Feature Selection For Fuzzy Multi-label Data

In the domain of machine learning,multi-label learning is an significant section.How to choose features to structure a classifier is the crux to processing multi-label data.According to the fuzzy rough set theory and the maximum-dependence and minimum-redundancy algorithm of information entropy,this paper proposes two feature selection methods for fuzzy multi-label data sets,and designs the algorithms of these two methods.1.Multi-label learning demand to take into account the internal relations among multi-label.For the sake of settle this matter better,in this dissertation,a fuzzy multi-label fuzzy rough set model is used for settling the feature selection of fuzzy multi-label data.Above all,the labels of multi-label data are fuzzified,and samples are sorted based on the distance between fuzzy labels to determine the probability that samples belong to different classes.Then,the k-nearest neighbor criterion is used to ascertain the similarity and heterogeneous samples of a sample,and the feature selection model of fuzzy multi-label data is established.Finally,the fundamental qualities of the model are analyzed,the importance of the features is defined,and experimental results bear out the validity of this way.2.In the feature selection of fuzzy multi-label data,the features selected according to the degree of dependence are likely to come into being redundancy.For the sake of settle this matter,in the dissertation,a new assessment method is proposed.In the first place,two factors affecting fuzzy multi-label features are introduced: feature dependence and feature redundancy.Dependence means the contribution of candidate features to each tag,while redundancy means the information overlap between candidate features and selected features under all tags.Then a new evaluation method is proposed,which links fuzzy information entropy with maximum-dependence and minimum-redundancy algorithm,and chose the feature subset with the optimal appraisal standard from the feature space.Finally,the experiment shows that the method is superior.