A novel behavioral model of the pasture-based dairy cow from GPS data using data mining and machine learning...

Area of Science:

• Animal behavior
• Precision agriculture
• Machine learning applications in animal science

Background:

• Understanding dairy cow behavior is crucial for enhancing productivity and welfare.
• Manual observation of grazing herds is labor-intensive and challenging.
• Global Positioning Systems (GPS) offer a potential solution for remote monitoring of cattle.

Purpose of the Study:

• To evaluate machine learning algorithms for predicting dairy cow behaviors using GPS data.
• To develop an accurate and robust model for classifying grazing, resting, and walking behaviors.
• To assess the performance of different algorithms in accounting for inter-cow variation and environmental factors.

Main Methods:

• Collected GPS data from 40 pasture-based dairy cows over four months.
• Extracted 43 features from raw GPS data for analysis.
• Employed machine learning algorithms, including JRip, with 10x10-fold cross-validation and an independent test set.

Main Results:

• The JRip rule-based classifier achieved a classification accuracy of 0.85.
• The model demonstrated a false positive rate of 0.10 and an F-measure of 0.76.
• The area under the receiver operating curve was 0.87, indicating strong predictive performance.

Conclusions:

• Machine learning, particularly the JRip algorithm, effectively predicts dairy cow behaviors from GPS data.
• The developed model shows promise for improving the assessment of pasture-based dairy cow behavior and welfare.
• This approach can reduce the need for manual observation, offering a scalable solution for herd monitoring.