RadViz and Identification of Clusters in Multidimensional Data

RadViz visualization makes it possible to map data from n-dimensional space into a plane. The paper reviews those specific properties of this method that are important for identification of clusters in the original multidimensional data. First, there is described an artificial data set which clearly points to a certain drawback of the original RadViz mapping. To resolve the identified problem there are suggested 2 minor modifications of the RadViz algorithm. Finally, it is proved that application of both suggested modifications guarantees that the upper mentioned problem does not re-appear. This claim is experimentally confirmed by a new visualization of the two original data sets using the modified mapping algorithm.