Machine learning driven prediction of formation energy of AxMyM'zO6 oxides

Abstract

Machine learning has emerged as a powerful tool for discovering new materials, offering significantly lower computational demands compared to traditional density functional theory calculations and experimental approaches. In this work, we apply machine learning to predict formation energies of AxMyM′zO6 oxides using a dataset of 350 compounds with 28 structural, elemental, and electronic descriptors. Four regression models such as Random Forest, Gradient Boosting, Support Vector
Regression, and CatBoost were trained and compared to obtain the accurate values. Among them, CatBoost achieved the highest accuracy (R2 = 0.83 and RMSE = 0.41 eV/atom), outperforming the other approaches. Feature analysis further revealed that electronegativity, ionization energy, and band gap are the dominant factors influencing the stability of AxMyM′zO6 oxides. These results demonstrate the effectiveness of machine learning for fast and reliable prediction of formation energies and
provide valuable guidance for the design of stable oxide materials suitable for energy devices.