In this study, the performance of lead-free double perovskite solar cells based on \(\mathrm {Cs_2AgBi_{0.75}Sb_{0.25}Br_6}\) is systematically investigated using SCAPS-1D device simulations, employing PCBM as the electron transport layer (ETL) and CFTS as the hole transport layer (HTL). The baseline device delivers a power conversion efficiency (PCE) of 17.59%, with an open-circuit voltage ( \(V_{oc}\) ) of 1.02 V, a short-circuit current density ( \(J_{sc}\) ) of 24.23 mA/cm \(^2\) , and a fill factor (FF) of 71.17%. A comprehensive parametric analysis is performed by varying transport-layer thicknesses, dopant concentrations, and absorber defect density to evaluate their influence on the photovoltaic response. Based on this analysis, a dataset of 256 samples is generated and used to train several machine learning (ML) models for efficiency prediction. Among the evaluated models, the Random Forest regressor demonstrates the best performance, achieving a coefficient of determination ( \(R^2\) ) of 0.997 and a root mean square error (RMSE) of 0.0565. The trained model is subsequently coupled with a differential evolution algorithm to identify an optimal device configuration, predicting a maximum PCE of 23.05%. Validation through SCAPS-1D simulations using the optimized parameters yields a PCE of 22.61%, with simultaneous improvements in \(V_{oc}\) , \(J_{sc}\) , fill factor, and reduced hysteresis. The close agreement between machine learning predictions and physical simulations confirms the reliability of the proposed framework and highlights the effectiveness of integrating device simulation with data-driven optimization for advancing high-performance lead-free perovskite solar cells.
