We trained successful neural networks to predict the formation energy and magnetic moment of 2-dimensional ferromagnetic materials of similar structure to Cr2Ge2Te6. We collect elemental data on materials with known and unknown properties from easily accessible sources to screen for viable materials for use in future research. Random forest regressors were used to identify the most important predictors of our target qualities, which perform better on predictions of formation energy than magnetic moment. We predict the properties of 1225 materials that are candidates for further research in two-dimensional magnetism and identify several potential sources of error in our models that can be targeted for further improvement.
"Machine Learning Applications for Materials Science: Predicting Properties of Two-Dimensional Magnetic Materials,"
Macalester Journal of Physics and Astronomy: Vol. 11:
1, Article 13.
Available at: https://digitalcommons.macalester.edu/mjpa/vol11/iss1/13