Relative reduct-based estimation of relevance for stylometric features

In rough set theory characteristic features, which describe classified objects, correspond to conditional attributes. A relative reduct is such an irreducible subset of attributes that preserves the quality of approximation of a complete decision table. For a decision table a single reduct or many reducts may exist. In typical processing one reduct is selected for the subsequent generation of decision rules, while others can be discarded. Yet when the set of reducts is analysed as a whole, observations and conclusions drawn can be used to evaluate relevance of attributes, which in turn can be employed in reduction of features not only for rule-based but also connectionist classifiers. The paper describes the steps of such procedure applied in the domain of stylometric processing of literary texts.