Abstract: By analyzing of attribute reduction based on the view of information theory in incomplete information system, it brings up a definition of conditional distribution information quantity. This information quantity comes from the distribution that condition attributes determinate tolerance class on the division of decision attribute. At the same time, new attribute significance is defined, and a heuristic attribute reduction algorithm is given. Experimental results shows that this algorithm is effective for attribute reduction in incomplete decision table.

Key words: roughness, incomplete decision table, distribution information quantity, attribute significance

摘要： 通过分析目前信息观下不完备信息系统属性约简，针对已提出的几种信息熵存在随着属性的增加系统分类能力减弱的不足，从条件属性确定的容差类在决策属性划分上的分布出发，给出不完备决策表的条件分布信息量的定义；同时，定义了新的属性重要度，并以此为启发信息设计属性约简算法。通过实验说明了该算法对不完备决策表属性约简是可行的。

关键词: 粗糙性, 不完备决策表, 分布信息量, 属性重要度