Estimation of inorganic crystal densities using gradient boosted trees

Abstract

Density is a fundamental material property that can be used to determine a variety of other properties and the material’s feasibility for various applications, such as with energetic materials. However, current methods for determining density require significant resource investment, are computationally expensive, or lack accuracy. We used the properties of roughly ∼15,000 inorganic crystals to develop a highly accurate machine learning algorithm that can predict density. Our algorithm takes in the desired crystal’s chemical formula and generates 249 predictors from online materials databases, which are fed into a gradient boosted trees model. It exhibits a strong predictive power with an R2 of ∼99%.