基于信息粒度的不完备信息系统空值补齐算法  

Null Value Estimation Method Based on Information Granularity for Incomplete Information System

作　　者：张霞[1,2] 储尚军[1] 许鸣珠[2,3] 

机构地区：[1]河北经贸大学,河北石家庄050061 [2]北京科技大学信息工程学院,北京100083 [3]石家庄铁道大学机械工程学院,河北石家庄050043

出　　处：《小型微型计算机系统》2011年第4期752-756,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(10972146)资助;河北省自然科学基金项目(E2010001089)资助;河北省教育厅科研计划项目(2009116)资助;河北省科技计划项目项目(10213595)资助

摘　　要：在现实世界中,不完备信息系统大量存在的,信息系统中空值的存在大大增加了信息表的不确定性,信息表无法产生更多潜在的有价值规则.处理不完备信息表的一种做法是先将空值补齐再提取规则,常用的空值补齐算法通常都是根据同属性其他值出现的频率高低估计空值,但是此方法不一定能保证规则的一致性.本文提出一种基于信息粒度的空值补齐方法GRCC,首先根据定义的信息粒度选择信息粒度最大的列,然后由相容类产生空值的属性值范围,最后利用MDL准则确定遗漏项的属性值,如此逐列进行填充直到完成全部信息表的补齐.经过实验,GRCC算法补齐的信息表比其它补齐方法产生的信息表产生更多高可信度和高支持度的规则,降低了信息表的不确定性.In actual life, there are lots of incomplete information systems. The existence of null value of information system greatly increases the uncertainty of information table and the information table can not produce more potencial valuable rules. One of the ways to deal with incomplete information system is to complete the null value first then extract rules. The commonly used algorithm to complete null value is mainly based on the appearing frequency of other values with same attribute to estimate the null value, but this method does not guarantee the consistency of rules. In this paper we propose a null value estimation algorithm GRCC based on information granularity,in-st select the column with maximum information granularity based on the defined information granularity, then produce the range of attribute value by compatible class, fmally determine the attribute value of missing terms by MDL principle. Using the above mentioned steps to fill the column one by one untill the whole information table is completed. By the experiment, we can see that the information table completed by GRCC produce more rules with high reliability and high support comparing with table completed by other methods, the uncertainty of information table is reduced.

关 键 词：不完备信息系统 空值补齐 信息粒度 相容类 最小描述长度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]