Behavioural Classification of Cattle Using Neck-Mounted Accelerometer-Equipped Collars.

Area of Science:

• Animal Science
• Machine Learning
• Wearable Technology

Background:

• Optimizing dairy cattle milk yields and reproductive success requires monitoring behaviors like illness onset and oestrus cycles.
• Accelerometer-based sensors offer automated, individual-level insights into cattle welfare and key behaviors.

Purpose of the Study:

• To develop and evaluate algorithms for classifying dairy cattle states using accelerometer data.
• To systematically reduce data dimensionality through feature selection for efficient model deployment.

Main Methods:

• Extracted knowledge-specific and generic time-series features from raw accelerometer data.
• Applied Mutual Information and Backward Feature Elimination for feature selection.
• Trained classification models using Hidden Markov Model, Linear Discriminant Analysis, and Partial Least Squares Discriminant Analysis.

Main Results:

• Linear Discriminant Analysis with Backward Feature Elimination provided the best balance of performance and computational complexity.
• The final model achieved a balanced accuracy of 0.83 with efficient feature extraction and inference times.
• Explored the trade-off between model performance, computational complexity, and memory footprint.

Conclusions:

• The proposed feature engineering methodology enables the deployment of cattle behavior classification models within operational constraints.
• Accelerometer-based systems, coupled with effective algorithms, are valuable tools for dairy cattle management and welfare monitoring.