Ensemble insights: Unlocking the recombination losses in perovskite solar cells using stacked classifier

Recombination losses in Perovskite solar cells (PSCs) have been triggered/witnessed to improve power conversion efficiency (PCE) through experimental and machine learning (ML) techniques in recent years. ML approaches possess significant potential for enhancing the PSCs efficiency by addressing the recombination losses at band to band (BB), Grain Boundaries (GB), and interface compared to the experimental setup. To maximize this potential, a versatile and optimal ML model is needed, to unveil these recombination losses within PSCs effectively. Herein, diverse ML models implemented, underwent through the best features, and seven-fold cross-validation to ensure generalizability and robustness to address the recombination losses. Furthermore, an optimal hyperparameter was determined through Optuna. Additionally, for the model interpretation, Shapley additive explanations (SHAP) is employed. Among them, the stacked classifier emerged as an exceptionally efficient classifier, achieving a remarkable accuracy of 0.94. In addition, the best model (stacked classifier) is validated with experimental data, revealing a strong agreement between the experimental outcomes and the prediction of the ML model showing an optimal route to address the recombination losses in PSCs. Finally, the web server is developed based on a stacked classifier to provide a user-friendly interface, and easy navigation, enabling the practical application for addressing the recombination losses in PSCs. Thus, the proposed approach improved accuracy by 11 % and correctly identified recombination losses with a 15.48 % improvement at GB and 15.79 % at interface compared to previously reported methodologies.