摘要
在多标签特征选择中,如果不考虑标签结构信息,只选取与单个标签相关的特征或只选取与整体标签集合相关的特征,则容易选取不重要特征,且也容易遗漏重要特征。为此,提出了一种基于标签组合的多标签特征选择算法——LGMLFS.该算法的主要特点是:(1)考虑标签集合的组结构,并按照标签相关性进行分组。(2)利用标签的相关性结构信息,选取与标签组合相关的特征。(3)度量特征与标签组合的相关性时,不仅考虑特征与组合内各标签的相关性,且同时考虑组合内各标签的重要性。该算法与其他算法对比的实验结果表明,在多个评价指标下,取得了较优的分类性能。
If we do not consider the label structure information and only select the features that are correlated with a single label or the overall label set,we will easy to select the unimportant features or miss the important ones during the process of selecting the features from multi-label datasets.Therefore,we propose a multi-label feature selection algorithm based on label group(LGMLFS).Its has three important characteristics:(1)We consider the structure of the label set and group labels using correlation degree among labels.(2)We select the features that are correlated with the label group according to the structure of label groups.(3)While measuring the correlation degree between the feature and the label group,we considered not only the correlation between feature and labels in the group but the importance of labels in the group.The experimental results show that our algorithm achieves better classification performance than other algorithms on many measures.
作者
孟威
周忠眉
MENG Wei;ZHOU Zhong-mei(School of Computer Science,Minnan Normal University,Zhangzhou 363000,China;Lab of Gramular Computing,Minnan Normal University,Zhangzhou 363000,China)
出处
《模糊系统与数学》
北大核心
2021年第1期144-154,共11页
Fuzzy Systems and Mathematics
基金
福建省自然科学基金资助项目(2018J01545)。
关键词
多标签分类
特征选择
标签组合
相关度
Multi-label Classification
Feature Selection
Label Group
Correlation