EXAMINATION OF CRITERIA FOR LOCAL MODEL PRINCIPAL COMPONENT REGRESSION

In analytical chemistry, principal component regression (PCR) is widel y used as a method for calibration and prediction. The motivation behi nd PCR is to select factors associated with predictive information and eliminate those associated with noise. The classical approach, referr ed to as top-down selection, chooses sequential factors based on singu lar value magnitudes, and the same factors are used for all future unk nown samples; i.e., a global model is formed. The number of factors ne eded is often determined through cross-validation on the calibration s amples or with an external validation set. Alternatively, a model deve loped specific to an unknown sample, i.e., a local model or sample-dep endent model, could offer improved accuracy. The idea behind sample-de pendent PCR is that factors associated with small singular values not included in a top-down PCR model can still contain relevant predictive information. This paper shows that local models generated by selectin g factors on a sample-by-sample basis often reduce prediction errors c ompared with those for the global top-down model. However, evidence is also provided that supports the use of global top-down models. Severa l criteria are proposed and examined for selecting factors on a sample -dependent basis. Observations and conclusions presented are based on two near-infrared data sets.