ABSTRACT

Dimensional Stacking is a technique for the display of high-dimensional data involving the recursive embedding of pairs of dimensions within one another. This paper describes strategies and techniques for searching a data set displayed in this manner to uncover structure. The process, analogous to television channel selection and fine tuning, creates a simple yet powerful environment for examining data. A semi-automated technique, which we call data-driven view selection, is also presented. Examples of the use of the system are shown and a comparison to projection pursuit techniques is presented. Several methods of view enhancement are also described, along with a study of the effects of adapting traditional image processing algorithms to operate on the N-dimensional data.

Keywords: high-dimensional graphics, plotting multivariate data, visualization, dimensional stacking