Acquiring a single decision tree representation of majority voting classifiers

摘要

This paper addresses two problems in Majority voting classifiers (MVCs) like Bagging: (1) no logical reasoning behind the decision and (2) a large amount of classification time and space for significant accuracy boosting To solve these problems, this paper proposes a method for learning a single decision tree that approximates MVCs. The method learns a DT from the original examples and the meta examples that are generated from each classifier joining MVCs. Experimental results show that the method has similar accuracy to Bagging and that the tree size by the method is as large as the size of two classifiers.