A Dual Machine‐Learning Framework for Predicting Phase Stability and Mixing Enthalpy Difference in MSi 2 ‐Type High‐Entropy Refractory Metal Silicides
High‐entropy refractory metal silicides (HERMS) are emerging as promising candidates for ultrahigh‐temperature structural applications, generally crystallizing in either the tetragonal C 11 b or hexagonal C 40 structure. The C 11 b phase is desirable for its higher hardness and fracture toughness, but predicting phase stability across multicomponent design spaces is computationally expensive using density functional theory (DFT). To address this, we propose a dual machine learning (ML) framework for (i) classification of stable crystal structure C 11 b versus C 40, (ii) quantitative prediction of the mixing enthalpy difference ( δ mix ), using only DFT‐free descriptors at inference. Using a dataset of 252 quaternary and quinary compositions, we demonstrate that a reduced descriptor set comprising valence electron concentration (VEC), electronegativity difference (Δ χ ), and atomic size difference ( δ r ) achieves high predictive accuracy for both classification and regression tasks. The classifier was validated using stratified 10‐fold cross‐validation (CV), achieving 97% balanced accuracy, a Matthews correlation coefficient (MCC) ≈ 0.96, and a receiver operating characteristic area under the curve (ROC‐AUC) ≈ 0.99, while the regression model reaches test R 2 ≈ 0.95 for δ mix . This computationally efficient strategy for screening and synthesizing C 11 b ‐stabilized HERMS compositions enables rapid exploration of multicomponent silicide design spaces, allowing researchers to reserve DFT calculations exclusively for ML‐selected candidates.