Machine learning based prediction of space group for Ba(Ce0.8-xZrx)Y0.2O3 perovskite-type protonic conductors

Abstract

Space groups of Ba(Ce0.8-xZrx)Y0.2O3 (0 ≤ x ≤ 0.8) perovskite-type protonic conductors of wide temperature range from 25 to 900 °C in dry and wet atmospheres are predicted using various machine learning (ML) models. For the prediction, the data including the chemical formulas of the crystal materials as well as space group obtained via high-temperature X-ray diffraction (HT-XRD) and high-temperature neutron diffraction (HT-ND) between 25 and 1000 °C were inputted. 4 different ML models have been constructed using a Random Forest (RF), a Multi-Layer Perceptron (MLP) and a XGBoost, with different data set using experimental data (907 sets of 41 compositions), Magpie descriptor, and additional Inorganic Crystal Structure Database (ICSD) literature data (110 sets of 18 compositions). In the classification problem of space group prediction and phase diagram mapping, accuracy = 0.940 and F1 score = 0.785 is achieved by RF model with coordination number (C.N.) = 9 of oxygen of A-site cation.