Brush-Based Ranking For Navigating Within High-Dimensional Datasets

The analysis of high-dimensional data means a big challenge, as most common visualization techniques do not scale well for displaying a large number of attributes at one time. Therefore, the initial questions arising when analyzing a new dataset typically concern the dimensions themselves in order to assess the relevance of various attributes and to identify clusters of similar (i.e., highly correlated) attributes. After considering this first step, entry-related tasks like detecting outliers or clusters of similar entries can be dealt with more efficiently in a second step. In this paper, we describe an approach which guides the user through a high-dimensional dataset by ranking dimensions and pairs of dimensions according to a large number of statistical summaries. The option to restrict the computations to subsets of the data (e.g., interactively defined by brushing a linked view) and to statistically compare various subsets makes this approach even more powerful and widely applicable, as illustrated by means of a biological dataset.