Abstract: The essential characters of continuous attribute discretization are analyzed in rough set theory. The idea that meeting for requirements of decision table in rough set and also the optimization of clustering algorithm is presented. And the heuristic search idea is introduced to solve the NP-Hard search problem of the cluster number setting for every continuous attributes. A new method——MDV search method is represented in the clustering process of SOM network for the continuous attributes discretization. The attribute redundancy rate is defined, and also its improved definition for the factual application. Finally, by the factual process for UCI database, the validity of MDV method is proved.

Key words: Rough set; Attribute discretization; Cluster; SOM

摘要： 分析粗糙集连续属性离散化问题的本质特点，提出满足粗糙集约简指标和优化算法相结合的离散化思想。引入启发式搜索策略，解决属性离散的NP-Hard 问题，建立连续属性SOM 自组织网络聚类的MDV(Maximum Discernibility Value)搜索方法，并给出属性约简的冗余度定义和计算方法。根据实际计算要求，对冗余度的定义进行改进。最后，通过UCI 数据库实例验证了MDV 方法的有效性。

关键词: 粗糙集；属性离散；聚类；SOM