Unsupervised feature selection based on feature relevance

Feature selection is an essential technique used in data mining and machine learning. Many feature selection methods have been studied for supervised problems. However feature selection for unsupervised learning is rarely studied. In this paper, we proposed an approach to select features for unsupervised problems. Firstly, the original features are clustered according to their relevance degree defined by mutual information. And then the most informative feature is selected from each cluster based on the contribution-information of each feature. The experimental results show that the proposed method can match some popular supervised feature selection methods. And the features selected by our method do include most of the information hidden in the overall original features.