Twin support vector regression with Huber loss

Article type: Research Article

Authors: Niu, Jiayi | Chen, Jing^* | Xu, Yitian^*

Affiliations: College of Science, China Agricultural University, Beijing, China

Correspondence: [*] Corresponding author. Jing Chen and Yitian Xu, College of Science, China Agricultural University, Beijing 100083, China. Tel.: +8610 62737077; Fax: +8610 62736563; E-mail: [email protected].

Abstract: ɛ-insensitive loss function is often employed in the twin support vector regression (TSVR). However, it can not effectively address the data with Gaussian noise. Huber loss function can suppress a variety of noise and outliers and yields great generalization performance. Motivated by this, we propose a novel twin support vector regression with Huber loss for the noise data in this paper. Experiments on nine benchmark datasets with different Gaussian noise show the validity of our proposed algorithm. Finally, we apply our method to the financial time series data that usually contain noise and outlier, and it also produces great performance.

Keywords: Twin support vector regression, Gaussian loss, Huber loss, Gaussian noise

DOI: 10.3233/JIFS-16629

Journal: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 6, pp. 4247-4258, 2017