首页 | 官方网站   微博 | 高级检索  
     

决策表分析的统计依据
引用本文:魏玲,张文修.决策表分析的统计依据[J].计算机科学,2005,32(4):19-21.
作者姓名:魏玲  张文修
作者单位:1. 西北大学数学系,西安,710069;西安交通大学理学院,西安,710049
2. 西安交通大学理学院,西安,710049
摘    要:给出了决策表的条件属性约简的非参数统计检验方法。首先,给出与决策表相应的列联表,进行条件属性与决策属性间相关性的显著性检验,在一定的显著性水平上,依据相关性显著与否,来判别该属性相对于决策行为是否冗余,从而获得属性约简;进而,来用Lanmbda系数对与决策属性显著相关的属性进行相关性度量,说明用条件属性对决策属性进行预测将消减误差的比例。并在列联表的基础上,获得决策表的一级规则。病例决策表的实验表明,该方法简单,有效。

关 键 词:决策表  列联表  相关性度量  显著性检验  一级规则

Statistical Evidence for Decision System Analysis
WEI Ling,ZHANG Wen-Xiu.Statistical Evidence for Decision System Analysis[J].Computer Science,2005,32(4):19-21.
Authors:WEI Ling  ZHANG Wen-Xiu
Affiliation:WEI Ling,ZHANG Wen-Xiu Faculty of Science,Xi'an Jiaotong University,Xi'an 710049 Department of Mathematics,Northwest University. Xi'an 710069
Abstract:The nonparametric test method is introduced to the reduction of decision table, which gives the statistical evidence for decision system analysis. Firstly, we transfer the raw decision system into a series of corresponding con- tingency table between each condition attribute and decision attribute; then, test the correlation significance between them to make sure if a condition attribute is useless and should be deleted. Secondly, for those condition attributes that are correlated with decision attribute, we do correlated measure using Lambda coefficient. The set of condition at- tributes that are correlated with decision attribute is regarded to be the reduction of condition attributes, and the cor- responding Lambda coefficient shows the proportionate reduction in error. At the same time, we get the first-order rules of the raw decision table based on the series of corresponding contingency table. Our test shows that our method is feasible and efficiently.
Keywords:Decision table  Contingency table  Correlated measure  Significance test  First-order rule
本文献已被 CNKI 维普 万方数据 等数据库收录!
点击此处可从《计算机科学》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司    京ICP备09084417号-23

京公网安备 11010802026262号