Using machine learning to improve bovine tuberculosis control in herd level outbreaks.

Area of Science:

• Veterinary Medicine
• Animal Health
• Machine Learning in Epidemiology

Background:

• Bovine tuberculosis (bTB) causes significant economic losses and lacks a definitive antemortem diagnostic test.
• Current control measures include the tuberculin skin test (CICT) and interferon gamma (IFN-γ) assay, but these have limitations.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting true tuberculosis infection in cattle.
• To enhance the quality assurance of existing bTB surveillance schemes.

Main Methods:

• Utilized data on tuberculin regression in cattle reactors following CICT.
• Applied machine learning techniques including Decision Trees, Bagging Trees, and Random Forests with balancing approaches (e.g., SMOTE).

Main Results:

• Random Forests (RF) trained with SMOTE balancing achieved the highest accuracy (0.90).
• The model identified the importance of IFN-γ assay components, suggesting threshold adjustments for large outbreaks.
• Combined RF and IFN-γ models showed potential for improved infection detection in breakdown herds.

Conclusions:

• Machine learning models, particularly RF, can significantly improve bTB detection accuracy.
• These models can enhance surveillance, reduce outbreak scale and duration, and support national disease control plans.
• The approach offers a mechanism for quality assurance of current bTB surveillance strategies.