Machine learning stability and band gap of lead-free halide double perovskite materials for perovskite solar cells

Abstract Perovskite solar cells have risen since 2013, which are urgently longing for lead-free perovskite materials discovery. Here, we propose a machine learning framework to investigate thermodynamic stability and band gap of lead-free halide double perovskites at high speed and high precision, analyze the importance of selected features and provide directions for discovering potential lead-free perovskites. Four different machine-learning algorithms are utilized, including random forest, ridge regression, support vector regression and XGBoost. XGBoost provides the highest predictive performance (R2:0.9935 and MAE:0.0126) for thermodynamic stability. Random forest provides the highest prediction performance (R2:0.9410 and MAE:0.1492) for band gap. Key features are extracted for exploring hidden structure-properties relationships. Thermodynamic stability and the most important feature of electronegativity are linearly correlated, and XGBoost performs best. Band gap and the extracted features of highest occupied energy level (hoe_b1) and cubic phase are non-linearly correlated, and random forest can well capture the non-linearly. This work demonstrates a great potential of machine learning for accelerating perovskite solar-cell materials discovery.