Evaluating the Effect of Polypropylene Fibers on the Rutting Resistance of Asphalt Binder via Artificial Intelligence Modeling

Abstract

This study presents an advanced artificial intelligence-based approach for predicting the rutting factor (G*/sinδ) of Polypropylene (PP) fiber-modified asphalt, a critical parameter governing the plastic deformation resistance of pavements at elevated temperatures. Utilizing an experimental dataset of 132 samples synthesized from reputable publications, the CatBoost algorithm was implemented and fine-tuned using two hyperparameter optimization strategies-Grid Search and Random Search, integrated with 5-fold cross-validation.The results demonstrate that the optimized CatBoost model achieves superior predictive accuracy, yielding a coefficient of determination (R2)of 0.9967 and a Root Mean Square Error (RMSE) of 0.5720 on the independent testing set. Model-agnostic interpretability analysis via Partial Dependence Plots identified the initial rutting factor of the base asphalt and the fiber content as the two most influential features affecting material performance. Notably, the study delineated an optimal interaction zone between the mixing temperature (160-180°C) and PP fiber dosage (3-5%) to achieve the most effective reinforcement network structure. This research not only validates the efficacy of the CatBoost algorithm in capturing the complex non-linear relationships of composite materials but also provides a robust predictive tool to optimize design protocols and minimize experimental costs in modified asphalt technology.