Interpretable hardness prediction of high-entropy alloys through ensemble learning

With the development of artificial intelligence, machine learning has a wide range of applications in the field of materials. The sparsity of data on the mechanical properties of high-entropy alloys makes it difficult to balance between the generalizability and interpretability in data-driven predictive models of material properties. A machine learning model was established based on the HEA hardness data of the Al-Co-Cr-Cu-Fe-Ni system, and several modeling features were screened out through a three-step parallel approach. Model ensemble was performed for RandomForest, XGBoost, LightGBM and CatBoost using the stacking ensemble algorithm, and the coefficient of determination(R²) of the model reached 0.93 after a ten-fold cross-validation. The ensemble learning is stable and accurate for predicting HEA hardness value, and is experimentally verified. The model and selected features can also be applied to different HEA systems as well as low hardness CrFeNi MEA. In addition, we further explained the large prediction deviation of MEA in the high hardness region. Further, the effects of HEA composition and phase formation on the hardness of HEA were qualitatively analyzed based on interpretable tools like SHAP values as well as PDP/ICE plots, respectively. Finally, the model not only has the generalization of ensemble learning, but also has certain interpretability.