High-dimensional visual analytics: interactive exploration guided by pairwise views of point distributions.

Area of Science:

• Data Visualization
• High-Dimensional Data Analysis
• Statistical Computing

Background:

• Multivariate data exploration presents challenges in organizing and interpreting complex visualizations.
• Existing methods for analyzing high-dimensional data lack systematic approaches for display organization and interactive guidance.

Purpose of the Study:

• To introduce a method for organizing multivariate displays.
• To guide interactive exploration of high-dimensional data.
• To enhance anomaly detection and data sorting capabilities.

Main Methods:

• Characterizing 2D distributions of orthogonal pairwise projections using measures like density, skewness, shape, and outliers.
• Applying statistical analysis to these characterizations.
• Developing algorithms for organizing scatterplots, identifying outlying distributions, and sorting multivariate displays.

Main Results:

• A systematic approach to organizing 2D scatterplots for coherent viewing of high-dimensional data.
• Effective identification of unusual marginal 2D distributions for anomaly detection.
• A method for sorting diverse multivariate displays, including trees, parallel coordinates, and glyphs, based on data characteristics.

Conclusions:

• The proposed method offers a robust framework for interactive exploration and analysis of high-dimensional datasets.
• This approach facilitates coherent data viewing, anomaly detection, and efficient organization of complex visualizations.
• The findings contribute to improved understanding and manipulation of multivariate data in various scientific domains.