Abstract: In order to elucidate the intricate, non-linear association between rolling process parameters and the mechanical properties of the eutectic Mg-8Li alloy, this work proposes a predictive framework that integrates machine learning and explainable analysis. A “rolling parameter-mechanical property” dataset constructed from experimental data is used to develop prediction models for ultimate tensile strength (UTS), yield strength (YS), and elongation (EL) based on six machine learning algorithms: support vector regression (SVR), random forest (RF), XGBoost, AdaBoost, extra trees, and CatBoost, with the Bayesian optimization framework and cross-validation strategy employed to enhance model generalization. The results indicate that the random forest model exhibits the best performance for UTS prediction (coefficient of determination R² = 0.9452, mean absolute error MAE = 2.07), the SVR model achieves the highest accuracy for YS prediction (R² = 0.9524, MAE = 3.38), and the AdaBoost model demonstrates the best performance for EL prediction (R² = 0.8676, MAE = 1.65). Additionally, by using the Pearson correlation coefficient (PCC) along with the Shapley additive explanations (SHAP) method, it is shown that the initial thickness negatively affects UTS and YS, while the reduction ratio positively impacts UTS and YS the most, and the rolling temperature has a complex negative effect on EL. Experimental validation confirms the strong generalization ability and engineering applicability of the optimal models. This work clarifies the synergistic mechanism of rolling parameters through a data-driven approach, providing theoretical support and technical pathways for achieving an intelligent rolling process design for the eutectic Mg-8Li alloy.

Key words: Machine learning, Mg-8Li alloy, Rolling process, Mechanical properties, Shapley additive explanations