Chemometrics and statistics: Neural networks

Abstract

Multivariate calibration based on first-order data, for example, near infrared (NIR) spectra, is now ubiquitous in many analyticalareas, with partial least-squares (PLS) regression being the most popular chemometric technique used for data processing.1 PLSis based on a linear relationship between suitable surrogate variables for the raw multivariate signals and analyte concentrations,although it may tolerate slight deviations from linearity by means of additional latent variables to those required to model linearsystems.