DISSIMILARITY-BASED BIPOLAR SUPERVISED CLASSIFICATION

Frequently, the set of classes of a supervised classification problem presents an structure related to the specific features of each application context. However, standard classification models does not use to consider such an structure in their learning and reasoning processes. By means of the introduction of a bipolar approach, this paper proposes a revision of the basic notions of supervised classifiers, aimed to extend their generalization power and adaptation to problems with an structured set of classes.