Exploring High-D Spaces with Multiform Matrices and Small Multiples.

Area of Science:

• Information Visualization
• Exploratory Data Analysis (EDA)
• Geovisualization

Background:

• Multivariate data analysis requires effective visualization tools.
• Existing methods like scatterplot matrices and small multiples have limitations.
• Integrating diverse visualization techniques can enhance exploratory data analysis.

Purpose of the Study:

• To introduce a novel approach for visual analysis of multivariate data.
• To develop a flexible, coordinated, multiview exploratory data analysis (EDA) toolkit.
• To enhance the exploration of complex relationships within high-dimensional datasets.

Main Methods:

• Leveraging a component-based architecture (GeoVISTA Studio) for a flexible EDA toolkit.
• Developing MultiForm Bivariate Matrix and Small Multiple plots for combined bivariate representations (e.g., scatterplots, bivariate maps, space-filling displays).
• Applying conditional entropy for variable selection and ordering, and implementing conditioning (dynamic query/filtering) for focused analysis.

Main Results:

• Demonstrated flexibility in depicting multivariate data using combined visualization forms.
• Successfully identified potentially interesting variable relationships using conditional entropy.
• Enabled focused exploration by removing the influence of known variables through conditioning.

Conclusions:

• The integrated approach offers a powerful and flexible toolkit for multivariate data visualization and EDA.
• The developed methods enhance the ability to discover relationships in high-dimensional data.
• The toolkit facilitates a more intuitive and efficient exploration of complex datasets, as shown in cancer data analysis.