Modeling of Marshall Stability of plastic-reinforced asphalt concrete using machine learning algorithms and SHAP

Abstract

Pavement engineering has consistently emphasized the need to improve the performance and longevity of asphalt concrete, a critical material in road construction. This study focuses on the predicting the Marshall Stability of plastic-reinforced asphalt concrete using machine learning algorithms. Machine learning algorithms, including Random Forest (RF), Gradient Boosting, Extreme Gradient Boosting (XGB), and Bagging Regressor (BR) were employed to predict the Marshall Stability. The machine learning models were evaluated using six performance metrics, including coefficient of correlation (R), root mean square error (RMSE), mean absolute error (MAE), relative absolute error (RAE), root relative squared error (RRSE), mean absolute percentage error (MAPE), and scatter index (SI), to ensure robust and reliable predictions. Additionally, SHapley Additive exPlanations (SHAP) analysis was conducted to perform a parametric identifying the influence of input parameters on Marshall Stability. Specifically, the XGB model achieved the best R values of 0.95 for training and 0.84 for testing, indicating strong correlations between predicted and actual MS values. Furthermore, SHAP analysis was conducted for the XGB model, which highlighted the significant influence of plastic size and bitumen content on MS prediction. SHAP and machine learning models can be used to optimize the composition of plastic-reinforced asphalt concrete for enhanced performance and sustainability. This research will provide practical guidance for pavement engineers and policymakers in utilizing waste plastic for sustainable infrastructure development.