Transforming high entropy alloy designing with predictive ensemble machine learning: A framework for accurate micro-hardness estimation and interpretability

The vast compositional space of high-entropy alloys (HEAs) makes conventional alloy design slow, expensive, and highly dependent on trial-and-error experimentation, highlighting the need for accurate and interpretable data-driven predictive frameworks. In this study, a comprehensive machine learning framework was developed to predict the microhardness of HEAs. A literature-derived dataset comprising approximately 1,900 experimentally reported hardness values spanning over more than two decades was collected. To construct a physics-informed feature space, thermodynamic, structural, and semi-empirical descriptors were generated using the High Entropy Alloy Predicting Software (HEAPs), while principal component analysis (PCA) was employed to reduce feature dimensionality and improve model robustness. Several state-of-the-art machine learning algorithms were systematically evaluated, with XGBoost achieving the highest predictive accuracy (MAE = 18.33, R² = 0.94). Although XGBoost yielded the best test-set performance, the proposed ensemble framework provided greater generalization under repeated 10-fold cross-validation (R² = 0.85), demonstrating reliable hardness prediction across diverse HEA compositions. Experimental validation using newly synthesized alloys demonstrated close agreement between predicted and measured hardness values, confirming the practical applicability of the proposed framework. Furthermore, SHAP and partial dependence plot (PDP) analysis identified the key thermodynamic and electronic descriptors governing hardness, providing valuable physical insights into the composition-property relationships of HEAs. The proposed framework integrates large-scale experimental data, physics-informed descriptors, interpretable machine learning, and experimental validation into a unified workflow, providing a reliable strategy for accelerating the design and optimization of next-generation high-performance high-entropy alloys.