Association of artificial intelligence-predicted milk yield residuals to behavioral patterns and transition success in multiparous dairy cows

Abstract

Data-driven health monitoring based on milk yield has shown potential to identify health-perturbing events during the transition period. As a proof of principle, we explored the association between the cow's residual milk yield, that is, the difference between the actual and expected milk yield, and the behavioral parameters of cows during the transition period, as measured by a neck and leg activity sensor. Cows from 8 Dutch commercial dairy farms were equipped with accelerometer sensors to study their time budgets, including eating, rumination, lying, and standing times. Daily sensor data of 2,689 lactations were used from 21 d prepartum until 21 d postpartum. The expected milk yield in the current transition period was predicted using a previously developed artificial intelligence model using low-frequency test day data from the previous lactation. The expected milk yields were subtracted from the actual productions to calculate the milk yield residuals in the transition period (MRT). Three milk residual categories (low, medium, high) were subsequently defined, and behavioral differences between the categories were studied. Postpartum eating times for cows in the high MRT category were consistently higher compared with the low MRT category, with differences ranging from 11.08 min (95% CI: 0.31-21.85 min) to 19.89 min (95% CI: 9.62-30.16min) per day. Rumination time in the 21 d after calving was lower in the category with the most negative milk yield residuals (low MRT) compared with both the other categories, with differences up to 36.12 min (95% CI: 24.63-47.62 min), while standing times after calving were highest in the low MRT category. Longer lying times were observed in the low MRT category on d 1 postpartum and at the end of the observation period (d 18-20). No significant differences were observed in eating, rumination, and standing times across the different MRT categories during the prepartum period. For lying times, significant effects were identified between the different MRT categories on certain days, with cows in the high MRT category exhibiting the longest lying times, with differences ranging between 19.13 min (95% CI: 0.47-37.80 min) and 24.82 min (95% CI: 7.32-42.32 min) compared with the low category. Cows with more negative milk yield residuals during the first 21 d after calving exhibited postpartum behavioral patterns associated with a compromised transition. Results of the present study suggest the potential application of MRT as a metabolic indicator for transitioning cows, which could support the development of new health monitoring tools.