Self-Weighted Supervised Discriminative Feature Selection.

Area of Science:

• Machine Learning
• Data Science
• Pattern Recognition

Background:

• Feature selection is crucial for dimensionality reduction and improving model performance.
• Existing sparse feature selection methods may overly suppress relevant features.
• There is a need for methods that select discriminative features effectively while ensuring minimal feature subsets.

Purpose of the Study:

• To propose a novel self-weighted orthogonal linear discriminant analysis (SOLDA) problem.
• To derive a self-weighted supervised discriminative feature selection (SSD-FS) method based on SOLDA.
• To enhance feature selection by ensuring minimal and discriminative feature subsets.

Main Methods:

• Introduced sparsity-inducing regularization to the SOLDA problem.
• Employed row-sparse projection for feature selection.
• Incorporated orthogonal constraints to minimize the number of selected features.

Main Results:

• The proposed SSD-FS method demonstrates superiority over existing sparse feature selection approaches.
• Row-sparse projection prevents over-suppression of relevant features.
• Orthogonal constraints ensure a minimal set of selectable features.
• The method effectively harnesses discriminant power to select discriminative features.

Conclusions:

• The novel SSD-FS method effectively selects discriminative features using a self-weighted orthogonal approach.
• The method is theoretically sound and experimentally validated.
• SSD-FS offers an improved approach to supervised feature selection.