Data-driven design for enhanced efficiency of Sn-based perovskite solar cells using machine learning

In this study, a novel three-step learning-based machine learning (ML) methodology is developed utilizing 26 000 experimental records from The Perovskite Database Project. A comprehensive set of 29 features encompassing both categorical and numerical data was utilized to train various ML models for various solar cell performance metrics, including open-circuit voltage (VOC), short-circuit current (JSC), fill factor (FF), and power conversion efficiency (PCE). The model accuracy was assessed using four key metrics: mean absolute error, mean square error, root mean square error, and R2 score. Among the constructed models, random forest (RF) emerged as the standout performer, boasting an R2 score of 0.70 for PCE. This RF model was then used for prediction on the large, optimized design pool of Sn-based perovskite data with intent to probe a viable non-toxic substitute to the standard Pb-based absorber. A three-step algorithm was tailored, which led to the discovery of a new set of feature combinations, showcasing a PCE improvement over the existing peak performance of Sn-based devices. The key aspects identified were device architecture, dimensionality, and deposition procedures for e