Comparative Study of Decision Tree with Different Evidential Uncertainty Measures

Abstract

The Decision tree is an effective classifier. In the decision tree, different levels of uncertainty may have to be handled. The theory of belief function is a useful tool to deal with uncertainty. There has been proposed several decision trees based on the theory of belief functions and one of these decision trees used a composite uncertainty measure as the criterion of selecting the appropriate attribute in the process of splitting the node. However, there have been various kinds of uncertainty measures in the theory of belief function, and the existing uncertainty measures may have different influence on the result of classification. In this paper, we use different uncertainty measures in the decision tree based on the theory of belief function to select the appropriate attribute. We provide comparative results and related analyses to check the impact of the uncertainty measure selection on the classification performance of the decision tree.