Mid-infrared predictions of cheese yield from bovine milk

Economically, cheese yield (CY) is very important. Todate, empirical or theoretical formulae allow estimating the theoretical CY from milk fat and casein or protein content of milk. It would be interesting to predict CY during milk recording directly without the need to estimate milk components. Through the BlueSel project, 157 milk samples were collected in Wallonia from individual cows and analyzed using a mid-infrared (MIR) MilkoScanFT6000 spectrometer. Individual laboratory cheese yields (ILCY) were determined for each sample and expressed as g of dry coagulum/100 g of milk dry matter. An equation to predict ILCY from MIR was developed using partial least squared regression (Winisi III). A first derivative pre-treatment of spectra was used to correct the baseline drift. To improve the repeatability of the spectral data, a file which contained the spectra of samples analyzed on 5 spectrometers was used during the calibration. During calibration, 23 outliers were detected a nd removed from the calibration set. The ILCY mean of the final calibration set was 63.9% with a SD of 11.2%. The calibration (C) coefficient of determination (R²) was equal to 0.76 with a standard error (SE) of calibration of 5.5%. A full cross-validation (CV) was preformed to assess the robustness. R²cv was 0.72 with a SECV of 6.0%. The similarity between R²c and R²cv as well as between SEC and SECV permits to consider robustness of the developed equation as good. Even if it is planned to improve the equation with additional samples, this first equation will permit to study ILCY in the Walloon dairy cattle.