Machine Learning-Driven Materials Design and Performance Prediction in Organic Solar Cells Emphasizing Ensemble Learning Models

Organic solar cells (OSCs) have progressed rapidly in recent years, driven by advances in donor polymers, non-fullerene acceptors, and increasingly complex binary and multicomponent blend architectures. Despite these achievements, device performance remains governed by strongly coupled molecular, morphological, and processing variables, making materials optimization inherently multidimensional and difficult to navigate using conventional trial-and-error approaches. The growing availability of experimental data and computational descriptors has therefore encouraged the integration of machine learning (ML) techniques into OSC research as a complementary strategy for accelerating materials discovery and device optimization. Among the available ML strategies, ensemble learning has proven particularly well suited to OSC systems, where datasets are often limited, heterogeneous, and derived from diverse experimental conditions. This review provides a focused and materials-oriented examination of ensemble learning applications in OSC research, spanning donor and acceptor screening, blend optimization, and stability-related prediction tasks. Bagging-based, boosting-based, and stacking-based approaches are discussed in relation to their roles in predicting key photovoltaic metrics, including power conversion efficiency (PCE), open-circuit voltage (Voc), short-circuit current density (Jsc), and related device parameters. The interplay between data sources, descriptor selection, and model performance is critically analyzed, with particular attention to model interpretability through feature-importance evaluation and other explainable learning techniques. Finally, current challenges are discussed, and ensemble learning is positioned as a practical and interpretable tool for accelerating rational OSC materials design.