Transfer Clustering Ensemble Selection.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Clustering ensemble (CE) methods enhance accuracy and robustness by integrating multiple clustering solutions.
• Clustering ensemble selection (CES) refines CE by reducing redundancy and noise, focusing on quality and diversity metrics.
• Existing CES strategies often rely on heuristic methods or fixed thresholds, limiting the optimal trade-off between quality and diversity.

Purpose of the Study:

• To propose a novel transfer clustering ensemble selection (TCES) algorithm.
• To leverage the relationship between quality and diversity from a source dataset and transfer it to a target dataset.
• To optimize the trade-off between quality and diversity for improved clustering performance.

Main Methods:

• Developed a transfer CES (TCES) algorithm utilizing three objective functions to transfer quality-diversity relationships.
• Implemented a multiobjective self-evolutionary process to optimize the defined objective functions.
• Constructed a transfer CE framework (TCE-TCES) integrating the TCES algorithm.

Main Results:

• The proposed TCE-TCES framework demonstrated the ability to find a superior trade-off between clustering quality and diversity.
• Experimental results on 12 transfer clustering tasks from the 20newsgroups dataset validated the effectiveness of TCE-TCES.
• The method achieved more desirable clustering outcomes compared to existing approaches.

Conclusions:

• The proposed TCES algorithm effectively transfers quality-diversity knowledge across datasets.
• The TCE-TCES framework offers a robust approach for optimizing clustering ensemble selection.
• This research contributes to advancing the field of ensemble learning in clustering.