关键词: 缺失概率分布的值不确定离散对象, 决策树, 可达概率区间, 条件熵区间

Abstract: This paper studies a decision tree for value-uncertain discrete objects missing probabilities, because it is difficult to obtain the probability distributions over uncertain data in applications. Firstly, the paper defines the (conditional) probability intervals, and proves that the (conditional) probability intervals are the reachable probability intervals. Secondly, based on the reachable probability intervals, it defines the (conditional) entropy intervals, and gives a method to compute the upper and the lower bounds of the (conditional) entropy intervals. Finally, it presents a new decision tree for uncertain data, in which the conditional entropy intervals are used to select the best attributes and objects are assigned to the branches with probability intervals. The decision tree can handle both value-uncertain discrete objects missing probabilities and certain discrete objects. Experiments with uncertain datasets based on UCI datasets show the satisfactory performance.

Key words: value-uncertain discrete objects missing probabilities, decision tree, reachable probability interval, conditional entropy interval