Does 3D really make sense for visual cluster analysis? yes!

Our paper takes the stance that a 3D shaded display can add a significant amount of information to the visualization of high-dimensional data. We believe that it makes better use of the innate cognitive capabilities of the human visual system which is highly optimized to recognize and reason with 3D shape information. As a first step we studied a variety of real-world datasets and confirmed that the extension from the traditional 2D space to 3D space is indeed justified - most datasets we studied had clusters residing in subspaces with more than two significant principal components. We then describe an interactive interface that allows users to navigate these 3D subspaces, expand the exploration to higher dimensionalities, and also transition among the distinct subspaces inhabited by different clusters in the data.