Practical selection of SVM parameters and noise estimation for SVM regression

  • Authors:

  • Vladimir Cherkassky;Yunqian Ma

  • Affiliations:

  • Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN;Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN

  • Venue:
  • Neural Networks
  • Year:
  • 2004

Quantified Score

Hi-index 0.02

Visualization

Abstract

We investigate practical selection of hyper-parameters for support vector machines (SVM) regression (that is, ε-insensitive zone and regularization parameter C). The proposed methodology advocates analytic parameter selection directly from the training data, rather than re-sampling approaches commonly used in SVM applications. In particular, we describe a new analytical prescription for setting the value of insensitive zone ε, as a function of training sample size. Good generalization performance of the proposed parameter selection is demonstrated empirically using several low- and high-dimensional regression problems. Further, we point out the importance of Vapnik's ε-insensitive loss for regression problems with finite samples. To this end, we compare generalization performance of SVM regression (using proposed selection of ε-values) with regression using 'least-modulus' loss (ε = 0) and standard squared loss. These comparisons indicate superior generalization performance of SVM regression under sparse sample settings, for various types of additive noise.