TreeKDE: clustering multivariate data based on decision tree and using one-dimensional kernel density estimation

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Clustering multidimensional data is a fundamental task in data analysis.
• Existing algorithms may struggle with automatic cluster number determination and defining cluster boundaries.
• There is a need for efficient and robust clustering methods.

Purpose of the Study:

• To present a new algorithm, TreeKDE, for clustering multidimensional data.
• To demonstrate the algorithm's capability for automatic cluster number determination.
• To showcase its efficiency and competitiveness against existing methods.

Main Methods:

• The TreeKDE algorithm utilizes a decision tree structure.
• It optimizes a one-dimensional kernel density estimator (KDE).
• Orthogonal projections of data onto coordinate axes are employed.

Main Results:

• TreeKDE automatically determines the number of clusters.
• It effectively defines cluster boundaries within rectangular regions.
• Comparative experiments show TreeKDE is efficient and competitive.

Conclusions:

• TreeKDE offers a simple and efficient approach to data clustering.
• The algorithm shows promise for further research and development.
• It provides a foundation for new clustering algorithms combining decision trees and KDE.