Improving bovine disease detection through multilabel classification.

Area of Science:

• Veterinary Medicine
• Agricultural Technology
• Data Science

Background:

• Dairy industry productivity is hampered by bovine health issues (lameness, mastitis, etc.), impacting milk yield and animal welfare.
• There is a critical need for intelligent, data-driven monitoring systems for early detection of these conditions.

Purpose of the Study:

• To propose a machine learning (ML)-based framework for early detection of bovine health events and diseases using multi-label classification.
• To identify conditions such as estrus, calving, lameness, mastitis, and acidosis through behavioral metrics.

Main Methods:

• Utilized a machine learning framework for multi-label classification of bovine health events.
• Analyzed key behavioral metrics from sensor-based monitoring (feeding, resting, locomotion, activity).
• Employed SMOTE (Synthetic Minority Over-sampling Technique) and Classifier Chains for multi-label prediction, leveraging label interdependencies.

Main Results:

• Tested on a large dataset of 2.35 million livestock behavioral records.
• An Extra Tree Classifier within a classifier chain configuration achieved 97% subset accuracy, 96% recall, 95% precision, and 96% F1-score.
• Demonstrated superior performance over standard binary relevance approaches, with a Hamming loss of 0.04.

Conclusions:

• Classifier chains combined with oversampling techniques effectively capture label correlations for improved bovine disease prediction.
• The developed ML framework offers a promising solution for early detection, enhancing dairy herd health management and animal welfare.