Reduction from Label Ranking to Binary Classification

This work focuses on a particular application of préférence learning, wherein the problem
is to learn a mapping from instances to rankings over a finite set of labels, i.e. label ranking. Our
approach is based on a learning réduction technique to reduce label ranking to binary classification.
The proposed réduction framework can be used with différent binary classification algorithms in order
to solve the label ranking problem. In particular, in this paper, we présent two variants of this réduction
framework, one where Multi-Layer Perceptron is used as binary classifier and another one where the
Dominance-based Rough Set Approach is used. In the latter, on the one hand it is possible to deal with
possible monotonicity constraints and on the other hand it is possible to provide such a mapping (i.e. a
label ranker) in the form of logical rules: if [antécédent] then [conséquent], where [antécédent] contains
a set of conditions, usually connected by a logical conjunction operator (AND), while [conséquent)
consists in a ranking (linear order) among labels.