Prediction of formation energies of UCr4C4-type compounds from Magpie feature descriptor-based machine learning approaches

Abstract

Lanthanide-doped UCr4C4-type phosphors have become one of the most promising materials for the narrow-band luminescence, due to their high photoluminescence quantum efficiency and good thermal stabilities. In this study, four machine learning regression models (i.e., the gradient boosted regression (GBR), support vector regression (SVR), random forest (RF) and K-nearest neighbor (KNN)) were established in combination with Magpie feature descriptors in order to calculate formation energies and then study thermal stabilities of phosphor hosts. Thereinto, the GBR model had the best performance (R2: 0.945, MAE: 0.106 eV/atom, MSE: 0.032 eV/atom) and then relatively accurately predicted formation energies of UCr4C4-type compounds. Besides, the Shapley additive explanation (SHAP) was implemented to interpret the prediction of the GBR and analyze the importance of feature descriptors. It is expected that the computing framework in the present work would provide a beneficial guidance for the study of the physical and chemical properties of inorganic phosphors.

Keywords

• UCr4C4-Type compounds
• Formation energies
• Machine learning
• Magpie feature descriptors
• Shapley additive explanation