Machine Learning Models for Efficient Property Prediction\nof ABX<sub>3</sub> Materials: A High-Throughput Approach

Recently, ABX<sub>3</sub> materials have garnered significant attention\ndue to their diverse applications in photovoltaics, catalysis, and\noptoelectronics as well as their remarkable efficiency in energy conversion.\nHowever, progress has been somewhat slow due to the high expenses\nof the experiment or the time-consuming density functional theory\n(DFT) calculation. In this study, we utilized the extreme gradient\nboosting (XGBoost) algorithm to facilitate the discovery and characterization\nof ABX<sub>3</sub> compounds based on vast data sets generated by\nDFT calculations. While the XGBoost algorithm provides a powerful\ntool for accelerating the discovery of ABX<sub>3</sub> compounds,\nit is crucial to acknowledge that different DFT approximation levels\ncan significantly impact the predicted band gaps, potentially introducing\ndiscrepancies when compared with experimental values. In the first\nstep, we predict the space group of 13947 oxides and halides using\nthe Open Quantum Materials Database and elemental features. Our analysis\nyields classification accuracies ranging from 82.39% to 99.14% across\nthese materials. Following this, XGBoost regression algorithms are\nemployed to interrogate the data set, enabling predictions of volume\n(achieving an optimal accuracy of 98.41%, with a mean absolute error\n(MAE) of 2.395 Å<sup>3</sup> and a root-mean-square error (RMSE)\nof 4.416 Å<sup>3</sup>), formation energy (an optimal accuracy\nof 97.36%, with an MAE of 0.075 eV/atom and an RMSE of 0.132 eV/atom),\nand band gap energy (an optimal accuracy of 87.00%, an MAE of 0.391\neV, and an RMSE of 0.574 eV). Finally, these prediction models are\nemployed to identify the possible space groups for each of the 1252\nnew ABX<sub>3</sub> formulas. Then, we predict the volume, the formation\nenergy, and the band gap energy for each candidate space group. Through\nthese predictive models, machine learning accelerates the exploration\nof new materials with enhanced performance and functionality.