Ensembles are among the most effective and successful methods for almost all supervised tasks. Not long ago, an ensemble approach has been proposed for quantification learning. The idea of such method is to exploit the prior knowledge about quantification tasks, building ensembles in which diversity is achieved by training each model with a different distribution. These training samples are generated taking into account the expected drift in class distribution. This paper extends this method proposing three new quantifier selection criteria particularly devised for quantification problems, where two of them are defined for dynamic ensemble selection. The experiments demonstrate that, in many cases, these selection functions outperform straightforward approaches, like averaging all models and using quantification accuracy to prune the ensemble. Moreover, the results show that performance heavily depends on the combination of the base quantification algorithm and the selection measure.

Ensemble methods are widely applied to supervised learning tasks. Based on a simple strategy they often achieve good performance, especially when the single models comprising the ensemble are diverse. Diversity can be introduced into the ensemble by creating different training samples for each model. In that case, each model is trained with a data distribution that may be different from the original training set distribution. Following that idea, this paper analyzes the hypothesis that ensembles can be especially appropriate in problems that: (i) suffer from distribution changes, (ii) it is possible to characterize those changes beforehand. The idea consists in generate different training samples based on the expected distribution changes, and to train one model with each of them. As a case study, we shall focus on binary quantification problems, introducing ensembles versions for two well-know quantification algorithms. Experimental results show that these ensemble adaptations outperform the original counterpart algorithms, even when trivial aggregation rules are used.