Impact of Purity Measures on Knowledge Extraction in Decision Trees

Abstract

Symbolic knowledge representation is crucial for successful knowledge extraction and consequently for successful data mining. Therefore decision trees and association rules are most commonly used symbolic knowledge representations. Often some sorts of purity measures are used to identify relevant knowledge in data. Selection of appropriate purity measure can have important impacton quality of extracted knowledge. In this paper a novel approach for combining purity measures and thereby altering background knowledge of extraction method is presented. An extensive case study on 42 UCI databases using heuristic decision tree induction as knowledge extraction method is also presented.