A framework for a decision tree learning algorithm with rough set theory

Abstract

In this paper, we improve the conventional decision tree learning algorithm using rough set theory. First, our approach gets the upper approximate for each class. Next, it generates the decision tree from each upper approximate. Each decision tree shows whether the data item is in this class or not. Our approach classifies the unlabeled data item using every decision trees and integrates the outputs of these decision trees to decide the class of unlabeled data item. We evaluated our method using mechanically-prepared datasets whose the proportion of overlap of classes in datasets differs. Experimental result shows our approach is better than the conventional approach when the dataset has the high proportion of overlap of classes and few data items which have the same set of attributes. We guess it is possible to get better classification rules from uncertain and dispersed datasets using our approach. However, we don’t use enough datasets to show this advantage in this experiment. In order to evaluate and enhance our approach, we analyze various and big datasets by our approach.