Machine Learning-Based Prediction of Complex Shear Modulus of Polymer-Modified Bitumen Aged Under Modified TFOT Conditions

The ageing of polymer-modified bitumen (PMB) significantly affects its rheological performance and service life in asphalt pavements. In this study, experimental data PMB 25/55–60 aged under a modified Thin Film Oven Test (TFOT) were restructured into a tidy dataset and analyzed using machine learning techniques. The input variables included temperature, angular frequency, and ageing condition, while the output variable was the complex shear modulus (G*). Two state-of-the-art regression models, Random Forest (RF) and Gradient Boosting Regressor (GBR), were trained and evaluated. Performance assessment revealed that GBR outperformed RF, achieving R² = 0.992, MAE = 1.07 × 10⁶ Pa, and RMSE = 2.04 × 10⁶ Pa, compared to RF with R² = 0.962. Condition-wise analysis further confirmed the robustness of GBR across different TFOT scenarios. Feature importance analysis identified temperature as the dominant factor influencing rheological behavior, followed by frequency and ageing condition. These findings demonstrate the potential of gradient boosting approaches for accurately predicting the rheological properties of aged PMB, providing a reliable tool for performance evaluation and supporting the development of predictive frameworks for pavement materials.