Machine Learning-Driven Prediction of the Glass Transition Temperature of Styrene-Butadiene Rubber

The glass transition temperature (T_g) of styrene-butadiene rubber (SBR) is a key parameter determining its low-temperature flexibility and processing performance. Accurate prediction of T_g is crucial for material design and application optimisation. Addressing the limitations of traditional experimental measurements and theoretical models in terms of efficiency, cost, and accuracy, this study proposes a machine learning prediction framework that integrates multi-model ensemble and Bayesian optimization by constructing a multi-component feature dataset and algorithm optimization strategy. Based on the constructed high-quality dataset containing 96 SBR samples, nine machine learning models were employed to predict the T_g of SBR and compare their prediction performance. Ultimately, a GPR-XGBoost mixed model was constructed through model ensemble, achieving high-precision prediction with R² values greater than 0.9 on both the training and test sets. Further feature attribution and local effect analysis were conducted using feature analysis methods such as SHAP and ALE, revealing the nonlinear influence patterns of various components on T_g, providing a theoretical basis for SBR formulation design and T_g regulation. The machine learning prediction framework established in this study combines high-precision prediction with interpretability, significantly enhancing the prediction performance of the T_g of SBR. It offers an efficient tool for SBR molecular design and holds great potential for promotion and application.