融合特征排序的多标记特征选择算法  

Multi-label feature selection via fusing feature ranking

作　　者：王晨曦[1] 林梦雷[2] 刘景华[2] 王娟[2] 林耀进[2] WANG Chenxi;LIN Menglei;LIU Jinghua;WANG Juan;LIN Yaojin2(Department of Computer Engineering, Zhangzhou Institute of Technology, Zhangzhou, Fujian 363000, China;School of Computer Science, Minnan Normal University, Zhangzhou, Fujian 363000, China)

机构地区：[1]漳州职业技术学院计算机工程系,福建漳州363000 [2]闽南师范大学计算机学院,福建漳州363000

出　　处：《计算机工程与应用》2016年第17期93-100,共8页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61303131;No.61379021);福建省自然科学基金(No.2013J01028);福建省高校杰出青年科研人才培育计划(No.JA14192);漳州市科技项目(No.ZZ2013J04;No.ZZ2014J14)

摘　　要：在多标记学习框架中,特征选择是解决维数灾难,提高多标记分类器的有效手段。提出了一种融合特征排序的多标记特征选择算法。该算法首先在各标记下进行自适应的粒化样本,以此来构造特征与类别标记之间的邻域互信息。其次,对得到邻域互信息进行排序,使得每个类别标记下均能得到一组特征排序。最后,多个独立的特征排序经过聚类融合成一组新的特征排序。在4个多标记数据集和4个评价指标上的实验结果表明,所提算法优于一些当前流行的多标记降维方法。In the framework of multi-label learning, feature selection is a powerful tool for solving the curse of dimensionality,which can improve the classification performance of multi-label classifier. In this paper, a multi-label feature selection algorithm via fusing feature ranking is proposed. First, it conducts adaptive graining samples based on different labels and employs the neighborhood of sample to compute the neighborhood mutual information between feature and label,which can measure the importance degree of feature. Then, all features are sorted in descending order by the value of their neighborhood mutual information under each label. Finally, it acquires a new feature rank by fusing all individual feature rank lists. Experiment is conducted on four data sets, and four evaluation criteria are used to measure the effectiveness. Experimental results show that the proposed algorithm is superior to several state-of-the-art multi-label feature selection algorithms.

关 键 词：特征选择 多标记分类 聚类融合 互信息 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]