Machine Learning and SHapley Additive exPlanation-Based Interpretation for Predicting Mastitis in Dairy Cows.

Area of Science:

• Veterinary Medicine
• Animal Science
• Machine Learning Applications

Background:

• SHAP analysis is underutilized in dairy cow disease prediction.
• Early detection of clinical mastitis is crucial for herd health and productivity.

Purpose of the Study:

• To evaluate machine learning models for predicting clinical mastitis in dairy cows.
• To identify key predictive features using SHAP analysis.

Main Methods:

• Utilized quantile regression to process data on activity, rumination, milk conductivity, and yield.
• Trained and validated eleven machine learning algorithms, including partial least squares.
• Applied SHAP analysis to interpret model predictions and feature importance.

Main Results:

• The partial least squares model achieved an AUC of 0.789, with high specificity (0.947) and precision (0.833).
• Nine variables from the 14-day pre-mastitis period were significantly associated with the disease.
• SHAP analysis identified specific features with positive and negative contributions to mastitis prediction.

Conclusions:

• Machine learning models, particularly partial least squares, can effectively predict clinical mastitis in dairy cows.
• SHAP analysis provides valuable insights into the factors driving mastitis prediction.
• Findings support the development of decision-support tools for dairy farm management.