Data Driven Modelling for Improving Herd-Level Bovine Tuberculosis Breakdown Predictions in GB Cattle

A data-driven model was used to predict herd-level bovine tuberculosis breakdowns in Great Britain (GB) with the aim of improving diagnosis. The results of single intra-dermal comparative cervical tuberculin (SICCT) tests were correlated with data related to infection risk, e.g., holding size and contact tracing. Four machine learning methods (Neural Network, Random Forest, Gradient Boosted Trees and Support Vector Classifier) were independently trained and optimised with data from 2012–2014 including 4,605–4,818 positive herd-level SICCT test results annually. The performance of the best predictive model was compared to the observed sensitivity and specificity of the herd-level SICCT test calculated on the 2015 testing data. This model performed significantly better in predicting breakdowns, increasing mean herd-level sensitivity from 61.3% to 67.6% (95% confidence interval (CI): 66.4–68.8%) and mean herd-level specificity from 90.5% to 92.3% (95% CI: 91.6–93.1%). The increased sensitivity of the test can help issue better-informed control measures.