Decision Support for Disease Management in Pig Production via Statistical Modelling

This paper exploits pneumonia treatments recordings from a Danish production facility to develop a decision support system (DSS) for on-site handling of the disease. Simple visualizations of the gathered data provide an intuitive retrospective tool to identify problematic areas within the facility or illustrate implicit strategies applied by the workers. To achieve an operational level DSS, a mathematical model is developed; on a daily basis each section of the production is assigned to one of the two categories: at risk or not at risk. Using logistic discrimination for this decision problem, the effect of past decisions, neighboring sections, and other covariates is included. Model parameters are estimated from a partition of the dataset and the predictive performance is measured on the remaining part of the data set. Risk of over-fitting the training set by introducing additional model parameters is illustrated. The results of the DSS are presented to the users using risk maps – a user-interface in analogy with the earlier intuitive plots and independent of model choice.