Exploring the composition - property relationship of magnetostrictive materials: A deep learning driven saturation magnetostriction coefficient predictor

Magnetostrictive materials garner significant attention due to their broad applications in spintronics and wearable devices. Designing materials with tailored saturation magnetostriction coefficients λ_s for specific applications remains a challenging task, primarily due to the complex composition–property relationships. In this study, we develop a deep learning-based predictor capable of accurately mapping material composition to λ_s. To better target RFe₂-type compounds, we refine the existing dataset through data augmentation and cleaning. The resulting model achieves an R² of approximately 0.95 on both training and testing sets, outperforming traditional machine learning approaches with lower root mean square error (RMSE) and mean absolute error (MAE) on testing sets. Furthermore, we examine the contributions of individual elements to the prediction outcomes. This computational strategy enables rapid screening of magnetostrictive compositions, reducing experimental iteration cycles by prioritizing candidates with target λ_s values.