A systematic review of machine learning approaches to bovine tuberculosis in cattle

Bovine tuberculosis (bTB) is a chronic zoonotic disease, caused by Mycobacterium bovis which despite years of eradication attempts, is still prevalent globally. Machine learning (ML), a technology allowing for computerised discovery of patterns in data, has been applied widely to the field of epidemiology for decades, more recently venturing into the detection and diagnosis of bTB with varying success. This systematic review was performed to identify the existing published research utilising ML with bTB datasets oriented directly or indirectly around the identification of infected animals or infected herds. The search strategy was formulated and compiled eligible publications and reviewed their approach, methodologies. A total of 19 publications were identified fitting the standardised criteria indicating this field is still in its early stages. The most frequently used models were Random Forests and Logistic Regression. Fifty two percent of the studies utilised more than one algorithm. One of the main challenges identified through this review is the frequent lack of transparency in the reported methodologies. We also recommend that model interpretability should be treated on par with the model performance as only 2 out of the 19 publications considered the importance of model interpretability. This review provides a useful foundation for understanding ML applications in bTB control and broader veterinary epidemiology and also demonstrates the need for specific reporting guidelines for ML in veterinary epidemiology to maximise the potential of these methods.