Behavior Classification of Cattle in a Virtual Fencing System Using Tri-Axial Accelerometers and Machine Learning

Area of Science:

• Animal Behavior
• Precision Livestock Farming
• Sensor Technology

Background:

• Virtual fencing offers a flexible alternative to physical fences in livestock management.
• Detailed behavioral assessments within virtual fencing systems are limited.
• Understanding cattle behavior is crucial for optimizing grazing management and animal welfare.

Purpose of the Study:

• To investigate the use of collar-mounted accelerometers and machine learning to characterize cattle behavior in a virtual fencing system.
• To estimate individual behavioral time budgets and analyze spatio-temporal activity patterns.
• To assess the impact of virtual fence warnings on cattle behavior.

Main Methods:

• Seven Angus cattle were monitored using tri-axial accelerometers, GNSS, and virtual fence warning logs.
• A random forest classifier was trained to identify behaviors: grazing/feeding, ruminating, lying, standing, and locomotion.
• Behavioral data were analyzed spatially and in relation to virtual fence interactions.

Main Results:

• The accelerometer-based model achieved high accuracy (mean 0.87) in classifying key cattle behaviors.
• Cattle exhibited increased grazing and reduced ruminating near virtual fence boundaries.
• Habituated cattle showed no consistent short-term behavioral disturbance following virtual fence warnings.

Conclusions:

• Accelerometer-based behavior classification provides non-invasive, fine-scale insights into cattle behavior in virtual fencing systems.
• Virtual fencing, in habituated herds, does not appear to cause significant disruption to measured behavioral patterns.
• Embedded sensor data holds potential for advanced animal-based behavioral monitoring in livestock management.