Machine learning for energy band prediction of halide perovskites

Abstract Halide perovskites have emerged as a class of highly promising photovoltaic materials with exceptional optoelectronic properties. The bandgaps of halide perovskites, along with the energy levels of the conduction band minimum (CBM) and valence band maximum (VBM), play a critical role in determining light absorption, interfacial energy alignment, charge carrier dynamics and photovoltaic performance of the corresponding solar cells. Herein, we developed high-accuracy machine learning (ML) models based on state-of-the-art algorithms to predict the CBM, VBM and bandgaps of halide perovskites. We primarily focus on properties calculated using the Heyd–Scuseria–Ernzerhof (HSE) functional. Among the tested ML models, the extreme gradient boosting regression (XGB), which outperformed five other shallow ML models as well as Transformer and multilayer perceptrons models, achieved a coefficient of determination ( R 2 ) of 0.8298 for CBM prediction ( R 2 of 0.8481 for VBM) and a mean absolute error (MAE) of 0.1510 eV (MAE of 0.1490 eV for VBM) on the test set. For HSE-derived bandgaps, the XGB model demonstrated an R 2 score of 0.8008 and an MAE of 0.2848 eV on the test set. In addition to HSE-derived bandgaps, we also incorporated predictions for bandgaps calculated using the Perdew–Burke–Ernzerhof (PBE) functional. For PBE-calculated bandgaps, the XGB model maintained best predictive performance, achieving an R 2 score of 0.9316 and an MAE of 0.1018 eV on the test set. Finally, we conducted shapley additive explanations analysis based on the optimal models to identify the key features influencing energy band properties of halide perovskites. Our findings statistically revealed the dominant factors affecting bandgaps, CBM and VBM energy levels in halide materials, which aligned with previous non-ML studies. This work provides meaningful insights for the rational design of halide perovskites with tailored energy band properties.