Voronoi Diagram Based Dimensional Anchor Assessment for Radial Visualizations

Selecting the most expressed dimensions from high dimensional data sets has motivated the design and application of a variety of statistical and machine learning techniques. Here, in our current work, we introduce a metric for assessing the effectiveness of these methods. Our formulation is based on the broad concepts of: (a) devising a formal method of partitioning a visualization's image space; (b) identifying regions that indicate the relative strength of the dimension selection based on how well they are populated by data images; and (c) similarily identifying those regions indicating a poor selection of dimensions. In particular, we explore assessing the quality of radial visualizations. Dimension selection in this class of visualizations strongly effects visualization quality and the sensitivity of cluster formation. We demonstrate the usefulness of Voronoi partitioning the RadViz image space; quantifying radial visualization quality is a direct measure of dimension selection. This work continues to develop and refine the formal theory behind the general class of Normalized Radial Visualizations, including RadViz.