Community Research and Development Information Service - CORDIS

Abstract

Today, PLS (partial least squares) regression has become an invaluable tool for multivariate analysis of chemical data. In analytical chemistry, PLS is frequently used for developing predictive models between different blocks of data, often denoted as multivariate calibration. In typical cases of multivariate calibration, the predictor variable matrix X contains a large number of variables (K = 100-10000) and most of the attention is paid to the predictive performance of the model. However, questions which frequently appear are: (i) do all variables contain important information for the model; (ii) is it possible to improve the predictive ability by variable selection? Recently, an interactive variable selection (IVS) technique has been developed for PLS. The method is based on dimensional-wise selective reweighting of single elements in the PLS weight vector W. Cross validation, a criterion for estimation of predictive quality, is used for guiding the selection procedure in the modelling stage. Improvements of up to 70 % in external PRESS over the classical PLS algorithm have been recorded. Both the philosophy and the use of IVS-PLS are presented together with its application to a typical case of multivariate calibration. The validity of the technique is demonstrated using a randomisation test.

Additional information

Authors: LINDGREN F, JRC Ispra (IT)
Bibliographic Reference: Paper presented: 3rd Symposium on Analytical Sciences, Paris (FR), March 28-30, 1995
Availability: Available from (1) as Paper EN 38931 ORA
Record Number: 199510665 / Last updated on: 1995-07-07
Category: PUBLICATION
Original language: en
Available languages: en