Ensemble Clustering via Co-Association Matrix Self-Enhancement.

Area of Science:

• Data Science
• Machine Learning
• Computer Science

Background:

• Ensemble clustering combines multiple base clustering results for superior performance.
• Current methods often depend on co-association (CA) matrices, which can be low-quality.
• Degraded CA matrix quality leads to diminished ensemble clustering performance.

Purpose of the Study:

• To propose a self-enhancement framework for CA matrices to improve ensemble clustering.
• To enhance the quality of CA matrices by integrating high-confidence information.
• To achieve more robust and accurate clustering results through improved CA matrix construction.

Main Methods:

• Extracting high-confidence (HC) information from base clusterings to create a sparse HC matrix.
• Propagating reliable information from the HC matrix to the CA matrix.
• Simultaneously complementing the HC matrix using CA matrix information to generate an enhanced CA matrix.
• Formulating the model as a symmetric constrained convex optimization problem solved via an alternating iterative algorithm.

Main Results:

• The proposed framework effectively enhances the CA matrix for better ensemble clustering.
• Experimental comparisons on ten benchmark datasets demonstrate superior performance over 12 state-of-the-art methods.
• The method shows effectiveness, flexibility, and efficiency in ensemble clustering tasks.

Conclusions:

• The developed CA matrix self-enhancement framework significantly improves ensemble clustering performance.
• The proposed optimization approach guarantees convergence and global optimum.
• The study provides a flexible and efficient solution for constructing high-quality CA matrices in ensemble clustering.