Machine-learning-based calving prediction from activity, lying, and ruminating behaviors in dairy cattle

Area of Science:

• Animal Science
• Agricultural Engineering
• Machine Learning

Background:

• Prepartum dairy cattle behavior monitoring is crucial for timely calving management.
• Automated sensors offer objective data collection for behavioral analysis.

Purpose of the Study:

• To characterize prepartum behavior in dairy cattle using automated activity, lying, and rumination monitors.
• To evaluate the potential of these monitors in predicting calving events.

Main Methods:

• Data collected from Holstein dairy cattle using HR Tag and IceQube sensors for 14 days prepartum.
• Mixed linear models analyzed daily and bihourly behavioral changes (activity, lying, rumination).
• Machine learning techniques (random forest, LDA, neural networks) assessed prediction accuracy.

Main Results:

• Significant changes in rumination time, total motion, lying time, and lying bouts were observed in the 14 days before calving.
• Extreme behavioral values occurred within the final 24 hours prepartum.
• A neural network combining both technologies achieved 100.0% sensitivity and 86.8% specificity for daily prediction.

Conclusions:

• Commercially available behavioral monitors show potential for calving prediction in dairy cattle.
• Machine learning analysis of monitored behaviors can provide valuable calving alerts.
• Automated monitoring systems can aid in optimizing dairy herd management and welfare.