Self-Weighted Unsupervised LDA.

Area of Science:

• Machine Learning
• Unsupervised Learning
• Data Analysis

Background:

• Clustering methods are complex with many parameters, making tuning difficult.
• Existing methods require laborious parameter adjustments.
• A need exists for concise unsupervised learning models with adaptive parameter learning.

Purpose of the Study:

• To develop a novel self-weighted unsupervised linear discriminative analysis (SWULDA) method.
• To create a concise model that adaptively learns parameters, eliminating manual tuning.
• To elucidate the relationship between k-means and linear discriminant analysis (LDA) within an unsupervised framework.

Main Methods:

• Developed a novel self-weighted unsupervised linear discriminative analysis (SWULDA) method.
• Integrated the principles of minimizing within-class scatter and maximizing between-class scatter.
• Employed a quadratic weighted optimization framework for adaptive parameter learning.

Main Results:

• The proposed SWULDA method avoids manual parameter adjustment.
• SWULDA demonstrates a clear link between k-means and LDA.
• Experimental validation on multiple datasets confirms the method's effectiveness.

Conclusions:

• SWULDA offers a simplified and effective approach to unsupervised clustering.
• The adaptive parameter learning framework enhances model usability.
• The method provides a unified perspective on clustering and dimensionality reduction techniques.