Knowledge discovery of bovine tuberculosis in the Eurasian badger using machine learning techniques

Bovine tuberculosis (Mycobacterium bovis) is a disease of cattle with severe consequences for agriculture in the British Isles. The Eurasian badger (Meles meles) is implicated in the spread and maintenance of bovine tuberculosis in the cattle population and various measures have been trialed in badgers to control infection. A five-year pilot Test, Vaccinate and Remove investigation (TVR) was carried out in a 100km² area of Northern Ireland that tested, vaccinated, and removed infected badgers. This study used machine learning techniques in order to predict whether a badger has bovine tuberculosis using data collected from the TVR study. Several machine learning models - Decision Trees, Random Forests, Logistic Regression, XGBoost - were created and attempted in order to classify the data with the highest accuracy. Synthetic Minority Oversampling Technique (SMOTE) was also carried out due to imbalance in the data. The C5.0 decision tree model was chosen as the final model. This model was the most appropriate choice as it achieved a very high AUC score with a value of 0.974 in training and 0. 962 in testing. It also had the benefit of being a white-box model. Almost all of the variables were found to be significant, including the visual diagnostic tests used in the study, thus supporting their importance. The final model gives confidence in current diagnostic tests to accurately identify infected badgers and helps to inform future diagnostic test regimes. This study represents one of the first applications of machine learning in wildlife disease control.