Multidimensional Visualization Based On Concentric Coordinates

Information visualization has changed the way in which people represent, analyze and comprehend data since it appeared in middle 1990s. And as one of the most classical research fields of information visualization, multidimensional visualization receives much concern all the time. A number of methods have been developed practically; however, there are still lots of issues to be solved urgently.In this paper, a new method named Concentric Coordinates is proposed to overcome the massive lines crossings and overlaps of the parallel coordinates when visualizing multidimensional dataset with large scale. In our approaches, axes are arranged as concentric circles rather than parallel lines and edges which represent each data item are drawn as segments of curves rather than poly-lines as it is done in the traditional parallel coordinates. After theoretical analysis, it has been proved that comparing to parallel layout concentric fashion can reduce crossings number 33% if the raw data could be transferred into complete bipartite graph. And for common data, some heuristics in horizontal crossing reduction are applied to solve the same problem in our concentric fashion. Experiments with random discrete dataset show that comparing to parallel coordinates, concentric coordinates could reduce the number of crossings by more than 15 percent overall with other similar situations, e.g. dealing with dataset in same scale and using the same heuristics. It is showed by empirical evidence as well that our approach could increase crossing angles partially. Fewer crossings and larger crossing angle are both the major criteria for human readability. That means our approach can enhance the perception and cognition of raw data.Besides the analysis in mathematics and geometry, the implement of concentric coordinates is introduced in detail, including the dimension representation and two ways to display data item based on Bezier curves and B-Spline curves respectively. And the usability is evaluated with real network data in port scan detection and protocol analysis.