Machine Learning Predictions of High-Curie-Temperature Materials

TL;DR

This paper develops machine learning models, particularly random forests, to predict high Curie temperature magnetic materials based solely on chemical composition, aiding the discovery of better room-temperature magnets.

Contribution

It introduces the use of large experimental datasets and compares multiple models and descriptors, identifying the most effective approach for predicting Curie temperatures.

Findings

Random forest models outperform k-NN in prediction accuracy.

Combining datasets improves model performance.

Systematic errors reveal over-prediction of low-$T_C$ materials.

Abstract

Technologies that function at room temperature often require magnets with a high Curie temperature, $T_\mathrm{C}$, and can be improved with better materials. Discovering magnetic materials with a substantial $T_\mathrm{C}$ is challenging because of the large number of candidates and the cost of fabricating and testing them. Using the two largest known data sets of experimental Curie temperatures, we develop machine-learning models to make rapid $T_\mathrm{C}$ predictions solely based on the chemical composition of a material. We train a random forest model and a $k$-NN one and predict on an initial dataset of over 2,500 materials and then validate the model on a new dataset containing over 3,000 entries. The accuracy is compared for multiple compounds' representations ("descriptors") and regression approaches. A random forest model provides the most accurate predictions and is not improved by dimensionality reduction or by using more complex descriptors based on atomic properties. A random forest model trained on a combination of both datasets shows that cobalt-rich and iron-rich materials have the highest Curie temperatures for all binary and ternary compounds. An analysis of the model reveals systematic error that causes the model to over-predict low-$T_\mathrm{C}$ materials and under-predict high-$T_\mathrm{C}$ materials. For exhaustive searches to find new high-$T_\mathrm{C}$ materials, analysis of the learning rate suggests either that much more data is needed or that more efficient descriptors are necessary.