General Plane-Based Clustering With Distribution Loss.

Area of Science:

• Machine Learning
• Data Mining
• Computational Statistics

Background:

• Existing plane-based clustering methods lack a unified framework.
• Current methods may not optimally capture complex data distributions.

Purpose of the Study:

• To propose a general model for plane-based clustering that encompasses existing methods.
• To develop a novel distribution-based clustering (DPC) method with improved data distribution capture.
• To analyze the theoretical properties and performance of the proposed general model and DPC.

Main Methods:

• Developed a general plane-based clustering model based on minimizing sample loss from within- and between-cluster information.
• Introduced a novel distribution loss function into the general model to create the DPC method.
• Analyzed theoretical termination conditions, proving finite convergence to a local or weak local solution.

Main Results:

• The general model integrates various existing plane-based clustering techniques.
• The proposed DPC method effectively captures data distribution due to its statistical properties.
• DPC demonstrates superior performance compared to state-of-the-art methods on diverse datasets.

Conclusions:

• The general model provides a unified framework for plane-based clustering.
• DPC offers a statistically robust and high-performing clustering solution.
• The proposed framework advances the field of plane-based clustering with practical implications.