Exploring structure-composition relationships of cubic perovskite oxides via extreme feature engineering and automated machine learning

In materials discovery, it is key to explore the structure-composition relationships and machine learning can be used as an effective tool. However, the complexity of conventional machine-learning and the lack of model interpretability make it difficult to derive simple descriptive formulas. Herein, we propose a new method to integrate extreme feature engineering and automated machine learning for exploring structure-composition relationships. Cubic perovskite oxides have attracted huge attention because of their excellent electrical-catalytic properties for electrocatalysts and fuel cells, exhibiting a universal formula of ABO(3) with matching ionic sizes of A-site and B-site cations. The lattice constant is the only fundamental parameter of cubic crystal structure. In this work, we explore the structure-composition relationships between lattice constant and ionic radius of cubic perovskite oxides. Extreme feature engineering is applied to construct new descriptors and an important subset containing nine descriptors is obtained by recursive feature elimination. The optimal descriptor ln(1 +vertical bar r(B)vertical bar) is found and the expression a = 2.2527ln(1 +vertical bar r(B)vertical bar) +2.8181 for the lattice constant a is obtained by linear regression algorithm, only related with B-site ionic radii. Thus, the influence of B-site ionic radii on the lattice constants is much more significant than others. The results indicate that the new method contributes to explore structure-composition relationships without a priori knowledge, accelerates materials design and optimization, and provides a new approach for materials discovery.