Reordering Sets of Parallel Coordinates Plots to Highlight Differences in Clusters
2022, Koh, Elliot, Blumenschein, Michael, Shao, Lin, Schreck, Tobias
Visualizing high-dimensional (HD) data is a key challenge for data scientists. The importance of this challenge is to properly map data properties, e.g., patterns, outliers, and correlations, from a HD data space onto a visualization. Parallel coordinate plots (PCPs) are a common way to do this. However, a PCP visualization can be arranged in several ways by reordering its axes, which may lead to different visual representations. Many methods have been developed with the aim of evaluating the quality of reorderings of given PCP view. A high-dimensional data set can be divided into multiple classes, and being able to identify differences between the classes is important. Then, besides overlaying the groups in a single PCP, we can show the different groups in individual PCPs in a small multiple fashion. This raises the problem of jointly reordering sets of PCPs to create meaningful reorderings of the set of plots. We propose a joint reordering strategy, based on maximizing the pairwise visual difference in PCPs, such as to support their contrastive comparison. We present an implementation and an evaluation of the reordering strategy to assess the effectiveness of the method. The approach shows feasible in bringing out pairwise difference in PCP plots and hence support comparison of grouped data.