Tackling the supervised label ranking problem by bagging weak learners

Preference learning is the branch of machine learning in charge of inducing preference models from data. In this paper we focus on the task known as label ranking problem, whose goal is to predict a ranking among the different labels the class variable can take. Our contribution is twofold: (i) taking as basis the tree-based algorithm LRT described in [1], we design weaker tree-based models which can be learnt more efficiently; and (ii) we show that bagging these weak learners improves not only the LRT algorithm, but also the state-of-the-art one (IBLR [1]). Furthermore, the bagging algorithm which takes the weak LRT-based models as base classifiers is competitive in time with respect to LRT and IBLR methods. To check the goodness of our proposal, we conduct a broad experimental study over the standard benchmark used in the label ranking problem literature.