Harnessing Machine Learning to Predict and Optimize the Performance of MASnI ₃ /FASnI ₃ Double‐Absorber Lead‐Free Perovskite Solar Cells

ABSTRACT

This work applies machine learning (ML) to predict and optimise the performance of a lead‐free MASnI ₃ /FASnI ₃ double absorber perovskite solar cell. SCAPS‐1D simulations were performed by varying absorber thickness and doping densities to generate a dataset for training ML models, including Support Vector Regression (SVR), Random Forest (RF), XGBoost (XGB), CatBoostRegressor and a stacked SVR‐RF ensemble. The optimised device achieved a PCE of 28.14%, a J _SC of 30.74 mA/cm ² , a V _OC of 1.07 V, and a FF of 86% at 0.5 µm MASnI ₃ and 0.3 µm FASnI ₃ thicknesses. Among the models, CatBoostRegressor showed the highest accuracy with R ² = 0.999156 and MSE = 0.002598. Performance evaluation for different training/testing ratios indicated that the 80:20 split gave the best balance between learning and generalisation. A Pearson correlation heatmap and histogram analysis further revealed a strong positive correlation between absorber thickness, J _SC , and PCE, validating the ML predictions. The integrated simulation‐ML approach offers a fast and reliable method for designing high‐efficiency, lead‐free perovskite solar cells.