A study published online on 17 September 2026 in the Journal of Dairy Science has built a data-driven method to classify the physiological profiles of dairy cows using milk mid-infrared spectra and machine learning.

The research team comprised Y. Chen, H. Atashi, S. Franceschini, C. Grelet, C. Nickmilder, P. Lemal, K. Wijnrocx, H. Soyeurt and N. Gengler. Their work applies a data-driven classification approach to a routine, non-invasive sample that dairy farms already collect at every milking: milk itself.

Why does milk mid-infrared spectroscopy matter?

Mid-infrared (MIR) spectroscopy is the standard technology used in milk recording schemes to determine fat, protein and lactose content. Because the same spectral signal carries information beyond gross composition, researchers have increasingly used it as a phenotyping tool for traits that are otherwise costly or impractical to measure at scale.

In this study, the authors directed that idea toward physiological profile classification. The goal is to assign cows to physiological status classes from spectral data, rather than relying on separate diagnostic sampling or manual observation.

For practitioners and herd managers, the appeal is straightforward. If classification works reliably, the information arrives with the routine milk sample, at herd level, with no extra handling of animals.

What does the study actually do?

The paper, titled "A data-driven physiological profile classification for dairy cows using milk mid-infrared spectra and machine learning," pairs MIR spectral data with machine learning models to sort cows into physiological profile categories. The authors describe the approach as data-driven, meaning the classes and the classification rules derive from the data rather than from fixed pre-set thresholds.

The Journal of Dairy Science published the article online on 17 September 2026. The journal did not assign the work to a regulatory track; this is basic and applied research, not a diagnostic product with agency clearance.

The study involves no drug, vaccine or device approval. Any move from this methodology to an on-farm decision-support tool would come later, through validation and commercial development.

Where could this fit at the farm level?

For food-animal practice, the practical interest lies in what routine milk recording could deliver at no additional sampling cost. A physiological profile classifier built on MIR spectra could, if validated, support monitoring of cow status across the lactation cycle at herd scale.

That would place physiological surveillance in the same data stream farms already use for payment testing and genetic evaluation. The burden would shift from individual animal diagnostics to population-level screening, with veterinarians acting on flagged animals.

Companion-animal practice sees no direct effect from this work. Public-health implications are likewise indirect at this stage: the study concerns cow physiology classification, not milk safety or residue monitoring.

What remains before adoption?

The publication marks a research step, not a product launch. Validation across herds, breeds and management conditions, plus integration into milk recording software, would determine whether the classifier reaches farm decision support.

The authors' dataset and model details appear in the Journal of Dairy Science article released on 17 September 2026.