A unified machine learning model for hardness prediction of metallic glasses and high entropy alloys

Accurate prediction of mechanical properties is key to accelerating the discovery and optimization of advanced structural alloys. Although machine learning (ML) techniques have been widely applied to predict hardness in metallic glasses (MGs) and high-entropy alloys (HEAs) individually, a unified framework capable of bridging these two structurally and chemically disordered systems remains lacking. Here, we report a unified ML model for Vickers hardness prediction covering both MGs and HEAs. Based on a curated dataset of 940 alloy samples, a three-stage feature selection pipeline identified five core physical descriptors from 12 initial candidates, after which Extra Trees was selected as the optimal model among 13 candidate algorithms. The unified model achieved an R ² of 0.89 under the stringent leave-one-alloy-family-out cross-validation, demonstrating robust generalization to previously unseen alloy families. Shapley Additive Explanations (SHAP) analysis identified atomic size mismatch ( Δ r ) as the dominant factor governing hardness across both alloy classes, with a pronounced hardness-enhancing effect emerging at approximately Δ r > 0.13 . This work demonstrates the feasibility of unified hardness modeling across distinct disordered alloy systems, and offers a data-driven approach for exploring the intrinsic property-governing mechanisms of multi-component disordered materials.