【摘要】 传统的基于粗集的属性约简须计算差别矩阵并生成大量的条件属性类 ,效率低 ,且很多算法还不完备 .为此 ,本文引入分类关联规则和相容分类关联规则的概念 ,给出基于分类关联规则的求解下近似和正区域的等价方法 ,从而提出基于分类关联规则的属性约简模型和算法 ,该模型将属性约简问题转化为求解一类特殊的分类关联规则集的问题 ,因而使得相应的算法可有效地改进属性约简挖掘效率 ,克服传统算法依赖于主存的限制 ,为属性约简提供了一种新的框架 .理论分析表明该算法是有效且可行的 .更多还原

【Abstract】 Conventional algorithms for attributes reduction based on rough sets need to compute the time-consuming discernibility matrix and generate lots of attribute classes, thus they are of low efficiency. Moreover, many algorithms are incomplete. In this paper, the concepts of classified association rules and compatible classified association rules are introduced, and equivalent models based on classified association rules for computing lower approximation and positive region are proposed. Furthermore, this paper gives a novel model and a complete algorithm —— EAMAR (efficient algorithm for minimal attributes reduction) for attributes reduction. The model only needs to mine a set of special classified association rules instead of generating lots of attribute classes, so it can effectively overcome the limitation of main memory and solve the problem of time consuming. Therefore, EAMAR provides a new framework for attributes reduction. Theoretical analysis results show that EAMAR is effective and efficient.更多还原