Interpretable machine-learning strategy for soft-magnetic property and thermal stability in Fe-based metallic glasses

Fe-based metallic glasses (MGs) have been extensively investigated due to their unique properties, especially outstanding soft-magnetic performance. Conventional design of soft-magnetic Fe-based MGs relies on trial-and-error and struggles to balance saturation flux density (Bs) and thermal stability because of the strong interplay between glass formation and magnetic interactions. Here, an interpretable eXtreme Gradient Boosting (XGBoost) machine-learning model based on intrinsic elemental properties (atomic size, electronegativity, valence electron count) predicts Bs and Tx (onset crystallization temperature) with 93.0% and 94.3% accuracy, respectively. Key features dictating Bs and Tx are derived from the model, revealing physical origins underlying high Bs and thermal stability. Guided by the model, several Fe-based MGs with high Tx (>800 K) and high Bs (>1.4 T) were developed. This demonstrates that an interpretable XGBoost approach can extract decisive parameters for Fe-based magnetic MGs and enable efficient design of high-performance glassy materials.