Machine Learning Prediction of Road Performance of Cold Recycled Mix Asphalt with Genetic Algorithm Hyperparameter...

Area of Science:

• Civil Engineering
• Materials Science
• Data Science

Background:

• Cold recycled mix asphalt (CRMA) is a sustainable pavement technology facing performance prediction challenges due to sensitivity to material composition and curing.
• Accurate prediction of CRMA performance is crucial for effective pavement rehabilitation and maintenance.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting critical CRMA performance indicators: dynamic stability (DS) and indirect tensile strength (ITS).
• To identify key factors influencing CRMA performance using interpretable ML techniques.

Main Methods:

• Trained four ML algorithms (ANN, XGBoost, RF, SVR) on a dataset of 436 CRMA samples.
• Optimized model hyperparameters using a genetic algorithm (GA).
• Interpreted model predictions and identified influential factors using the SHAP method.

Main Results:

• GA-optimized XGBoost demonstrated superior predictive accuracy for both DS (R²=0.9793) and ITS (R²=0.9694).
• Curing temperature, recycled asphalt pavement (RAP) content, and curing time were identified as the most significant factors influencing CRMA performance.
• The study established an accurate and interpretable ML framework for CRMA performance prediction.

Conclusions:

• Machine learning, particularly GA-optimized XGBoost, offers a robust approach for predicting CRMA performance.
• Understanding the influence of factors like curing conditions and material composition enables optimized mix design for durable and sustainable pavements.