A Visual Representation of Clusters Characteristics using Edge Bundling for Parallel Coordinates

Abstract

This paper proposes an edge bundling approach applied on parallel coordinates to improve the visualization of cluster information directly from the overview. Lines belonging to a cluster are bundled into a single curve between axes, where the horizontal and vertical positioning of the bundling intersection (known as bundling control points) to encode pertinent information about the cluster in a given dimension, such as variance, standard deviation, mean, median, and so on. The hypothesis is that adding this information to the overview improves the visualization overview at the same that it does not prejudice the understanding in other aspects. We have performed tests with participants to compare our approach with classic parallel coordinates and other consolidated bundling technique. The results showed most of the initially proposed hypotheses to be confirmed at the end of the study, as the tasks were performed successfully in the majority of tasks maintaining a low response time in average, as well as having more aesthetic pleasing according to participants’ opinion. KeywordsEdge Bundling, Parallel Coordinates, Data Clustering.