Machine-Learning Techniques Can Enhance Dairy Cow Estrus Detection Using Location and Acceleration Data.

Area of Science:

• Animal Science
• Agricultural Technology
• Machine Learning

Background:

• Estrus detection in dairy cows is crucial for reproductive management.
• Traditional methods can be labor-intensive and inaccurate.
• Technological advancements offer potential for improved detection.

Purpose of the Study:

• To evaluate the efficacy of combining location, acceleration, and machine learning for dairy cow estrus detection.
• To compare four machine learning algorithms for identifying estrus indicators.
• To determine the optimal time window for prediction.

Main Methods:

• Continuous monitoring of 12 dairy cows for 12 days using neck-mounted devices.
• Collection of location and acceleration data (25,684 records).
• Application of K-nearest neighbor (KNN), back-propagation neural network (BPNN), linear discriminant analysis (LDA), and classification and regression tree (CART) algorithms.

Main Results:

• Neck tag positioning accuracy was static at 0.25 ± 0.06 m and dynamic at 0.45 ± 0.15 m.
• Machine learning models achieved high performance metrics (e.g., sensitivity up to 99.4%, accuracy up to 95.4%) across different time windows.
• The BPNN algorithm with a 0.5-hour time window demonstrated the highest predictive capability for estrus detection.

Conclusions:

• Integration of location, acceleration, and machine learning significantly enhances dairy cow estrus detection.
• The BPNN algorithm shows promise as a reliable tool for automated estrus identification.
• This technology can lead to more efficient and effective dairy herd management.