基于优势-等价关系的几种约简及规则抽取方法  

Reductions Based on Dominance-Equivalence Relations and Rule Extraction Methods

作　　者：李艳[1] 孙娜欣[1] 赵津[1] 王华超[1] 

机构地区：[1]河北大学数学与计算机学院河北省机器学习与计算智能重点实验室,保定071002

出　　处：《计算机科学》2011年第11期220-224,共5页Computer Science

基　　金：国家自然科学基金(60903088);河北省自然科学基金(F2009000227;A2010000188;F2010000323);河北省第二批百名优秀人才支持计划;河北大学博士基金资助

摘　　要：考虑了条件属性引入优势关系而决策属性上引入等价关系的不协调目标信息系统。分析了这种基于优势-等价关系的相容约简、最大分布约简及正域约简三者之间的关系。此外,结合劣势关系抽取规则以提高规则的覆盖率,改进了基于优势关系的正域约简抽取规则(PDRIS)的方法。最后给出算例,并在UCI数据集上进行了大量的试验,以与PDRIS进行比较。Considered the inconsistent target information systems which respectively induce dominance relation on condition attribute set and equivalence relation on decision attribute set.Under this dominance-equivalence relation,the relationships were analyzed among three different reductions,compatible reduction,maximum distribution reduction and positive domain reduction.The rule extraction method based on positive domain reduction（PDRIS） was improved by introducing inferior relations together with dominance relations to extract rules with larger coverage.Finally,an example was given to illustrate our rule extraction method,and the experimental results on 13 UCI data were demonstrated to make comparisons between the proposed method and the PDRIS method.

关 键 词：粗糙集 优势关系 等价关系 正域约简 规则抽取 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]