无监督环境下基于聚类集成的特征选择

Unsupervised Feature Selection Using Clustering Ensemble Method

无监督学习环境下的特征选择往往无法取得像有监督学习环境下那样令人满意的效果。文章提出了一种无监督环境下特征选择的方法,能够根据特征之间的相关性对特征进行聚类,通过选择聚类中那些与该类内其他特征相关性最大的特征作为精简特征集,因此不需要进行特征空间的搜索,同时可以获得更有意义的分类信息。

The result of the feature selection in unsupervised learning is not as satisfactory as that in supervised learning.This paper presents an feature selection method in unsupervised learning,which is able to group features based on their interdependence. By selecting a subset of feature which have high multiple-interdependance with others within cluster,no search in original features is needed and significant classification information can be obtained.