Abstract:
Predicting the bandgap of perovskite absorbers properly is vital for an effective perovskite solar cell design. According to the literature, the bandgap of perovskites is the most dominant parameter affecting solar cell performance. The primary objective of this study was to develop a suitable machine-learning (ML) model to predict the bandgap of any halide-based perovskite absorber. First, three ML algorithms, such as linear regression, random forest regression, and gradient boosting regression, were developed based on the elemental properties of each site present in the perovskites. Different error metrics such as root mean squared error (RMSE), coefficient of determination (R2), and cross-validation scores were calculated to compare their performances. Among the standalone ML models, the gradient-boosting model achieved the best performance. Then, these base models were used to construct the ensemble voting regression model utilizing weighted averages according to the base models’ performances. The voting regressor model outperformed the three baseline models, with the lowest RMSE (0.076) and highest fitting accuracy of R2 (0.95). To reduce the toxicity of lead in MAPbI3, three potential replacements of Ge, Sn, and Si were considered for bandgap prediction and Sn based MAPb1-xSnxI3 was found the best. Finally, a suitable composition of MAPb0.75Sn0.25I3 from the predicted dataset and pristine MAPbI3 was selected for SCAPS-1D analysis. It was found that pristine MAPbI3 performed well in a single junction solar cell with the highest power conversion efficiency of 21.65%. However, replacing the Pb site with 25% Sn reduced the performance showing maximum efficiency (PCE) of 18.58%. The MAPb0.75Sn0.25I3 composition needs further attention to enhance its performance. This study gives an effective tool for the experimentalists to make the task of predicting bandgap efficient, making the whole design process smooth and easier.