‘Multi View Graphing’: Linked Multi Visualization utilising Brushing, Binning and Clustering

Visualization can provide a distinctly advantageous overview of data, enabling the rapid identification of anomalies, patterns or correlations that would not otherwise be obvious. Different visualization techniques each offer their own unique insight into the same data; however the similarities that exist between them are not always clear. High data density can also be a very evident issue when exploring data using visualization. The densest datasets can ensure that even well suited visualization methodologies succumb to usability issues. The most powerful data analysis environments are arguably those that provide interactive exploration, however visual feedback in such environments is sometimes undesirably limited. The concept of linking different visualization styles using interactive techniques, such as brushing, is currently evident in multiple publically available software environments. To explore the concept of linked visualization a prototype application was produced, allowing up to four unique visual styles to be generated using the same data, at the same time. Current brushing methodologies were extended and included, in order to provide the ability to affect each visualization from within every other. The issue of data density was tackled through the use of a novel approach to binning based around a uniform grid. Visual cues were used extensively throughout the prototype, ranging from representing a brushing area through to defining the basic starting parameters of a clustering algorithm. Three distinctly different test cases are presented to demonstrate the techniques showcased within the prototype, each in conjunction with external collaborators. Results suggest that using linked multi visualization is a more effective method of data analysis, offering greater insight than using a lone visualization technique. In tackling data density, the grid based binning has the ability to offer an easily disseminated overview of even extremely cluttered visualizations. The extensive use of visual cues in the prototype vindicated the theory that offering clear feedback within interactive environments is of the utmost importance. Also the interactive definition of clustering parameters via visual cues shows promise as a concept but one requiring further research. This study highlights that the current trend towards linking multiple visualization techniques within advanced data analysis environments is correct; it also introduces novel brushing, binning and clustering concepts worthy of further investigation.