Merging K-means with hierarchical clustering for identifying general-shaped groups

Area of Science:

• Data Science
• Machine Learning
• Computational Statistics

Background:

• Clustering algorithms partition data into similar groups.
• Hierarchical clustering offers tree-like structures but is computationally intensive for large datasets.
• K-means clustering is efficient but limited to identifying spherical clusters.

Purpose of the Study:

• To develop a hybrid non-parametric clustering approach.
• To overcome the limitations of existing hierarchical and K-means clustering methods.
• To identify general-shaped clusters in large datasets.

Main Methods:

• A hybrid approach combining K-means and hierarchical clustering.
• Initial data partitioning into spherical groups using K-means.
• Subsequent merging of groups using hierarchical methods with a data-driven stopping criterion.

Main Results:

• The hybrid method successfully identifies general-shaped clusters.
• The approach is applicable to large datasets, addressing computational complexity.
• Demonstrated good performance on simulated and real-world datasets.

Conclusions:

• The proposed hybrid clustering method effectively identifies complex data structures.
• This approach offers a scalable and versatile alternative for clustering.
• It has the potential to reveal hidden patterns in diverse datasets.