半监督Relief-F特征选择算法  

Semi-supervised Relief-F Feature Selection Algorithm

作　　者：靳炳烨 王锋[1] 魏巍[1] JIN Bingye;WANG Feng;WEI Wei(School of Computer and Information Technology,Shanxi University,Shanxi Taiyuan 030006,China)

机构地区：[1]山西大学计算机科学与技术学院,山西太原030006

出　　处：《河北师范大学学报（自然科学版）》2023年第4期348-353,共6页Journal of Hebei Normal University：Natural Science

基　　金：国家自然科学基金(62276158);山西省回国留学人员科研资助项目(2021-007)。

摘　　要：数据规模的不断增加,使得为数据库中全部样本做标记变得尤为困难,数据集也因此呈现出了明显的弱标记性.为此,针对大规模少数标记数据集的特征选择问题,基于经典的Relief-F算法,通过综合考虑有标记样本与无标记样本对数据样本近邻的影响,重新定义样本近邻的搜索策略,提出了一种面向符号数据的半监督特征选择算法.为进一步分析新算法的有效性,仿真实验中选取了5组UCI数据集,并引入机器学习中3个常用分类器对新算法和对比算法的特征选择结果的分类性能作了分析和比较,实验结果很好地验证了本文中提出的新算法的有效性和可行性.With the increase of data size,it is very difficult to determine labels for all objects in databases.Data sets present weak markedness as well.Hence,for feature selection on partial labeled data sets,by reference to classical Relief-F algorithm,an effective semi-supervised feature selection algorithm is proposed to deal with partial labeled data.In this algorithm,on the basis of considering labeled samples and unlabeled samples,a kind of new search strategy for finding nearest neighbors is introduced.For analyzing effectiveness of the new algorithm,five UCI data sets and three commonly used classifiers are employed to illustrate classification performance of the new proposed algorithm.The comparison and analysis results show that the new semi-supervised feature selection algorithm is effective and feasible.

关 键 词：特征选择 Relief-F算法 半监督学习 距离度量 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]