关键词: 明确区 非明确区 明确度 深度拟合率

Abstract:

In the process of constructing decision trees based on RST, the common method is to calculate the size of the boundary for every conditional attribute  relative to a decision classification and then choose the attribute with the minimum size of the boundary as the current node of the tree. However, this method only aims to solve the problem of two-valued classification while invalid to the multi-valued one. In this paper, we propose a new concept of e  xplicit region based on RST which will be used as the eriteria for selecting attributes. Among all the candidate attributes, the attribute which has the  largest explicit region will be figured out as the current node according to the whole decision classification. Furthermore, we also propose a new pruning strategy based on explicit region and the method proves to be brief and effective.

Key words: （explicit region, implicit region, explicit degree, depth-fitting ratio）