Combining multiple clusterings using evidence accumulation.

Area of Science:

• Data Science
• Machine Learning
• Computational Statistics

Background:

• Combining multiple clusterings (clustering ensemble) is crucial for robust data analysis.
• Diverse clustering algorithms and parameter settings yield varied data partitions.

Purpose of the Study:

• To introduce and evaluate an evidence accumulation (EAC) framework for combining multiple clustering results.
• To develop a theoretical basis for analyzing and evaluating clustering ensemble strategies.

Main Methods:

• Generated a clustering ensemble from various algorithms, parameters, or data representations.
• Applied the EAC concept using a voting mechanism to create a similarity matrix from partitions.
• Utilized hierarchical agglomerative clustering on the similarity matrix for final data partitioning.
• Analyzed clustering stability using bootstrapping and mutual information.

Main Results:

• The EAC method produced a consistent clustering by aggregating evidence from individual partitions.
• Experimental results demonstrated competitive or superior performance compared to other combination strategies and individual algorithms.
• The theoretical framework provided a basis for evaluating the EAC strategy's effectiveness.

Conclusions:

• Evidence accumulation offers a robust approach to clustering ensemble generation.
• The proposed EAC framework enhances data organization and improves clustering accuracy.
• The method shows promise for analyzing synthetic and real-world datasets.