Unsupervised Feature Selection based on Constructing Virtual Cluster’s Representative

The data readability, complexity reduction of learning algorithms and increase predictability are the most important reasons for using feature selection methods, especially when there exist lots of features. In recent years, unsupervised feature selection techniques are well explored. In this paper, we proposed an unsupervised feature selection algorithm using multivariate-symmetrical-uncertainty based feature clustering, Feature Selection-based Virtual Feature Representative (FSVFR). The main idea of FSVFR is as follows: First, it selects the cluster centers based on the similarity density of the neighbors of each feature; after assigning the features to the clusters, the virtual representative is generated in such a way that contains maximum common information with cluster’s members and minimum similarity with other representatives. These steps continues until there is no more change in the representatives. Second, a feature that has the most common information in each cluster is selected as its representative. The experimental results on benchmark datasets demonstrate the effectiveness of our approaches as compared to the two common methods.