Feature Selection for Clustering by Exploring Nearest and Farthest Neighbors

摘要

Feature selection has been explored extensively for use in several real-world applications. In this paper, we propose a new method to select a salient subset of features from unlabeled data, and the selected features are then adaptively used to identify natural clusters in the cluster analysis. Unlike previous methods that select salient features for clustering; our method does not require a predetermined clustering algorithm to identify salient features, and our method potentially ignores noisy features, allowing improved identification of salient features. Our feature selection method is motivated by a basic characteristic of clustering: a data instance usually belongs to the same cluster as its geometrically nearest neighbors and belongs to a cluster different than those of its geometrically farthest neighbors. In particular, our method uses instance-based learning to quantify features in the context of the nearest and the farthest neighbors of every instance so that clusters generated by the salient features maintain this characteristic.