Abstract: The attribute reduction and value reduction of rough set are discussed in this paper.The discernibility matrix is extended to value reduction and the attribute significance is redefined based on attribute dependence.A algorithm for classification rule extraction based on discernibility matrix and attribute significance is proposed，which keeps the same classification ability，comes to the smallest attribute reduction and gets the effective rules after the value reduction.Compared with the existed algorithm，less time complexity and less space complexity are acquired with this method.Finally，experiment sets on clothing sales verify the effectiveness of the algorithm.

Key words: discernibility matrix, attribute significance, rule generation, attribute reduction

摘要： 研究了Rough集理论中的属性约简和值约简问题，将分辨矩阵引入值约简中，从属性依赖度的角度重新定义了属性重要度，提出了基于分辨矩阵和属性重要度的分类规则提取算法。该算法在保持分类能力不变的前提下，得到最小属性约简，再经过值约简后得到精确的规则，与现有算法相比，能减少时间和空间耗费。实验结果验证了该算法的有效性。

关键词: 分辨矩阵, 属性重要度, 规则提取, 属性约简