Efficient clustering aggregation based on data fragments.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Clustering aggregation, or clustering ensembles, combines multiple clustering results for improved accuracy.
• Existing point-based algorithms are computationally inefficient for large datasets.

Purpose of the Study:

• To develop an efficient fragment-based approach for clustering aggregation.
• To theoretically validate the fragment-based method using established goodness measures.
• To introduce and evaluate new clustering aggregation algorithms.

Main Methods:

• Developed a fragment-based approach where data fragments are subsets not split by clustering results.
• Proved theoretical underpinnings for fragment-based aggregation using two common goodness measures.
• Designed and implemented three new fragment-based clustering aggregation algorithms.

Main Results:

• The proposed fragment-based algorithms exhibit lower computational complexity compared to point-based methods.
• Experimental results on public datasets demonstrate comparable accuracy to existing algorithms.
• The new algorithms are efficient for large-scale clustering aggregation tasks.

Conclusions:

• The fragment-based approach provides a computationally efficient alternative for clustering aggregation.
• This method maintains accuracy while significantly reducing processing time for large datasets.
• The study offers practical advancements in scalable clustering ensemble techniques.