Optimizing the performance of CH3NH3SnBr3–based PSCs with SCAPS-1D and machine learning techniques

Perovskite solar cells (PSCs) are considered among the most promising photovoltaic (PV) technologies of the future due to their excellent optoelectronic features, adjustable bandgaps, and cost-efficient fabrication. Yet, large-scale deployment is hindered by the instability of devices and the toxicity of lead-based absorbers. In this work, we investigate a non-toxic PSC structure, FTO/ZnO/CH 3 NH 3 SnBr 3 /CuI/Au, through numerical simulations combined with machine learning (ML). SCAPS-1D modeling under AM1.5G illumination yielded an efficiency of 30.42%, with V OC of 0.99 V, J SC of 35.53 mA/cm 2 , and FF of 86.28%, reflecting efficient charge transport and reduced recombination. To enhance prediction and optimization, a dataset of 750 cases was generated by varying absorber thickness, absorber acceptor density, HTL acceptor density, and ETL donor density. Four ML techniques; Extreme Gradient Boosting (XGB), Random Forest (RF), K-Nearest Neighbours (k-NN), and Artificial Neural Networks (ANN) were applied to model the nonlinear relationship between these parameters and device efficiency. Among them, XGB delivered the highest accuracy, with R 2 = 99.93%, MSE = 0.0034, and RMSE = 0.0581. Shapley Additive Explanation (SHAP)-based feature importance analysis revealed that absorber and HTL acceptor densities strongly influence efficiency, while ETL donor density and absorber thickness have a lesser effect. Process outline depicted schematically for CH 3 NH 3 SnBr 3 –based PSC. • Development of perovskite solar cell. • Enhanced the efficiency of the FTO/ZnO/CH 3 NH 3 SnBr 3 /CuI/Au solar cell structure. • Prospects of the renewable and sustainable energy sector. • The SCAPS 1D software and ML provides reliable PCE information. • Necessary initiative to assess the different parameters influence.