Does the age of milk affect its mid-infrared spectrum and predictions?

•Infrared-spectroscopy plus chemometrics was used to predict milk of different age.•Infrared-predicted traits of milk were recorded longitudinally on the same samples.•The most reproducible traits across the sample ages were fat, protein, and casein.•14 latent variables allowed to achieve the best accuracy in discriminant analysis.•Discriminant ability was promising with an accuracy of 77% in the validation set. Milk of dairy species commonly undergo standardized official analyses, these that may require chemical preservation and transportation to a certified laboratory. In this context, storage duration is an important factor that can potential affect both milk chemical analyses and its mid-infrared spectrum. We analysed milk samples at different time points/ages to assess repeatability and reproducibility of mid-infrared predicted traits (e.g., fat and protein). Using spectral data, we also evaluated the ability of spectroscopy coupled with chemometrics to discriminate samples according to their age. Although the main components of milk remained consistently reproducible across age (days), changes in the spectrum due to sample aging and deterioration of the matrix were detectable. Using a discriminant analysis, we achieved a classification accuracy of 77% in validation. Predicting milk age using mid-infrared spectra is feasible, allowing for sample monitoring within circuits where maximum reliability is needed, e.g., bulk or individual milk samples for legal/official use or payment systems.