Abstract: Core is an essential component of attribute reduction. Though analyzing the significancy of every attribute in core, it is found that the significanfy of every attribute is different. The attributes with less siginicancy can affect the accuracy of classifier. Therefore, in this paper, Quasi-core is proposed by deleting the attributes with less significancy than most of attributes in core, and a new heuristic attribute reduction algorithm is designed based on the Quasi-core. Experimental result shows that the reduct with less attributes are obtained by the algorithm, and classifier constructed based the reduct has higher classifier accuracy.

Key words: rough set, core, attribute reduction, positive region

摘要： 核是属性约简中的必不可少的部分。通过对核中属性重要程度的差异进行分析，可以发现一些核属性相对于决策的重要度很小，这些属性一定程度上影响了基于约简结果构造的分类器的分类精度。通过将核中一些对决策贡献很小的属性去除，提出了准核的定义，并基于准核构造了一种新的启发式属性约简算法，利用该算法获得的约简中属性数量更少，基于这种约简构造的分类器分类精度更高，实验结果表明了算法的有效性。

关键词: 粗糙集, 核, 属性约简, 正域