Ordinal classification based on the sequential covering strategy

Abstract

Ordinal classification is a supervised learning problem. The distinctive feature of ordinal classification is that there is an order relationship among the categories to learn. In this paper, we present a fuzzy rule learning algorithm based on the sequential covering strategy applied to ordinal classification. This proposal modifies a nominal classification algorithm, called NSLV, to adapt it to this kind of problems. To take into account the order relationship among the categories, a new fitness function and a new concept of negative examples for a rule are proposed. Moreover, we introduce a new rule evaluation model for ordinal classification problems. Experimental results show that the proposed algorithm offers a better performance compared to other ordinal algorithms.