Discovery and verification of two-dimensional organic–inorganic hybrid perovskites via diagrammatic machine learning model

Abstract

Two-dimensional (2D) organic–inorganic hybrid perovskites (OIHPs) have drawn increased attention due to rich physical properties such as ferroelectricity and photovoltaic properties. Nevertheless, it is challenging to discover novel 2D OIHPs within the vast chemical composition space. Herein, a diagrammatic machine learning model was employed to improve this issue. We collected 179 OIHPs with a variety of organic cations and screened out 6 features from 10,622 descriptors. Subsequently, a decision tree model was created to predict the dimensionality of OIHPs, achieving a LOOCV accuracy of 0.94 and a test accuracy of 0.89, respectively. Then, one candidate from a virtual space with 8256 samples was successfully synthesized, which was consistent with the prediction of the model. Finally, three rules were produced by visualization of the tree structure to generally discriminate 2D from non-2D OIHPs. It is believed that the diagrammatic model has reliability in identifying 2D OIHPs and will serve further property studies of OIHPs in the future.

Keywords

• Machine learning
• Organic-inorganic hybrid perovskites
• Diagram
• Decision tree