| Abstract Scope |
In this study, we present an interpretable machine learning framework for rapid screening of magnetic materials suitable for aerospace applications. We analyze 33,668 materials from the North East Materials Database (NEMAD), supplemented with entries from the NovoMag and MagNet databases. Our models predict key magnetic properties - including Curie temperature, coercivity, and magnetization - using compositional and crystal structure descriptors. By integrating multiple datasets, we enhance both predictive reliability and interpretability. Among 21,325 materials with reported Curie temperatures, 5,023 (23.6%) exceed 500 K, with an average of 712.8 K, identifying promising candidates for high-temperature aerospace applications. Feature analysis indicates that rare earth and transition metal content are the most influential predictors of thermal stability, while cubic spinel structures achieve the highest average Curie temperature (549.7 K). Random Forest models achieved an R2 score of 0.423 for Curie temperature prediction with mean absolute error of ±135.1 K, providing reliable interpretable guidance. |