Použití symbolické regrese pro Monte-Carlo výběr atributů

Abstract

Master's thesis aim for possible use of symbolic regression for the feature selection. The thesis proposes use of genetic programming for evolving a classifiers. As follows the classifiers are analyzed according to the proposed methods in order to obtain relevance of attributes. Proposed methods are compared to some reference methods for feature selection. Graph representation of feature relations is proposed, partly reflecting significance of the attributes.