Analysis of Factors Influencing the Penetration Value of Polyethylene-Modified Bitumen Using a Machine Learning Approach

Abstract

This study presents an advanced machine learning-based approach for analysis of factors affecting the penetration of polyethylene(PE)-modified asphalt binders. Acomprehensive experimental dataset comprising 204 samples was synthesized from previous studies, covering seven critical input variables: PE content, melting point, PE type, mixing temperature, mixing time, mixing speed, and the penetration of the base bitumen. The performance of five machine learning algorithms (ExtraTrees, LGBM, Gradient Boosting, Bagging, and AdaBoost) was evaluated through 30 Monte Carlo simulations and 5-fold cross-validation to ensure stability and objectivity. The results indicate that the ExtraTrees model achieved superior performance, with a coefficient of determination R2= 0.982 and a root mean square error RMSE = 6.473 on the testing dataset. Explainable artificial intelligence techniques, utilizing Partial Dependence Plots (PDP), elucidated the non-linear interaction mechanisms of the variables; specifically, the base asphaltpenetration (X7) and PE content (X1) were identified as the two most decisive factors influencing the penetration of PE-modified asphalt binder. Finally, an intuitive user interface was developed to provide a rapid and accurate prediction tool, facilitating the optimization of the mix design process for modified asphalt binders in practice.