Supervised Feature Selection With Orthogonal Regression and Feature Weighting.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• Effective feature selection is crucial for model performance and understanding complex data.
• Existing feature selection methods often fail to retain sufficient discriminative information.

Purpose of the Study:

• To propose a novel supervised orthogonal least squares regression model with feature weighting for enhanced feature selection.
• To address the limitations of traditional methods in retaining discriminative information.

Main Methods:

• A novel supervised orthogonal least squares regression model with feature weighting was developed.
• Generalized power iteration and augmented Lagrangian multiplier methods were employed to solve the optimization problem.

Main Results:

• The proposed method demonstrated superior performance in reducing feature dimensionality.
• Experimental results indicated better classification outcomes compared to traditional feature selection methods.

Conclusions:

• The developed method effectively reduces dimensionality and enhances classification.
• Both theoretical proofs and experimental results validate the method's effectiveness and superiority.