Developing Hybrid Machine Learning Models to Determine the Dynamic Modulus (E*) of Asphalt Mixtures Using Parameters...

Area of Science:

• Civil Engineering
• Materials Science
• Computational Intelligence

Background:

• Accurate characterization of asphalt mixture dynamic modulus (E*) is crucial for pavement design and performance evaluation.
• Empirical models often lack the precision required for complex material behaviors.

Purpose of the Study:

• To evaluate the efficacy of six machine learning (ML) models optimized by a novel modified beetle antennae search (MBAS) algorithm for predicting asphalt mixture E*.
• To compare the performance of these ML models against traditional empirical models.

Main Methods:

• Six ML models (BP, SVM, DT, RF, KNN, LR) were employed.
• A newly developed MBAS algorithm was utilized for hyperparameter tuning of the ML models.
• Model performance was assessed using statistical coefficients and Monte Carlo simulation, focusing on Root Mean Square Error (RMSE).

Main Results:

• The MBAS algorithm demonstrated satisfactory hyperparameter tuning for the ML models.
• Fast convergence and significantly lower RMSE values were achieved by five ML models (BP, SVM, DT, RF, KNN).
• The Random Forest (RF) model exhibited the highest accuracy, efficiency, and robustness in predicting asphalt mixture E*.

Conclusions:

• The MBAS-optimized ML models show significant potential to replace traditional empirical models for asphalt mixture E* characterization.
• The Random Forest model, optimized via MBAS, is a highly accurate and robust tool for predicting asphalt mixture dynamic modulus.