Robust least squares twin bounded support vector machine with a generalized correntropy-induced metric

Yuan, Chao; Zhou, Changsheng; Pan, Honghao; Arvin, Farshad; Peng, Jigen; Li, Haiyang

Abstract

The least squares twin support vector machine (LSTSVM), which aims to seek nonparallel hyperplanes by solving two linear equations, has received extensive attention in statistical theory as a powerful and widely used method for addressing classification problems. However, LSTSVM may show a decline in performance when confronted with corrupt and noisy data, and its robust discriminative ability cannot be guaranteed. To address this issue, a robust framework known as the least squares twin bounded support vector machine with a generalized correntropy-induced metric (CLSTBSVM) is proposed to suppress the impact of noise and outliers. Firstly, we define a novel non-second-order statistical local similarity measure, which is an extension of the classical correntropy-based metric and replaces the squared L2-norm distance in LSTSVM. Some basic properties related to this robust metric are analyzed. Moreover, a probability weight is assigned to each sample to evaluate whether it is a normal data point. The physical interpretation of this is straightforward: a normal point is given a probability value of 1, while an abnormal point is assigned a weight of 0. An efficient iterative algorithm is designed for solving the optimal problem. This algorithm is easy to implement, and its convergence to an optimum is theoretically guaranteed. Finally, experiments on various types of datasets demonstrate the robustness of CLSTBSVM.

Citation

Yuan, C., Zhou, C., Pan, H., Arvin, F., Peng, J., & Li, H. (2025). Robust least squares twin bounded support vector machine with a generalized correntropy-induced metric. Information Sciences, 699, Article 121798. https://doi.org/10.1016/j.ins.2024.121798

• Journal Article Type: Article
• Acceptance Date: Dec 18, 2024
• Online Publication Date: Dec 20, 2024
• Publication Date: 2025-05
• Deposit Date: Feb 25, 2025
• Journal: Information Sciences
• Print ISSN: 0020-0255
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Volume: 699
• Article Number: 121798
• DOI: https://doi.org/10.1016/j.ins.2024.121798
• Public URL: https://durham-repository.worktribe.com/output/3503407