Automated Cluster Elimination Guided by High-Density Points.

Area of Science:

• Data Science
• Machine Learning
• Computational Statistics

Background:

• Determining the optimal number of clusters in cluster analysis is a significant challenge.
• Existing methods using clustering validity indices (CVIs) are complex and often yield conflicting results.
• This complexity hinders efficient and accurate cluster analysis.

Purpose of the Study:

• To introduce a novel clustering algorithm, self-regulating possibilistic C-means with high-density points (SR-PCM-HDP), for simplified and efficient cluster number determination.
• To enhance clustering efficiency and validity, particularly for datasets with challenging distributions.
• To overcome the limitations of existing CVI-dependent methods.

Main Methods:

• Developed density-based knowledge extraction (DBKE) to estimate initial cluster numbers and identify high-density points, enhancing Density Peak Clustering (DPC).
• Introduced SR-PCM-HDP with a parameter balancing high-density points and cluster centers for improved convergence and reduced sensitivity.
• Redefined possibilistic C-means (PCM) parameter adjustment for adaptive cluster elimination and formation.

Main Results:

• SR-PCM-HDP accurately determines cluster numbers and ensures clustering validity.
• The algorithm demonstrates superior performance on datasets with overlapping or imbalanced distributions.
• Experimental results show effectiveness compared to 13 state-of-the-art algorithms.

Conclusions:

• SR-PCM-HDP offers a more efficient and accurate approach to cluster analysis by simplifying cluster number determination.
• The algorithm effectively handles complex datasets, outperforming existing methods.
• This novel approach provides a robust solution for unsupervised learning challenges.