Abstract
An unsupervised machine learning approach (ML) of rumination time (RT) data was used to evaluate metabolic and lactation performance in a cohort of Simmental dairy cows (29) around parturition. The k-means clustering (unsupervised ML) was used to generate clusters based on RT (Hr-Tags; SCR by Allflex) over 24 h from 1 to 7-d post-calving. Two large clusters were obtained: high (High-RT, n = 13) and low (Low-RT, n = 12). Milk production was recorded until 42 d in lactation and milk samples collected weekly. Blood samples were collected from -21 to +28 d relative to calving. Data were subjected to PROC MIXED of SAS. Compared with Low-RT, High-RT cows had lower plasma BHB (3, 7, and 14 d), haptoglobin (3 and 7 d), and non-esterified fatty acids (1 and 3 d). High-RT cows had greater tocopherol from 3 to 28 d, fructosamine and albumin at 7 d, and retinol at 3 and 7 d after calving than Low-RT cows. Compared with multiparous Low-RT, milk production was greater in multiparous High-RT cows, but no differences were detected for primiparous. Only for multiparous, High-RT cows displayed a greater LFI than Low-Rt cows. The close relationship between RT and the physiological state at the onset of lactation supports the use of RT as an indicator of metabolic and inflammatory adaptations to the negative energy balance of cows after parturition. At the farm level, these outcomes provide information to farmers that can be helpful in management decisions for cow health, complementing traditional methods.