Efficient Discovery of Lead-Free A ₃ BX ₆ Halide Perovskites via Machine Learning

Metal halide perovskites have been widely utilized in optoelectronic devices due to their exceptional optoelectronic properties. These properties, along with their potential applications, are fundamentally governed by their bandgap and formation energy. In this study, machine learning (ML) was employed as a pivotal approach to efficiently explore A3BX6 perovskites with high stability and promising photoelectric properties. Among the evaluated 12 ML algorithms, the GBR algorithm demonstrated optimal performance and was selected to predict the bandgap and formation energy. Two regression models, namely B_GBR_BOA for bandgap prediction and F_GBR_GA for formation energy prediction, achieved correlation coefficient (R2) scores of 0.986 and 0.992, respectively, while SHapley Additive exPlanations (SHAP) analysis revealed the corresponding critical features. Meanwhile, 461 potential perovskites with bandgaps in the range of 1–4 eV and formation energies lower than −1 eV/atom were screened out from 2,280 virtual candidates. Furthermore, both density functional theory (DFT) calculations and experimental investigations were carried out to verify the promising predictions of ML. The predicted Rb3BiI6 was successfully synthesized and applied in efficient photodetection and photocatalysis for the first time. This work provides a novel strategy for the efficient discovery of lead-free halide perovskites with promising optoelectronic properties and high stability, facilitating the rational design of high-performance optoelectronic devices.