摘要： 针对覆盖算法中识别精度与泛化能力存在的矛盾，在信息论观点的Rough集理论基础上，提出覆盖熵概念，以决策属性相对于分类器的条件信息熵为约束条件，在确保算法分类能力不降低的情况下，对一组覆盖中信息熵最大的覆盖进行约简，减少了分类器的不确定因素。实验结果证明，该算法具有很好的识别精度与泛化能力，对模糊、不确定的数据也具有较好的处理能力。

关键词: 覆盖熵, 覆盖约简, 条件信息熵

Abstract: Aiming at the conflict between validity and extensibility of the covering algorithm, on the basis of information view of rough set theory, this paper presents the concept of covering entropy. It takes the conditional information entropy of decision attribute to classifier as constraint condition. Under the condition of ensuring the classified ability of the algorithm, the uncertainty of the classifier is decreased by reducing the covering which has the largest covering entropy in a group of coverings. Experimental results prove that the reducing covering algorithm based on conditional information entropy has a good validity and extensibility, and has a good ability of dealing with fuzzy and uncertain data.

Key words: covering entropy, coverage reduction, conditional information entropy

中图分类号: 

• TP311