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

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.