Maximum margin and global criterion based-recursive feature selection.

Area of Science:

• Machine Learning
• Data Science
• Computational Statistics

Background:

• Recursive Feature Elimination (RFE) and its variants face limitations in high-dimensional feature selection.
• Existing RFE methods use feature ranking criteria inconsistent with maximum-margin theory and local computations.
• This leads to suboptimal feature importance assessment and reduced selection accuracy.

Purpose of the Study:

• To address the limitations of RFE in high-dimensional feature selection.
• To propose a novel feature ranking criterion, Maximum Margin and Global (MMG), that aligns with maximum-margin theory and global feature importance.
• To introduce efficient algorithms and alpha seeding strategies for optimal feature subset selection.

Main Methods:

• Developed the Maximum Margin and Global (MMG) criterion for feature ranking.
• Introduced an optimal feature subset evaluation algorithm utilizing the MMG criterion.
• Implemented two alpha seeding strategies to enhance computational efficiency.

Main Results:

• The proposed MMG criterion and algorithms demonstrate superior performance compared to state-of-the-art methods across ten benchmark datasets.
• Extensive experiments confirm the effectiveness of the MMG criterion in accurately assessing global feature importance.
• The alpha seeding strategies significantly reduce computational costs while maintaining high accuracy.

Conclusions:

• The MMG criterion offers a more theoretically sound and globally aware approach to feature selection in high-dimensional spaces.
• The proposed algorithms and seeding strategies provide an efficient and accurate solution for feature selection tasks.
• This research advances feature selection methodologies, offering practical benefits for machine learning applications.