Capped L_2,p-norm metric based robust least squares twin support vector machine for pattern classification.

Area of Science:

• Machine Learning
• Pattern Classification
• Robust Statistics

Background:

• Least squares twin support vector machine (LSTSVM) is effective for pattern classification.
• LSTSVM's squared L2-norm metric is sensitive to outliers, potentially magnifying their influence.
• Robustness in classification is crucial for reliable data analysis.

Purpose of the Study:

• Propose a novel robust least squares twin support vector machine (CL2,p-LSTSVM) framework for binary classification.
• Introduce a capped L2,p-norm distance metric to mitigate the impact of noise and outliers.
• Enhance the robustness of LSTSVM while retaining its efficiency.

Main Methods:

• Developed the CL2,p-LSTSVM framework utilizing a capped L2,p-norm distance metric.
• Designed an iterative algorithm to optimize the non-convex metric, solving two linear systems per iteration.
• Extended the framework for nonlinear and semi-supervised classification tasks.

Main Results:

• The capped L2,p-norm metric effectively reduces the influence of outliers and noise.
• Experimental results on artificial, UCI, and image datasets demonstrate superior robustness compared to state-of-the-art methods.
• The proposed iterative algorithm shows efficient convergence and manageable computational complexity.

Conclusions:

• CL2,p-LSTSVM offers a robust and effective solution for binary classification problems with outliers.
• The novel capped L2,p-norm metric provides a flexible way to control robustness.
• The method's extensions to nonlinear and semi-supervised learning broaden its applicability.