Classifiers based on granular structures from rough inclusions

Lech Polkowski and Piotr Artiemjew.

Granular computing initiated by L.A.Zadeh aims at computing with granules of knowledge i.e. with classes of objects similar with respect to a chosen representation of knowledge. It is assumed that classes of sufficiently similar objects behave satisfactorily similarly in solutions of problems of decision making, classification, fusion of knowledge, approximate reasoning. In this work, we support this assumption with a study of classifiers based on granular structures. Granules of knowledge are defined here by means of rough inclusions as proposed by Polkowski and from granules computed in this way in a data set, granular reflections are produced which are basis for inducing classifiers. We demonstrate a few basic rough inclusions and we show that classifiers obtained from granulated according to them data sets yield results better than the standard exhaustive rough set classifier.

PDF full paper