Using multiple attribute-based explanations of multidimensional projections to explore high-dimensional data. (August 2021)

• Record Type: Journal Article
• Authors: Tian, Zonglin; Zhai, Xiaorui; van Driel, Daan; van Steenpaal, Gijs; Espadoto, Mateus; Telea, Alexandru
• Is Part Of: Computers & graphics. Volume 98(2021)
• Page Start: 93
• Page End: 104
• Publication Date: 2021-08
• Subjects: Dimensionality reduction -- Explanatory techniques -- High-dimensional data analysis; Computer graphics -- Periodicals; 006.6
• DOI: 10.1016/j.cag.2021.04.034
• Languages: English
• ISSNs: 0097-8493

Abstract

Highlights: We provide additional examples of how the five explanatory views in [Van Driel et al] and [Da Silva et al] can be combined in a visual analytics fashion to find relevant insights in high-dimensional datasets that cannot be found using a single view. We illustrate the above process on five non-synthetic datasets, and correlate the obtained insights with ground-truth information independently extracted by other researchers from three of these datasets. Of these datsets, a single one has been used in the earlier work, the other being new. Also, the correlation of obtained insights with the ground-truth information is new. We present a new explanatory method, variance ratio, for computing local dimensionality. We discuss in detail the parameter settings and dependency on the used projection techniques of our proposed explanatory visualization. Graphical abstract: The Graphical Abstract includes the first row of Figure 3 in the paper. From left to right, these images are the explanation of the wine quality dataset using the dimension contribution, variance, and dimension correlation. Abstract: Multidimensional projections (MPs) are effective methods for visualizing high-dimensional datasets to find structures in the data like groups of similar points and outliers. The insights obtained from MPs can be amplified by complementing these techniques by several so-called explanatory mechanisms. We present and discuss a set of six such mechanisms that explain MPs in terms of similar dimensions, local dimensionality, and dimension correlations. We implement our explanatory tools using an image-based approach, which is efficient to compute, scales well visually for large and dense MP scatterplots, and can handle any projection technique. We demonstrate how the provided explanatory views can be combined to augment each other's value and thereby lead to refined insights in the data for several high-dimensional datasets, and how these insights correlate with known facts about the data under study.