*5.1.10 Ensembles of pruned sets (EPS)*

The PS method cannot create the new multi-label sets, which have not been seen in the training dataset. Consequently, it presents a problem when working with datasets where labelling is particularly irregular or complex. To solve this problem, an ensemble of PS [24] is proposed. The build phase of EPS is straightforward. Over m iterations, a subset of the training set is sampled and a PS classifier with relevant parameters is trained using this subset. For prediction, the threshold t is used, and different multilabel predictions are combined into a final prediction. This final label set may not have been known to any of the individual PS models, allowing greater classification potential.
