*5.1.9 Pruned sets (PS)*

PS [24] consists of creating the new training dataset P from the original training dataset D by pruning infrequently label sets. This operation is controlled by a parameter p, which determines how often a label combination must occur for it not to be pruned. This algorithm summarizes this operation:


The pruned instances are reintroduced into the training in the form of new instances with smaller and more commonly found label sets. This allows the preservation of the instances and information about their label set. However, the size of

training dataset is increased, and the average number of labels per instance becomes lower, which can in turn cause too few labels to be predicted at classification time.
