Adopting Star Plot for Visualization of High Dimensional Multivariate Data

Abstract

The Star Plot is one of popular methods for visualization of multivariate data. This method displays each data record as a star-shaped icon by mapping all variables (dimensions) on radiating rays (axes) originated from a single point. The number of such icons is equal to the number of data items (records). As the number of dimensions and the size of data set increase, Star Plot visualization soon becomes too cluttered because many rays have to be accommodated within small circular area and individual icons also become too small. To overcome these problems associated with visualization of high-dimensional multivariate data, we propose different ways of effectively using the Star Plot method. First, instead of displaying multiple star-shaped icons, one for each data item, we plot all data items together. With this overlapping, the entire display space can be used for rendering. Second, we shift the origin outward to produce a circle in the center, the circumference of which provides the origin points for the respective axes and increase the spacing between rays toward low-value ends. Third, to accommodate a very large number of dimensions, we adopt a multi-level approach in which we partition the dimensions into groups and plot corresponding rays in multiple concentric circular rings. For instance, our two-level Star Plot draws a subset of rays (e.g., 1/4th of total dimensions) in inner small circle and the rest (3/4th dimensions) in outer larger ring. The GUI support allows the user choose desired Star Plot option and also dynamically adjust the dimension partitioning and the ring boundary. We have demonstrated the effectiveness and usefulness of the proposed Star Plot extension by visualizing three multivariate data sets of varying number of dimensions.