Abstract

Journal of Near Infrared Spectroscopy
Volume 12 Issue 2, Pages 93–100 (2004)
doi: 10.1255/jnirs.412

Least-squares support vector machines for chemometrics: an introduction and evaluation

R.P. Cogdill^a,* and P. Dardenne^b
aDuquesne University Center for Pharmaceutical Technology, 410 Mellon Hall, 600 Forbes Ave., Pittsburgh, PA 15282, USA. E-mail: cogdillr@duq.edu
^bCentre de Recherches Agronomiques de Gembloux, Chaussée de Namur, B-5030 Gembloux, Belgium

Support vector machines (SVM) are a relatively new technique for modelling multivariate, non-linear systems, which is rapidly gaining acceptance in many fields. There has been very little application or understanding of SVM methodology in chemometrics. The objectives of this paper are to introduce and explain SVM regression in a manner that will be familiar to the NIR and chemometrics community, and provide some practical comparisons between least-squares SVM regression and more traditional methods of multivariate data analysis. Least squares support vector machines (LS-SVM) were compared to partial least squares (PLS), LOCAL and artificial neural networks (ANN) for regression and classification using four, diverse datasets. LS-SVM was shown to be the most effective algorithm, and required the lowest number of calibration samples to achieve superior predictive performance.

Keywords: chemometrics, support vector machines, artificial neural networks, linear regression, non-parametric regression, radial basis function

---

Full-text article (215 kB) (subscribers only)

Buy article on-line for £20 (get immediate access)

Alerting Service

 RSS Feed

Permalink: http://dx.doi.org/10.1255/jnirs.412
QR Code (what is this?):

---