Automating classification of veterinary biosecurity recommendations using machine learning.

Area of Science:

• Veterinary Medicine
• Machine Learning
• Data Science

Background:

• The ProAction® program mandates biosecurity risk assessments for Canadian dairy farmers.
• Veterinarians provide personalized biosecurity recommendations, generating substantial text data.
• Automating the classification of these recommendations can enhance farm management.

Purpose of the Study:

• To develop and evaluate a machine learning model for classifying veterinary biosecurity recommendations into 12 categories.
• To assess the performance of different machine learning algorithms for this classification task.
• To determine the model's effectiveness in classifying new, unseen data.

Main Methods:

• Text data from 11,250 veterinary recommendations (2018-2021) were translated to French for consistency.
• Three algorithms (Multinomial Naïve Bayes, Support Vector Machine, Random Forest) were trained and compared.
• Performance was evaluated using precision, recall, and F1-score, with a separate validation on new data (Cohen's Kappa).

Main Results:

• The Support Vector Machine (SVM) algorithm demonstrated the highest performance and efficient processing.
• The trained SVM model achieved a Cohen's Kappa of 0.88 on a new dataset, indicating strong agreement with human classification.
• The model successfully classified biosecurity recommendations from diverse Canadian dairy herds.

Conclusions:

• Machine learning offers a powerful tool for automating the classification of biosecurity recommendations in dairy farming.
• This technology can support timely and informed decision-making for improved herd management and biosecurity.
• The developed SVM model shows significant potential for practical application in veterinary and agricultural sectors.