Multicategory angle-based large-margin classification.

Area of Science:

• Machine Learning
• Computer Science
• Data Science

Background:

• Large-margin classifiers are widely used for classification tasks.
• Current multicategory large-margin classifiers often employ a sum-to-zero constraint, which can lead to computational inefficiencies.

Purpose of the Study:

• To propose a new multicategory angle-based large-margin classification framework.
• To develop a more computationally efficient method for multiclass classification.

Main Methods:

• Introduced a novel angle-based classification framework for multicategory problems.
• Utilized a simplex-based prediction rule, avoiding the sum-to-zero constraint.
• Generalized existing binary large-margin classifiers for multicategory applications.

Main Results:

• The proposed angle-based classifiers demonstrate more efficient computation compared to existing methods.
• Theoretical and numerical studies confirm the effectiveness of the angle-based approach.

Conclusions:

• The angle-based framework provides a computationally efficient and effective alternative for multicategory large-margin classification.
• This method offers a natural generalization pathway for binary large-margin classifiers to multiclass problems.