ABSTRACT
This study employs Machine Learning (ML) techniques to optimize the performance of Perovskite Solar Cells (PSCs) by identifying the ideal materials and properties for high Power Conversion Efficiency (PCE). Utilizing a dataset of 3000 PSC samples from previous experiments, the Random Forest (RF) technique classifies and predicts PCE as the target variable. The dataset includes various features encompassing cell architecture, substrate materials, electron transport layer (ETL) attributes, perovskite characteristics, hole transport layer (HTL) properties, back contact specifics, and encapsulation materials. ML-driven analysis reveals novel, highly efficient PSC configurations, such as Fe2O3/CsPbBrI2/NiO-mp/Carbon, CdS/FAMAPbI3/NiO-C/Au, and PCBM-60/Phen-NaDPO/MAPbI3/asy-PBTBDT/Ag. Additionally, the study investigates the impact of crucial parameters like perovskite bandgap, ETL thickness, thermal annealing temperature, and back contact thickness on device performance. The predictive model exhibits high accuracy (86.4 % R2) and low mean square error (1.3 MSE). Notably, the ML-recommended structure, SnO2/CsFAMAPbBrI/Spiro-OmeTAD/Au, achieves an impressive efficiency of around 23 %. Beyond performance improvements, the research explores the integration of ML into the manufacturing and quality control processes of PSCs. These findings hold promise for enhancing conversion yields, reducing defects, and ensuring consistent PSC performance, contributing to the advancement of this renewable energy technology.