Numerical and machine learning analysis of Rb2AgInBr6 lead-free double perovskite solar cells

Abstract To address the environmental concerns and stability limitations associated with conventional lead-based perovskite photovoltaics, this study proposes a lead-free double perovskite solar cell based on an optimized $$\hbox {ITO}$$ / $$\hbox {C}_{60}$$ / $$\hbox {Rb}_{2} \hbox {AgInBr}_{6}$$ / $$\hbox {Cu}_{2}\hbox {NiSnS}_{4}$$ / $$\hbox {Pt}$$ heterostructure. A hybrid computational framework combining SCAPS-1D numerical simulation under AM1.5G illumination with machine-learning (ML) analysis was developed to systematically investigate the influence of key device parameters, including absorber thickness, band gap, acceptor density, and defect density, on photovoltaic performance. Among the evaluated regression algorithms, the Random Forest model exhibited the best predictive performance, achieving a cross-validation score of 0.5280, with an $$R^2$$ value of 0.4717 and an RMSE of 4.8109. SHAP-based interpretability analysis was further employed to elucidate the contribution and dependence of the input parameters on the predicted efficiency. Following systematic optimization of the device parameters, the proposed architecture achieved a simulated power conversion efficiency (PCE) of 33.55%, with a short-circuit current density ( $$J_{sc}$$ ) of 34.094 mA/cm $$^2$$ , an open-circuit voltage ( $$V_{oc}$$ ) of 1.2276 V, and a fill factor (FF) of 80.16%. These results demonstrate the potential of integrating SCAPS-1D simulations with interpretable machine-learning techniques as a computational framework for optimizing lead-free double perovskite solar cells and guiding the development of sustainable photovoltaic devices.