Iterative Discovery of Multiple AlternativeClustering Views.

Area of Science:

• Data Science
• Machine Learning
• Computational Statistics

Background:

• Existing clustering algorithms typically yield a single solution, limiting exploratory data analysis.
• Data analysts often require multiple, diverse clustering outcomes for comprehensive interpretation.
• Identifying alternative clusterings within different data subspaces (views) is crucial for deeper insights.

Purpose of the Study:

• To develop a novel approach for generating multiple, diverse clustering solutions.
• To enable the discovery of alternative clusterings within different data subspaces.
• To provide data analysts with enhanced tools for exploratory data analysis.

Main Methods:

• An optimization procedure is employed to simultaneously identify subspaces and their corresponding clusterings.
• The algorithm incorporates objective functions for both cluster quality and novelty.
• Simultaneous and iterative discovery modes are explored and compared.

Main Results:

• The proposed approach successfully generates multiple, distinct clustering solutions.
• The method effectively identifies relevant data subspaces associated with alternative clusterings.
• Experimental comparisons demonstrate the efficacy of the novel approach against existing methods.

Conclusions:

• The novel approach enhances exploratory data analysis by providing multiple clustering solutions and identifying subspace structures.
• This method offers a significant advancement over single-solution clustering algorithms.
• The ability to discover multiple clusterings and their associated subspaces facilitates a more thorough understanding of complex data.