| Abstract: | A chemometric study on the prediction of the main nutritional aspects of milk has been carried out by using fourier transform infrared spectroscopy (FTIR) attenuated total reflectance (ATR) measurements of commercially available milk samples of different types. Whole, semi and skimmed milks, enriched or not with calcium, vitamins or modified by alteration of lipid or sugar composition were considered. After evaluating different strategies for data acquisition and ATR cleaning between samples, hierarchical cluster analysis (HCA) was carried out for classification of samples in order to choose the calibration set. The prediction capabilities of partial least squares (PLS) data treatment were evaluated in order to obtain information about total fat, total protein, total carbohydrates (CH), calories and calcium. On using the mean square error of cross-validation and prediction as control variables, a critical evaluation were made about the calibration set to be used, the spectral range to be considered and the data treatment (PLS-1 or PLS-2) to be performed. By selecting a calibration set of 33 samples the properties of 48 samples were predicted with relative precision of triplicates of 0.062, 0.040 and 0.039% w/v for total fat, protein and carbohydrates, and 0.66 kcal/100 ml for calories, and 2.1 mg of Ca/100 ml. The mean difference (dx−y) between predicted and actual values and standard deviation of mean differences (sx−y), were of 0.06 (0.38), 0.03 (0.18) and −0.15 (0.41), being sx−y values between brackets, for total fat, proteins and carbohydrates, 0.06 (3.8) kcal/100 ml for calories and −4.5 (9) mg/100 ml for calcium.The sensitivity and selectivity of the methodology developed were evaluated on terms of the net analyte signal. Selectivity factors ranging from 2 to 7.6% have been calculated for the five parameters considered. |
