Extending Dimensions in Radviz based on mean shift

Abstract

Radviz is a radial visualization technique which maps data from multiple dimensional space onto a planar picture. The dimensions placed on the circumference of a circle, called Dimension Anchors (DAs), can be reordered to reveal different patterns in the dataset. Extending the number of dimensions can enhance the flexibility in the placement of the DAs to explore more meaningful visualizations. In this paper, we describe a method which rationally extends a dimension to multiple new dimensions in Radviz. This method first calculates the probability distribution histogram of a dimension. The mean shift algorithm is applied to get centers of probability density to segment the histogram, and then the dimension can be extended according to the number of segments of the histogram. We also suggest using the Dunn's index to find the optimal placement of DAs, so the better effect of visual clustering could be achieved after the dimension expansion in Radviz. Finally, we demonstrate the usability of our approach on visually analysing the iris data and two other datasets.