Predictive modelling of deviation from expected milk yield in transition cows on automatic milking systems

Area of Science:

• Animal Science
• Dairy Production
• Veterinary Medicine

Background:

• The transition period (calving to early lactation) is critical for dairy cows, with 30-50% experiencing disease.
• Metabolic and infectious diseases reduce early lactation milk production, impacting farm economics.
• Deviation from expected milk yield is used retrospectively to assess transition cow health.

Purpose of the Study:

• To evaluate if early lactation data (1-3 days in milk) from automatic milking systems (AMS) can predict 30-day cumulative milk yield deviation in multiparous cows.
• To assess if predicted yield deviations can classify cows for improved transition management.

Main Methods:

• A retrospective cohort study analyzed production, rumination, and activity data from 31 commercial AMS.
• A 3-step analysis involved calculating expected yield, determining 30-Day Yield Deviation (YD), and using machine learning to predict YD from early lactation data.
• Cows were grouped into RED (0% YD) based on YD.

Main Results:

• Machine learning models predicted 30-Day Yield Deviation with a mean absolute error of 9% after external validation.
• Classification of cows in the RED group (large negative deviations) achieved 99% specificity, 35% sensitivity, and 67% balanced accuracy.
• Early lactation milk yield, rumination, and activity patterns from days 1-3 in milk showed utility in predicting yield deviation.

Conclusions:

• Dairy cow milk yield, rumination, and activity from 1-3 days in milk can predict deviations in 30-day cumulative milk yield.
• Current prediction models lack the necessary sensitivity for reliable classification of cows to facilitate improved transition cow management.