A novel multiseed nonhierarchical data clustering technique.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Traditional clustering algorithms like K-means and Forgy struggle with elongated or nonconvex data shapes.
• These methods rely on single seed points, limiting their ability to capture complex cluster geometries.

Purpose of the Study:

• To propose a novel multiseed clustering algorithm for improved data grouping.
• To address the limitations of existing methods in handling non-spherical cluster shapes.

Main Methods:

• A density-based representative point selection algorithm identifies initial seed points.
• A minimal spanning tree guided technique assigns multiple seeds to a single cluster.
• A border point detection algorithm identifies cluster shapes to determine elongation.

Main Results:

• The proposed multiseed algorithm demonstrates superior performance in clustering complex shapes.
• Experimental results validate the effectiveness of the density-based seeding and MST-guided assignment.
• The border detection accurately identifies elongated cluster characteristics.

Conclusions:

• The novel multiseed clustering algorithm offers a significant improvement over single-seed methods.
• This technique provides a robust solution for clustering datasets with challenging, non-convex geometries.
• The algorithm's ability to detect cluster shape enhances its applicability in various data analysis scenarios.