Accuracy-Complexity Tradeoff Analysis in Data Mining by Multiobjective Genetic Rule Selection

Abstract

In this paper, we demonstrate that multiobjective genetic rule selection improves the accuracy-complexity tradeoff curve of extracted rules in data mining. That is, it significantly decreases the number of extracted rules without degrading their classification accuracy. First we briefly explain heuristic rule extraction and multiobjective genetic rule selection. Then we examine the classification accuracy of extracted rules through computational experiments. Experimental results show that the classification accuracy of extracted rules strongly depends on parameter specifications in heuristic rule extraction. It is also shown that appropriate parameter specifications are problem-dependent. Finally we demonstrate that multiobjective genetic rule selection significantly decreases the number of extracted rules without degrading their classification accuracy.