Automated machine learning guides discovery of ABO3-type oxides for effective water splitting photocatalysis

Abstract

The search for suitable perovskite oxides for water splitting is challenging due to their vast compositional space. This study employs the TPOT automated machine learning approach to predict the photocatalytic properties of 5329 ABO₃-type perovskite oxides based on 14 features. The process streamlines the steps typically associated with conventional machine learning, reducing computational time by 90 % compared to DFT and narrowing the screening scope. Regression and classification models were developed to predict band edge positions and band gap types. Following TPOT optimization, the prediction error was reduced by 42.4 %. Finally, 57 candidate materials were identified, providing potential for experimental synthesis.