摘要
针对Rough集中刻画属性分类能力的测度正区域等仅能反映属性可辨识对象集大小,不能反映属性对样本的划分状况影响分类的其它因素的问题,提出了Rough集中度量属性分类贡献能力的综合测度———属性分类粗糙度,对其特性进行了分析,给出了用该测度以及信息增益等分别作为决策树算法选择属性的启发式对UCI几个数据集的挖掘结果.理论分析和实验表明,属性分类粗糙度更全面地刻画了属性对分类的综合贡献能力,且具有计算更为简单等特点.
For the problem that the measures for measuring attribute classification ability in Rough set can only reflect the size of the object set discriminated by attributes but the synthetic contribution ability of attributes for classification, a new synthetic measures --attribute classification rough degree （ACRD） is proposed for measuring attribute classification contribution ability in Rough set. The idea of constructing ACRD and characteristics of ACRD are described. The results of mining for 6 data sets of UCI machine learning repository are given by using ACRD, information gain, information gain ratio and normalized gain as the heuristics of decision tree algorithm respectively. Theoretic analysis and experiments show that ACRD synthetically reflects the contribution ability of attributes for classification and is simpler than the other three measures in calculation.
出处
《系统工程学报》
CSCD
北大核心
2006年第5期508-514,共7页
Journal of Systems Engineering
基金
天津市教委高校科技发展基金资助项目(020714)
天津理工大学科技发展基金资助项目(LG03029)
关键词
ROUGH集
属性分类粗糙度
决策树
分类
测度
rough set
attribute classification rough degree
decision tree
classification
measure
