Border-Peeling Clustering.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Traditional clustering methods often struggle with adjacent clusters of varying densities.
• Existing non-parametric techniques like DBSCAN define cluster cores based on direct density estimations.

Purpose of the Study:

• To present a novel non-parametric clustering technique based on a layered cluster structure.
• To introduce a method that reveals latent cluster cores through progressive border point removal.

Main Methods:

• The proposed technique identifies border points by analyzing local neighborhood densities.
• A progressive peeling process removes border points, exposing the underlying cluster cores.
• This method adapts to local densities for effective cluster separation.

Main Results:

• The technique successfully separates adjacent clusters, even those with differing densities.
• Experiments on large labeled datasets, including high-dimensional deep features, demonstrate effectiveness.
• The method shows competitive performance against state-of-the-art non-parametric clustering algorithms.

Conclusions:

• The novel non-parametric clustering technique offers a robust alternative for complex datasets.
• Its layered approach and border point peeling mechanism provide adaptive and effective cluster separation.
• The method demonstrates strong performance and competitiveness with existing state-of-the-art techniques.