Automated Cattle Behavior Classification Using Wearable Sensors and Machine Learning Approach.

This paper focuses on automating the classification of in-house cattle behavior using collar tags equipped with tri-axial accelerometers to collect data on feeding and ruminating behaviors. The accelerometer data is divided into time intervals (10, 30, 60, and 180 s), and we extract 15 essential posture-related features to create a labeled dataset for behavior classification. We evaluate four machine learning algorithms (Random Forest, Extreme Gradient Boosting, Decision Tree, and Logistic Regression) on this dataset using leave-one-out cross-validation. The results indicate that shorter time intervals result in better prediction performance. Random Forest and Decision Tree algorithms perform well, striking a good balance between sensitivity and specificity. This proposed approach holds promise for real-time behavior classification and has the potential to benefit livestock management and enhance animal welfare.