SMMP: A Stable-Membership-Based Auto-Tuning Multi-Peak Clustering Algorithm.

Area of Science:

• Data Mining
• Machine Learning
• Computational Statistics

Background:

• Existing single-prototype clustering algorithms struggle with complex cluster shapes.
• Multi-prototype algorithms often require manual parameter tuning for cluster number estimation and shape detection.
• These limitations lead to time-consuming and potentially inaccurate clustering results.

Purpose of the Study:

• To propose a novel stable-membership-based auto-tuning multi-peak clustering (SMMP) algorithm.
• To enable fast, automatic, and effective multi-prototype clustering without iterative processes.
• To address the challenges of automatic cluster number estimation and complex shape detection.

Main Methods:

• Employs a dynamic association-transfer method to learn point representativeness to sub-cluster centers, utilizing density peak clustering.
• Uses a border-link-based connectivity measure for high-fidelity sub-cluster similarity evaluation.
• Leverages stable membership states under varying thresholds to automatically determine the number of sub-clusters and clusters.

Main Results:

• SMMP achieves fast, automatic, and effective multi-prototype clustering without iteration.
• The algorithm successfully handles complex-shaped clusters.
• Demonstrates effectiveness on both synthetic and real-world large datasets.

Conclusions:

• SMMP offers a significant advancement in multi-prototype clustering, particularly for complex data structures.
• The iterative-free and auto-tuning nature of SMMP makes it a valuable tool for large-scale data analysis.
• This method overcomes key limitations of existing clustering algorithms, enhancing efficiency and accuracy.