SOMFlow: Guided Exploratory Cluster Analysis with Self-Organizing Maps and Analytic Provenance.

Area of Science:

• Data Science
• Computer Science
• Human-Computer Interaction

Background:

• Clustering is crucial for uncovering hidden patterns in data.
• Current visual-interactive clustering methods often require extensive user tuning.
• Iterative refinement is essential for achieving meaningful cluster analysis.

Purpose of the Study:

• To present a multi-stage visual analytics (VA) approach for iterative cluster refinement.
• To introduce SOMFlow, an implementation utilizing Self-Organizing Maps (SOM) for time series data analysis.
• To enhance the understanding of clustering results and the interactive analysis process.

Main Methods:

• Developed a multi-stage Visual Analytics (VA) approach for iterative cluster refinement.
• Implemented SOMFlow using Self-Organizing Maps (SOM) for time series data analysis.
• Integrated a visual flow graph to track and compare analytical decisions and intermediate results.

Main Results:

• The SOMFlow system provides a visual platform for analyzing intermediate clustering results.
• Quality and interestingness measures are leveraged to guide analysts in discovering patterns.
• Pair analytics experiments demonstrated the approach's effectiveness in interactive data analysis.

Conclusions:

• The proposed VA approach facilitates iterative cluster refinement for complex datasets.
• SOMFlow enhances analyst understanding of clustering outcomes and the analytical workflow.
• The method supports effective exploration and interpretation of time series data structures.