Attribute Reduction Approach to Neighborhood Decision Agreement

Number of views： 11

分享到:

摘要:

基于邻域决策错误率的属性约简可以在删除冗余属性的同时,提升邻域分类器的留一验证分类精度.但这种约简方式并未充分考虑邻域分类结果在约简前后的差异.为解决这一问题,借助联合分布矩阵,提出了邻域决策一致性的概念,构建了邻域决策一致性与邻域分类精度的调和平均值,并将其作为约简求解的度量准则.在12个UCI数据集上的实验结果表明,所提出的新约简不仅能够有效地提升邻域分类器的决策一致性,而且在多数情况下能够进一步提高邻域分类器的留一验证分类精度.

Attribute reduction based on neighborhood decision error rate can improve the leave-one-out classification accuracy of neighborhood classifier via deleting redundant attributes.Nevertheless,such approach does not fully take the difference between classification results before and after reduct into account.To solve such problem,from the viewpoint of joint distribution matrix,the neighborhood decision agreement is proposed and a new criterion for attribute reduction is constructed,which is the harmonic mean of neighborhood decision agreement and neighborhood classification accuracy.The experimental results on 12 UCI data sets show that the new criterion based reduct can not only improve the decision agreement of neighborhood classifier,but also the leave-one-out classification accuracy of neighborhood classifier will also be increased in most cases.

作者:

李智远杨习贝徐苏平陈向坚王平心

机构地区:

江苏师范大学科文学院江苏科技大学计算机科学与工程学院南京理工大学经济管理学院江苏科技大学数理学院

出处:

《betway官方app 学报：自然科学版》 CAS 北大核心 2017年第5期68-73,共6页

基金:

国家自然科学基金(61572242 61503160 61502211 61471182) 江苏省高校哲学社会科学基金(2015SJD769) 中国博士后科学基金(2014M550293) 江苏省青蓝工程人才项目

关键词:

邻域分类器邻域决策错误率邻域决策一致性约简

neighborhood classifier neighborhood decision error rate neighborhood decision agreement reduct

分类号: