Identifying the best ML model for predicting the bandgap in a perovskite solar cell†

Perovskite solar cells (PSCs) have gained attention for their characteristics of high efficiency and commercial viability. However, the efficiency of a PSC depends on various factors. One such important parameter is the bandgap of the active layer as it plays an important role in PSCs with regards to the amount of light absorption. Thus, it influences the overall performance of the solar cell. It is important to predict the bandgap of the active layer in PSCs to achieve an effective fabrication process. In this study, we compared six machine learning (ML) models to predict the bandgap. The models were created using a dataset of 500 devices, such as MAPbI₃, FAPbI₃, CsSnI₃ and CsMAPbI₃, obtained from The Perovskite Database Project. These models were further validated using a different dataset of 50 devices. The models were created using ML methods: random forest, gradient boosting regressor, k-nearest neighbours (KNN), AdaBoost, Gaussian process regressor, and bagging. The feature parameters considered for the models were the A coefficient, B coefficient, and C coefficient, out of various other parameters such as the perovskite dimension, perovskite thickness, perovskite deposition temperature, and perovskite deposition time. The random forest model showed better results compared to other models with a low mean absolute error (MAE) of 0.000775, low mean squared error (MSE) of 0.00000920, and high coefficient of determination (r²) of 0.9994.