A Novel Minimum Spanning Tree Clustering Algorithm Based on Density Core.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Clustering analysis is a key unsupervised learning technique with broad applications.
• Density-based clustering methods show promise but struggle with multidensity and complex shapes.
• Existing methods often exhibit parameter sensitivity and limitations in handling intricate data structures.

Purpose of the Study:

• To propose a novel clustering algorithm, MST-DC, designed to overcome the limitations of traditional density-based methods.
• To enhance clustering performance on datasets with varying densities and complex geometries.
• To offer a robust alternative to existing clustering algorithms for pattern recognition and data analysis.

Main Methods:

• The proposed MST-DC algorithm utilizes reverse nearest neighbors to identify core objects.
• A minimum spanning tree algorithm is employed to cluster these core objects.
• Non-core objects are subsequently assigned to the nearest identified cluster based on their closest core object.

Main Results:

• Experimental evaluations on synthetic and real-world datasets demonstrate the effectiveness of MST-DC.
• MST-DC shows superior performance compared to established algorithms including Kmeans, DBSCAN, DPC, DCore, SNNDPC, and LDP-MST.
• The algorithm successfully addresses challenges posed by multidensity and complex-shaped data.

Conclusions:

• The MST-DC algorithm presents a significant advancement in clustering analysis, particularly for complex datasets.
• Its density-core-based approach combined with minimum spanning tree clustering offers improved accuracy and robustness.
• MST-DC provides a superior alternative for various applications requiring effective unsupervised learning and pattern recognition.