Machine learning prediction of magnetic properties of Fe-based metallic glasses considering glass forming ability

Fe-based metallic glasses (MGs) have shown great commercial values due to their excellent soft magnetic properties. Magnetism prediction with consideration of glass forming ability (GFA) is significance for developing novel functional MGs. However, theories or models established based on condensed matter physics exhibit limited accuracy and some exceptions. In this work, 618 MGs samples collected from published works, machine learning (ML) were well trained predict saturated magnetization (Bs) GFA was treated as a feature using the experimental data supercooled liquid region (ΔTx). Three ML algorithms, namely eXtreme gradient boosting (XGBoost), artificial neural networks (ANN) random forest (RF), studied. Through selection hyperparameter tuning, XGBoost showed best predictive performance randomly split test dataset determination coefficient (R2) 0.942, mean absolute percent error (MAPE) 5.563%, root squared (RMSE) 0.078 T. A variety importance rankings derived by that ΔTx played an important role in models. This work proposed method can simultaneously aggregate other features thermodynamics, kinetics structures properties accuracy.