Performance prediction and analysis of perovskite solar cells using machine learning

The conventional way to develop perovskite solar cells (PSCs) is generally based on trial and error and time-consuming synthesis methods. This motivates the adoption of machine learning (ML) models for performance prediction of PSCs. In this work, four ML models have been chosen out of 24 prediction models created to forecast open circuit voltage (Voc), short circuit current density (Isc), fill factor (FF), and power conversion efficiency (PCE) of PSCs. The prediction model derived from Multi-layer Perceptron algorithm demonstrates the highest level of accuracy and RMSE values for predicting PCE, Voc, Isc, and FF are as low as 0.58 %, 0.054 V, 1.01 mA cm−2 and 0.021, respectively. Through Shapley Additive exPlanations theory, the factors affecting the performance parameters of PSCs are analysed. Among 15 distinct features, hole mobility of hole transport layer, electron mobility of electron transport layer, formamidinium in cations and Br in anions, grain size and band gap of the perovskite absorber play the most vital role in improving the performance of PSCs. Herein, four new distinct attributes: grain size, tolerance factor, and electron and hole mobility values of perovskite absorber layer, have been included to the dataset and analysed to predict the performance of PSCs. These results suggest that ML techniques effectively forecast the performance of PSCs and minimize the synthesis cost and time towards the fabrication of efficient cells for commercialization.