Feature Selection for Visual Clustering

VISTA, a dynamic information visualization system, which allows the user to interactively observe potential clusters in a series of continuously changing visualizations, incorporates the algorithmic clustering results, and serves as an effective validation and refinement tool for irregularly shaped clusters. The validation and visual clustering is performed by tuning the parameter alpha for dominating attributes. Identification of the dominating attributes for visual tuning and visual distance computation process becomes tedious, when the dimensionality of the dataset increases. One common approach to solve this problem is dimensionality reduction. Most of the existing dimensionality reduction techniques are ldquowrapper methodsrdquo, that requires the use of class labels. This paper proposes a novel filter feature selection method for interactive visual clustering based on the outlier score of the features. The experimental results show that the proposed method works well in finding the dominating attributes and eases the visual distance computation process.