Prediction of ferromagnetic materials with high Curie temperature based on material composition information

Abstract

The search for ferromagnetic materials with high Curie temperature (<i>T</i>_c) is a hot issue in condensed matter physics. In this work, an effective machine learning model of Curie temperature based on material component information is established to predict a variety of ferromagnetic materials with high Curie temperature. Based on the collected data of 1568 ferromagnetic materials, and taking the component information of ferromagnetic materials as descriptors, in this work four efficient machine learning models are constructed, namely support vector regression, kernel ridge regression, random forest and extremely randomized trees, through hyperparameter optimization and ten-break cross-validation. Of them, extremely randomized tree model has the best prediction performance, and its cross-validation <i>R</i>² score can reach 81.48%. At the same time, the extremely randomized tree model is also used to predict 36949 materials in the materials project database, and 338 ferromagnetic materials with <i>T</i>_c greater than 600 K are found in this work. The method proposed in this paper can help obtain ferromagnetic materials with high Curie temperature and accelerate the process of ferromagnetic material design.