LAUSR

Ensemble learning breaks the bottleneck of weak learners and is usually significantly more accurate than base learners. The overall power conversion efficiency of all-organic dye-sensitized solar cells is difficult to obtain by either calculations or experiments. To achieve high-accuracy models, various ensemble learning methods are investigated. Three types of global ensemble models, including homogeneous and heterogeneous ensembles, are constructed, which outperformed the best single base learner, a support vector machine model (MAE: 0.52; $Q^{2}$ : 0.76); in particular, a novel local heterogeneous ensemble model (MAE: 0.34 and $Q^{2}$ : 0.91) achieved high accuracy and generalization. This paper shows ensemble learning model is capable of exploring complicated quantitative structure activity relationship, where the features are distant from targets.