Robust least squares support vector machine with the absolute error function

• 발행기관: 한국자료분석학회
• 저자명: Kang-Mo Jung
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.26 No.6, 1661~1670쪽, 전체 10쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2024.12.31

국문 초록

This paper presents a novel approach to improve the robustness of the least squares support vector machine (LS-SVM) for classification tasks by addressing its sensitivity to outliers in support vectors. While LS-SVM is computationally efficient due to its reliance on matrix operations, its use of a squared error loss function makes it highly sensitive to outliers, and it typically requires all data points as support vectors, which reduces model interpretability and efficiency. To overcome these limitations, we propose a LS-SVM variant that employs an absolute error loss function, which reduces the influence of outliers on the decision boundary. However, optimizing the absolute error function poses challenges due to its non-smooth nature. To address this, we apply the split-Bregman iterative method, which efficiently handles non-differentiable optimization problems by decomposing them into sub-problems that converge rapidly. Our method’s performance is evaluated through experiments on benchmark datasets, comparing it to LS-SVM and weighted LS-SVM in terms of classification accuracy, robustness to noise. Results indicate that the proposed model outperforms LS-SVM and weighted LS-SVM by maintaining accuracy in the presence of outliers, leading to improved model interpretability and efficiency.

목차

1. Introduction

2. Least squares support vector classifier

3. The absolute error loss function

4. Numerical experiments

5. Concluding Remarks

References

키워드

Absolute error loss function Classification Least squares support vector machine Robust methods Weighted squares support vector machine

참고문헌

• Proceedings of the International Joint Conference on Neural Networks
• Knowledge Based Systems
• Machine Learning
• SIAM Journal Imaging Science
• Journal of the Korean Data Analysis Society
• Journal of the Korean Data Analysis Society
• Journal of the Korean Data Analysis Society
• Journal of the Korean Data Analysis Society
• Neurocomputing
• Neural Processing Letters
• IEEE International Conference on Data Mining Workshops
• Springer Verlag
• Knowledge Based System
• Neurocomputing
• SIAM Journal of Imaging Science
• Journal of the Korean Data Analysis Society