Articles

The goal of building a multivariate calibration model is to predict a chemical or physical property from a set of predictor variables, e.g. analyte concentration or octane number from a near infrared (NIR) spectrum. A good multivariate calibration model should be able to replace the laborious, possibly imprecise reference method. The quality of a model therefore primarily depends on its predictive ability. Other properties such as interpretability of the model coefficients might also be of interest, but here the focus is on the problem of quantifying the predictive ability.

A.M.C.Davies,^a and Tom Fearn,^b

^aNorwich Near Infrared Consultancy, 75 Intwood Road, Cringleford, Norwich NR4 6AA,UK
^bDepartment of Statistical Science, University College London, Gower Street, London

Tony Davies

Norwich Near Infrared Consultancy , 75 Intwood Road, Cringleford, Norwich NR4 6AA, UK

Tom Fearn

Department of Statistical Science, University College London, Gower Street, London,WC1E 6BT, UK

A.M.C. Davies

Norwich Near Infrared Consultancy, 75 Intwood Road, Cringleford, Norwich, NR4 6AA, UK

At the end of the last column we promised that this time we would show how matrix algebra can be used for real computational tasks. The chosen task is Fourier transformation (FT) of a near infrared (NIR) spectrum. Those who know Tony Davies will not be surprised at this choice of subject but in the third lesson the reason for wanting to do the obvious will become apparent.

The very first Tony Davies Column.