Representative Point-Based Clustering With Neighborhood Information for Complex Data Structures.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Clustering complex data structures with varying densities, shapes, and noise presents significant challenges.
• Existing clustering methods often overlook the crucial role of neighborhood information.

Purpose of the Study:

• To introduce a novel clustering algorithm, representative point-based clustering with neighborhood information (RPC-NI).
• To address limitations in discovering clusters within complex data structures by incorporating neighborhood information.

Main Methods:

• Developed a new local centrality metric integrating neighborhood density and topological convergence to identify core representative points.
• Defined a density-adaptive distance for evaluating dissimilarities between representative points.
• Constructed a minimum spanning tree (MST) using these distances and applied an MST-based clustering algorithm.

Main Results:

• RPC-NI successfully identifies representative points by comprehensively utilizing neighborhood information.
• The algorithm adapts to arbitrary cluster shapes, varying densities, and different sizes due to multiple representative points per cluster.
• Extensive experiments show RPC-NI outperforms baseline algorithms in clustering accuracy and robustness.

Conclusions:

• Neighborhood information is critical for discovering clusters in complex data structures.
• RPC-NI offers a robust and accurate approach for challenging clustering tasks.
• The method's adaptability makes it suitable for diverse and complex datasets.