Predicting riboflavin (vitamin B2) and cobalamin (vitamin B12) concentrations in Danish cow milk using Fourier transform mid infrared spectra

This study assessed the potential of Fourier transform mid-infrared (FT-MIR) spectra to estimate vitamins B₂ and B₁₂ in bovine milk. The aim of this study was to evaluate 3 different sampling methods for calibration models to estimate the B₂ and B₁₂ vitamins in milk. Milk samples from 317 Holstein cows were analyzed for vitamin concentrations. Partial least squares regression was used, with data split into calibration and validation sets (80/20) using quartile-based stratification. Three different sampling scenarios were evaluated: Scenario 1 was trained without examples from the extreme values, Scenario 2 included a few (1/3 Q4) extreme values in the training set, Scenario 3 used random training sets. To assess the accuracy of the developed scenarios were repeated 5-fold cross-validation (3 repetitions). This entire process was replicated 1000 times to ensure robust performance assessment. In the calibration dataset, coefficients of determination ranged from 0.10 to 0.27 for vitamin B₂ and from 0.19 to 0.42 for vitamin B₁₂. In the validation dataset, coefficients of determination ranged from 0.01 to 0.25 for vitamin B₂ and from 0.27 to 0.42 for vitamin B₁₂. Overall, FT-MIR-based prediction performance was superior for vitamin B₁₂ compared to vitamin B₂. However, the limited sample size and the restriction to Holstein breed alone may have influenced the results, highlighting the need for further research with larger and more diverse datasets.