Towards a Systematic Combination of Dimension Reduction and Clustering in Visual Analytics

Area of Science:

• Data Science
• Computer Science
• Information Visualization

Background:

• Dimension reduction and clustering algorithms are key visual analytics tools for analyzing data similarity and group structures.
• Historically, these algorithms were used independently, but recent trends integrate them into single visualization systems.
• Current integrated approaches are often disconnected, lacking interdependence between dimension reduction and clustering.

Purpose of the Study:

• To provide an overview of combining dimension reduction and clustering algorithms within visualization systems.
• To identify and discuss the inherent challenges in developing integrated visualization systems that utilize both algorithm families.
• To inform the design of more cohesive and interdependent algorithmic combinations.

Main Methods:

• Literature review of existing visual analytics systems incorporating dimension reduction and clustering.
• Analysis of design decisions in concurrent algorithm application: algorithm selection, processing order, and interaction design.
• Identification of challenges in creating interdependent algorithmic frameworks.

Main Results:

• Existing systems often combine dimension reduction and clustering in an ad hoc or parallel manner.
• Significant design decisions are required for effective integration, including algorithm choice and processing sequence.
• A lack of interdependence between the two algorithm families is a common issue.

Conclusions:

• Combining dimension reduction and clustering offers potential for enhanced data analysis but presents significant design challenges.
• Developing integrated systems requires careful consideration of algorithmic interdependence and user interaction.
• Further research is needed to create more cohesive and effective frameworks for combining these techniques.