Abstract
Knowledge granulation as proposed by Zadeh consists in making objects under discussion into classes called granules; objects within a granule are similar one to another to a satisfactory degree relative to a chosen similarity measure. Rough mereology as developed by Polkowski in a series of works is especially suited to tasks of granulation as it does propose a systematic way to construct similarity measures in data sets and offers as well theoretic tools for granule formation in the form of an adaptation of the idea of mereological classes defined by Le´sniewski in his mereology theory. In this article, which extends our contributions to the Special Session on Rough Mereology organized by Polkowski and Artiemjew as a part of the Conference on Rough Sets and Knowledge Technology RSKT 2008, we give a fairly detailed account of basic ideas of rough mereology, a description of basic similarity measures called rough inclusions along with the idea of granulated data sets (granular reflections of data sets); then we follow with the idea on how to construct classifiers from granular data, and finally we present some results of granular classification on real data sets. In what follows, we restrict ourselves to a closed world of a given decision system, leaving aside metaphysical questions of relations between this system and the overwhelming universe of all feasible objects.