基于联合表示值的特征选择方法  

Collaborative representation score based feature selection method

作　　者：张志武[1] 荆晓远[2,3] 吴飞[2] 

机构地区：[1]南京邮电大学计算机学院,江苏南京210023 [2]南京邮电大学自动化学院,江苏南京210023 [3]武汉大学软件工程国家重点实验室,湖北武汉430072

出　　处：《南京邮电大学学报（自然科学版）》2017年第1期85-89,共5页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金(61073113;61272273);江苏省普通高校研究生科研创新计划(CXZZ12_0478)资助项目

摘　　要：提出了基于联合表示值的过滤型特征选择方法。该方法利用训练样本构造一个图,图的邻接关系和边的权重是由l_2范数最优化问题求解,在迭代的特征选择过程中,根据当前特征的l_2联合表示值对特征进行排序,丢弃最不重要的特征,并用剩余特征重构l_2图。NASA软件缺陷数据库和UCI机器学习数据库上的实验结果表明,该方法比传统的特征选择方法取得更好的性能。A filter-type feature selection method based on collaborative representation score is proposed. A l2 graph is constructed on training samples, and a l2-norm based optimization is used to determine the graph adjacency structure and corresponding graph weights. At each iteration step, the features are sorted according to their current collaborative representation scores, and the least relevant features are discarded and the remainders are used to update the l2 graph. Experimental results on the NASA Metrics Data Pro- gram and UCI Machine Learning Data Sets demonstrate that the proposed method outperforms the repre- sentative feature selection methods.

关 键 词：联合表示值 特征选择 降维 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]