Fast clustering algorithm based on MST of representative points.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Minimum Spanning Tree (MST)-based clustering is effective for diverse and irregular data shapes.
• Existing MST algorithms face computational challenges due to processing entire datasets.

Purpose of the Study:

• To develop a more computationally efficient and accurate MST-based clustering algorithm.
• To address the limitations of traditional MST clustering methods regarding dataset size and cluster number determination.

Main Methods:

• Proposed R-MST algorithm utilizes representative points for MST construction, reducing computational load.
• Introduced an improved representative point selection strategy based on density and nearest neighbor distance for better distribution in sparse regions.
• Developed an adaptive method using mutual neighbors to identify inconsistent edges and automatically determine the number of clusters.

Main Results:

• R-MST demonstrates improved efficiency compared to traditional MST clustering algorithms.
• The algorithm shows enhanced accuracy in clustering, particularly on datasets with varying densities.
• Automatic determination of the number of clusters eliminates the need for prior knowledge.

Conclusions:

• R-MST offers a significant advancement in MST-based clustering by optimizing computational efficiency and accuracy.
• The adaptive approach for cluster number determination makes R-MST more practical for real-world applications.
• The representative point strategy effectively handles datasets with heterogeneous densities.