LAUSR

Abstract A new approach to constructing global calibration models for optical spectroscopic determination of fat content in normalized milk (homogenized milk with standardized nutrient content) in the region 400–1100 nm has been developed. Following the global modelling approach attempting to cover all possible variability of analyzed object, calibration models were built on a large set of commercial milk samples from many producers, collected in two different countries and varying from each other in nutrient composition and methods of technological processing. Suggested method employs spectral information related to the light scatter effect by colloidal particles of the milk constituents. Due to the application of advanced variable interval selection method, the root mean-square error of cross-validation (RMSECV) value of 0.12% in the range 1.5–6.0% of fat content has been achieved in the calibration by partial least-squares (PLS) regression method. Selected spectral intervals can be used for the development of an inexpensive and compact optical sensor for milk analysis based on light-emitting diodes (LEDs). Global calibration models developed in this study can be used for the practical analysis of fat in various normalized milk samples.