DimLift: Interactive Hierarchical Data Exploration Through Dimensional Bundling.

Area of Science:

• Data Science
• Information Visualization
• High-Dimensional Data Analysis

Background:

• Exploratory data analysis is crucial for identifying patterns in datasets.
• High-dimensional datasets present challenges for pattern discovery.
• Existing dimensionality reduction techniques can obscure important dimensions.

Purpose of the Study:

• Introduce DimLift, a novel visual analysis method.
• Enable the creation and interaction with dimensional bundles.
• Facilitate the discovery of subtle relationships in complex datasets.

Main Methods:

• Dimensional bundles are generated via iterative dimensionality reduction or user-driven approaches.
• Dimensional bundles group dimensions that contribute similarly to dataset variance.
• Interactive exploration and reconstruction are performed using layered parallel coordinates plots.

Main Results:

• DimLift effectively lifts interesting and subtle relationships to the surface.
• The method handles complex scenarios with missing and mixed data types.
• A case study on clinical cohort data demonstrates the technique's power.

Conclusions:

• DimLift offers a powerful approach for visual analysis of high-dimensional data.
• The method enhances the identification of patterns in complex datasets.
• Applications span clinical, nutrition, and ecological domains.