Interpretable Machine Learning Predicts Glass‐Forming Regions Across Diverse Inorganic Systems

ABSTRACT Accurately predicting the glass‐forming region (GFR) remains a fundamental challenge in materials science, as conventional empirical approaches are inefficient and lack generalizability. Here, we introduce an interpretable machine learning (ML) framework that utilizes only chemical composition to forecast GFRs across diverse inorganic systems. By employing XGBoost and Shapley additive explanations (SHAP), the model identifies glass‐forming compositions with high accuracy ( F 1‐score > 0.89) and reveals the roles of key physicochemical descriptors, such as valence and electronegativity, in controlling glass formation. The approach bridges atomic‐level features (e.g., local cation environments) and microstructural evolution (e.g., network connectivity) to macroscopic glass‐forming ability, as demonstrated in systems like oxyfluoride glasses, where AlF 3 content dictates structural transition. Crucially, the model demonstrates exceptional generalizability, validated across chemically distinct systems including oxides, oxyfluorides, and metallic glasses, highlighting universal principles underlying vitrification. This work provides a robust, transferable platform that not only accelerates the design of industrially relevant glasses, such as quaternary laser glasses and bulk metallic glasses, but also refines the theoretical understanding of glass formation through mechanistic, composition–structure–property insights.