Development of a Novel Classification Approach for Cow Behavior Analysis Using Tracking Data and Unsupervised Machine...

Area of Science:

• Animal Science
• Machine Learning
• Geographic Information Systems

Background:

• Global Positioning Systems (GPS) enable remote livestock monitoring for well-being and pasture use.
• Supervised machine learning for behavior identification is labor-intensive due to required animal observations.

Purpose of the Study:

• To develop an automated method for identifying cattle behaviors using unsupervised learning techniques.
• To eliminate the need for human observations in analyzing livestock behavior from GPS data.

Main Methods:

• A two-step framework was designed: time series segmentation of GPS data followed by cluster analysis and labeling.
• Unsupervised learning techniques were applied to GPS tracking data from five cows in a rangeland pasture.
• Cattle movement pathways were clustered based on velocity and distance from water, then classified into walking, grazing, and resting behaviors.

Main Results:

• Six distinct behavior clusters were identified from cow movement data.
• The framework successfully classified behaviors into walking (mean velocity 44 m/min), grazing (13 m/min), and resting (2 m/min).
• Predicted diurnal patterns revealed typical grazing bouts in the early morning and evening.

Conclusions:

• The proposed unsupervised framework effectively predicts cattle behavior from unlabeled GPS tracking data.
• This approach offers a labor-saving alternative to traditional supervised methods for livestock behavior analysis.
• The findings demonstrate the applicability of advanced machine learning for ecological and agricultural monitoring.