SCAPS-1D simulation and machine learning prediction of photovoltaic performance in K2AgSbCl6 lead-free double perovskite solar cells

In this study, the non-toxic, lead-free double perovskite K 2 AgSbCl 6 is investigated as an efficient photovoltaic absorber through an integrated framework combining SCAPS-1D device simulations and machine learning (ML) modeling. Among the evaluated ML algorithms, CatBoost demonstrated the highest predictive accuracy, achieving a testing R 2 of 0.9795, a testing RMSE of 0.5261, and a 5-fold cross-validation score ( R 2 ) of 0.9651, closely followed by XGBoost. SHAP feature attribution analysis revealed that the acceptor density and the absorber layer/electron transport layer (AL/ETL) interface defect density are the primary factors influencing device performance. The optimized solar cell architecture, ITO/C 60 /K 2 AgSbCl 6 /CBTS/Pt, achieved a short-circuit current density ( J s c ) of 17.0818 mA/cm 2 , an open-circuit voltage ( V o c ) of 1.3868 V, a fill factor (FF) of 81.22%, and an overall power conversion efficiency (PCE) of 19.24%. These findings highlight the potential of data-driven optimization strategies for accelerating the development of high-performance lead-free double perovskite solar cells.