Rough Set Approach to Pattern Extraction from Classifiers

Abstract

We discuss the impact of inductive reasoning on the rough set to concept approximation. In inductive reasoning one cannot define inclusion degrees of object neighborhoods directly into the target concepts but only into some neighborhoods relevant to such concepts. Such degrees together with degrees of inclusion of patterns in target concepts make it possible to define outputs of classifiers for new classified objects. We show how among formulas used for classifier construction from decision rules one can search for new patterns relevant for the incremental concept approximation.