A pairwise label ranking method with imprecise scores and partial predictions

Abstract

In this paper, we are interested in the label ranking problem. We are more specifically interested in the recent trend consisting in predicting partial but more accurate (i.e., making less incorrect statements) orders rather than complete ones. To do so, we propose a ranking method based on pairwise imprecise scores obtained from likelihood functions. We discuss how such imprecise scores can be aggregated to produce interval orders, which are specific types of partial orders. We then analyse the performances of the method as well as its sensitivity to missing data and parameter values. © 2013 Springer-Verlag.