Sparse solution of least-squares twin multi-class support vector machine using ℓ₀ and ℓ_p-norm for classification and...

Area of Science:

• Machine Learning
• Computer Science

Background:

• Multi-class classification challenges exist, particularly with high-dimensional data.
• Existing methods like Twin K-class Support Vector Classification (Twin-KSVC) and its improvements (ILSTKSVC) influence classification accuracy but often neglect feature selection.

Purpose of the Study:

• To propose a novel method, ℓp-norm least-squares twin multi-class support vector machine (PLSTKSVC), that simultaneously performs classification and feature selection.
• To address the limitations of prior studies by integrating feature selection into the multi-class classification framework.

Main Methods:

• Developed PLSTKSVC, inspired by ILSTKSVC and cardinality-constrained optimization, using 0
• The optimization problems are solved efficiently by solving systems of linear equations for an approximate solution.
• Investigated the properties of optimum solutions under specific assumptions.

Main Results:

• The proposed PLSTKSVC method demonstrated superior performance across several real-world datasets.
• Achieved higher classification accuracy compared to existing strategies.
• Successfully reduced the number of features, indicating effective feature selection.

Conclusions:

• PLSTKSVC offers a robust and efficient approach for simultaneous multi-class classification and feature selection.
• The method shows significant potential for applications involving high-dimensional datasets where both accuracy and dimensionality reduction are crucial.