Prediction of magnetocaloric properties of Fe-based amorphous alloys based on interpretable machine learning

Abstract

This study developed a machine learning model to accurately predict the isothermal magnetic entropy change (-SM) in amorphous alloys, a key parameter for evaluating magnetocaloric performance. Four machine learning algorithms were compared, and the (Extremely Randomized Trees) ETR algorithm demonstrated exceptional performance with an (R-squared) R2 value of 0.90 and a (Mean Absolute Percentage Error) MAPE of 13.31 % on the test set. Feature selection techniques, including Pearson correlation coefficient (PCC) and Recursive feature elimination (RFE), identified a subset of 7 important features: (Applied Field) Mf, δr, ΔH, ΔTm, ΔS, Tm‾, and Ec‾. The Shapley Additive Explanations (SHAP) method provided insights into feature importance and critical values. Design strategies for new alloys, using the FeZrB system as an example, were proposed based on the predictive model. The model's generalization ability was validated on other amorphous alloy systems, showcasing its wide applicability. This research contributes to the field of amorphous alloys and suggests future directions for machine learning applications.