Premium
Prediction and spectral profile estimation in multivariate calibration
Author(s) -
Trygg Johan
Publication year - 2004
Publication title -
journal of chemometrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.47
H-Index - 92
eISSN - 1099-128X
pISSN - 0886-9383
DOI - 10.1002/cem.860
Subject(s) - calibration , chemometrics , multivariate statistics , interpretation (philosophy) , matrix (chemical analysis) , mathematics , partial least squares regression , transformation (genetics) , transformation matrix , linear regression , statistics , biological system , computer science , chemistry , physics , chromatography , biochemistry , kinematics , classical mechanics , biology , gene , programming language
Direct and indirect calibration have been compared with respect to both prediction and model interpretation. This included their ability to estimate the pure spectral profile of each known constituent in a mixture of different metal–ion complexes. In the examples, the predictions by indirect calibration, represented by the PLS and O‐PLS methods, were consistently better than those of direct calibration, exemplified by the CLS method. It was further demonstrated that indirect calibration is equally capable to direct calibration in estimating the pure spectral profiles, as long as the unknown systematic variation is properly handled. A linear transformation of the regression coefficient matrix, given by K = B(B T B) −1 , is all that is needed. Note that this does not only apply to spectral data, but any situation where the Y‐variables can be assumed to additively contribute to the variation in the X matrix. Throughout the examples, the O‐PLS method was able to maintain good spectral profile estimates and predictions. This indicates that O‐PLS may be the approach for simultaneous good prediction and interpretation of complex multivariate systems. Copyright © 2004 John Wiley & Sons, Ltd.