高泛化能力的变精度粗糙集的决策树算法  

Decision tree algorithm using variable precision rough set

作　　者：高鹏[1] 王道平[1] 徐章艳[2] 李凡[3] 

机构地区：[1]北京科技大学经济管理学院,北京100083 [2]广西师范大学计算机系,广西桂林541004 [3]华中科技大学计算机科学与技术学院,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2008年第8期74-77,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60463003,60663001)

摘　　要：现有基于变精度粗糙集模型的决策树生成算法具有如下不足:有些叶子结点上覆盖的实例数太少,导致这些叶子结点的泛化能力太小而没有意义;不能很好地处理不一致的实例集.为解决上述问题,引入属性是否具有决策类这一概念,较好地避免了决策树的过剩生长,使得生成的决策树有较好的泛化能力.给出新的终止条件,即时地终止不一致实例集的生长.在此基础上,给出新的终止条,提出了一种新的基于变精度粗糙集的决策树生成算法.用一实例说明了新算法的效率得到提高.The decision tree generation algorithms based on the variable precision rough set model have the following shortcomings. Firstly, some leaf nodes may cover few instances. Thus the generalization ability of these nodes is not so high enough to make sense. Secondly, inconsistent instances can not be processed properly. To deal with the above problems, the definition of decision class was first introduced to solve the problem of over-expanding of decision tree properly. Therefore the ability of generalization of decision tree was also improved. Then a new method, which could stop the expansion of inconsistent instances in time, was also proposed. At the same time, a new decision tree generation algorithm based on variable precision rough set was presented. Finally, an example was used to illustrate the advantages of this new algorithm.

关 键 词：决策树 算法 变精度粗糙集 信息增益 不一致实例集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]