Outbreak Prediction in Swine Populations with Machine Learning

Abstract The pork industry is an essential part of the global food system, providing a significant source of protein for people around the world. A major factor restraining productivity in the pork industry is disease outbreaks in pigs throughout the production process: widespread outbreaks can lead to losses as high as 10% of the U.S. pig population in extreme years. In this study, we present a model to predict the emergence of outbreaks of swine farms throughout the production process. We capture direct contact, spatio-temporal and historical predictors, each represented through a set of features, and then train and evaluate machine learning algorithms on our extracted feature sets. We perform a feature selection to determine the smallest subset of features that provides good performance and use the results to interpret the most valuable features and produce the most generalizable model to address issues caused by the curse of dimensionality. Finally, we evaluate the model's ability to predict outbreaks in both the near and distant future, which allows for advance warning of disease outbreak. We evaluate our model on two swine production systems; our results demonstrate good ability to predict outbreaks in both systems with a balanced accuracy of 0.798 on any disease in the first system and balanced accuracies of 0.638, 0.709, and 0.701 on porcine reproductive and respiratory syndrome, influenza A virus, and Mycoplasma hyopneumoniae in the second system, respectively.