Semisupervised Feature Selection via Generalized Uncorrelated Constraint and Manifold Embedding.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• Ridge regression is common in supervised and semisupervised learning.
• Direct application of ridge regression to semisupervised learning lacks closed-form solutions and manifold structure.
• Existing methods struggle to efficiently leverage data topology.

Purpose of the Study:

• To propose a novel semisupervised feature selection method (SFS).
• To address limitations of standard ridge regression in semisupervised learning.
• To enhance feature selection by incorporating manifold structure and ensuring a closed-form solution.

Main Methods:

• Developed a semisupervised feature selection (SFS) method.
• Introduced a generalized uncorrelated constraint to ridge regression.
• Embedded manifold structure and a full rank constraint on the projection matrix.
• Utilized an adaptively obtained scale factor for flexibility.

Main Results:

• The proposed SFS method achieves an elegant closed-form solution.
• The method effectively preserves data topology through manifold structure.
• Full rank constraint prevents excessive row sparsity.
• Experimental results demonstrate superiority over state-of-the-art semisupervised feature selection techniques.

Conclusions:

• The novel SFS method offers significant advantages for semisupervised feature selection.
• The generalized uncorrelated constraint is key to achieving closed-form solutions and preserving manifold structure.
• This approach provides a more robust and efficient alternative to existing methods.