parameters of MAM with varying degrees of accuracy, as documented in the literature. Consequently, this paper examines the literature on the use of ML models for predicting Marshall Parameters, particularly MS and MF of MAM. This study identified several ML models, such as support vector machine (SVM), K-nearest neighbor, artificial neural networks, and random forest (RF), previously employed in this domain. The study also looked at various performance metrics used in evaluating the predictive accuracy of MS and MF. Some of the metrics include Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R²). The paper also highlighted the potentials of ML models in reducing costs, time and labour as well as improving prediction accuracy. In addition the study also address challenges such as over fitting and the need for more quality and open source datasets. Recommendations for future research include the development of standardized datasets and the exploration of synthetic data to enhance model reliability and generalizability.