Fast and Effective Active Clustering Ensemble Based on Density Peak.

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Semisupervised clustering methods often use random pairwise constraints, leading to redundancy and instability.
• Existing active clustering methods lack comprehensive querying criteria and efficient label updating strategies.

Purpose of the Study:

• To propose novel active clustering algorithms that improve the efficacy of pairwise constraints.
• To enhance the representativeness and informativeness of selected instances for more robust clustering.
• To develop efficient label updating mechanisms and an ensemble framework for superior cluster separation.

Main Methods:

• Proposed an active density peak (ADP) clustering algorithm considering both representativeness and informativeness.
• Designed a fast-update-strategy for efficient label updates.
• Developed an active clustering ensemble framework incorporating local and global uncertainties with weighted voting consensus.

Main Results:

• Experimental comparisons on real-world datasets demonstrated the effectiveness of the proposed ADP clustering and ensemble framework.
• The methods showed improved performance over state-of-the-art techniques in active clustering.

Conclusions:

• The proposed active clustering methods effectively address limitations in existing approaches.
• The integration of representativeness, informativeness, and ensemble techniques leads to more stable and accurate clustering results.