Brushing Moments in Interactive Visual Analysis

Abstract

We present a systematic study of opportunities for the interactive visual analysis of multi-dimensional scientific data that is based on the integration of statistical aggregations along selected independent data dimensions in a framework of coordinated multiple views (with linking and brushing). Traditional and robust estimates of the four statistical moments (mean, variance, skewness, and kurtosis) as well as measures of outlyingness are integrated in an iterative visual analysis process. Brushing particular statistics, the analyst can investigate data characteristics such as trends and outliers. We present a categorization of beneficial combinations of attributes in 2D scatterplots: (a) k th vs. (k + 1) th statistical moment of a traditional or robust estimate, (b) traditional vs. robust version of the same moment, (c) two different robust estimates of the same moment. We propose selected view transformations to iteratively construct this multitude of informative views as well as to enhance the depiction of the statistical properties in scatterplots and quantile plots. In the framework, we interrelate the original distributional data and the aggregated statistics, which allows the analyst to work with both data representations simultaneously. We demonstrate our approach in the context of two visual analysis scenarios of multi-run climate simulations.

Representative Factor Generation for the Interactive Visual Analysis of High-Dimensional Data

“…There are few other works where similar dual analysis methods already proved to be useful, such as in parameter space exploration [4], temporal data analysis [3], and multi-run simulation data analysis [24]. Kehrer et al [23] The structure of high-dimensional datasets and the relations between the dimensions have been investigated in a few studies, also. Seo and Shneiderman devise a selection of statistics to explore the relations between the dimensions in their Rank-by-Feature framework [33].…”

“…uniq values are usually higher for continuous dimensions and lower for categorical dimensions. We use a method based on robust statistics [23] to determine %out values. In order to investigate if the dimensions follow a normal distribution, we also apply the Shapiro-Wilk normality test [31] to the dimensions and store the resulting p-values (pVal shp ) in S. Higher pVal shp indicate a better fit to a normal distribution.…”

On Computationally-Enhanced Visual Analysis of Heterogeneous Data and Its Application in Biomedical Informatics

“…Performing the high-dimensional data analysis on derived attributes is a strategy utilized in a number of studies. Kehrer et al [49] integrate statistical moments and aggregates to interactively analyze collections of multivariate data sets. In the VAR display by Yang et al [48], the authors represent the dimensions as glyphs on a 2D projection of the dimensions.…”

“…In this setting, the user interacts with the computational mechanism either through modifying parameters or altering the data domain [46], [47], [48], [49] [50]…”

Hierarchical visual filtering, pragmatic and epistemic actions for database visualization

“…Instead of creating multiple views by means of visual filtering, they define a collective filtering based on selections originating from multiple views. Kehrer et al [10] argue in favor of statistical summarizations (mean, variance, skewness, and kurtosis), each in a dedicated view, as a means for interpreting visually filtered data. Also recent, Turkay et al [19] create multiple views by simultaneously filtering data and data dimensions assisted by statistical summaries.…”