Evaluating the mechanical behavior of plastic waste modified asphalt using optimized machine learning approaches.

Area of Science:

• Civil Engineering
• Materials Science
• Environmental Science

Background:

• Growing plastic waste necessitates sustainable solutions in pavement construction.
• Recycled plastics in asphalt mixtures present performance prediction challenges due to complex relationships.

Purpose of the Study:

• To develop accurate machine learning models for predicting Marshall Stability (MS) and Marshall Flow (MF) in plastic-modified asphalt.
• To optimize these models using Particle Swarm Optimization (PSO) for enhanced predictive power.

Main Methods:

• Utilized a dataset of 210 plastic-modified asphalt samples.
• Employed machine learning algorithms: Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGB), and Light Gradient Boosting Machine (LGBM).
• Optimized models with Particle Swarm Optimization (PSO) and validated using SHAP, PDP, ICE plots, and Taylor diagrams.

Main Results:

• The PSO-optimized XGBoost model achieved high accuracy, with R² values of 0.82 for MS and 0.83 for MF.
• Identified optimal ranges for plastic particle size (2.5-4 mm), bitumen content (5.3-5.5%), and plastic content (20-30%).

Conclusions:

• The developed models provide reliable predictions for plastic-modified asphalt performance.
• Findings facilitate sustainable waste plastic utilization and cost-effective asphalt mix design for durable pavements.