Comparing the knowledge quality in rough classifier and decision tree classifier

Abstract

This paper presents a comparative study of two rule classifier; rough set (Rc) and decision tree (DTc). Both techniques apply different approach to perform classification but produce same structure of output with comparable result. Theoretically, different classifiers will generate different sets of rules via knowledge even though they are implemented to the same classification problem. Hence, the aim of this paper is to investigate the quality of knowledge produced by Rc and DTc when similar problems are presented to them. In this case, four important performance metrics are used as comparison, the accuracy of classification, rules quantity, rules length and rules coverage. Five dataset from UCI Machine Learning are chosen and then mined using Rc toolkit namely ROSETTA while C4.5 algorithm in WEKA application is chosen as DTc rule generator. The experimental result shows that Rc and DTc own capability to generate quality knowledge since most of the results are comparable. Rc outperform as an accurate classifier, produce shorter and simpler rule with higher coverage. Meanwhile, DTc obviously generates fewer numbers of rules with significant difference.