A physical model inspired density peak clustering.

Area of Science:

• Data Mining
• Machine Learning
• Computational Biology

Background:

• Clustering is crucial in data mining for bioscience and network analysis.
• Density peak clustering (DPC) is widely used but sensitive to data structure and prone to allocation errors.
• Existing DPC algorithms struggle with datasets of varying cluster densities and shapes.

Purpose of the Study:

• To address limitations of traditional density peak clustering algorithms.
• To propose a new clustering algorithm, potential-field-diffusion-based density peak clustering (PFD-DPC).
• To improve accuracy and robustness in identifying clusters within complex datasets.

Main Methods:

• Introduced a potential field concept and a novel density measure based on potential field diffusion.
• Defined new criteria for identifying similar points and employed distinct allocation strategies for dissimilar points.
• Conducted extensive experiments on synthetic and real-world datasets.

Main Results:

• The proposed PFD-DPC algorithm accurately selects cluster centers using the novel density measure.
• PFD-DPC effectively avoids data point allocation errors by differentiating similar and dissimilar points.
• Demonstrated excellent clustering performance across diverse datasets (varying size, dimension, shape, density, and nonconvexity).

Conclusions:

• PFD-DPC offers a robust and accurate clustering solution, outperforming existing methods.
• The algorithm is highly suitable for complex and challenging datasets, including those with variable densities and nonconvex structures.
• This advancement in clustering technology has significant implications for data mining applications in various scientific fields.