A correlated-variables model for monitoring individual growing-finishing pig's behavior by RFID registrations

de Bruijn, B.G.C.; de Mol, R.M.; Hogewerf, P.H.; van der Fels, J.B.


Feeding and drinking behavior of individual growing-finishing pigs can be used for health and welfare monitoring by real time warning systems. The objective of this study was to develop and test a model detecting deviating feeding and drinking behavior of individual growing-finishing pigs based on Radio-Frequency Identification (RFID) registrations. Feeding and drinking behavior of growing-finishing pigs was recorded through low frequency RFID readings via ear tags. In three 16-week batches, twelve pens containing each twelve pigs were equipped with one drinker and two feeders. Four readers (each with eight antennas) recorded feeding and drinking activity of each individual pig in the pens. All tag readings were combined into visits, and subsequently visits into meals. The analyzed variables were number of meals per day, average interval between meals and maximum interval between meals per day for both feeding and drinking. A correlated-variables model with a Kalman filter was developed, generating alerts when the daily level was deviating from the expected level. For illustration of model results, the model was validated with culling recordings, which included all pigs that died or were euthanized during the experiment. Most cases of culled pigs corresponded with alerts given by the model, but sensitivity was hard to determine due to low number of cases and specific circumstances of the culled pigs. The specificity of the model was similar for feeding and drinking behavior, and was highest for number of meals (93–99%), followed by the average interval between meals (90–96%) and the maximum interval between meals (86–97%), depending on desired confidence interval. The developed model is promising for early detection of health problems in growing-finishing pigs, but further validation should occur on a bigger scale in order to increase accuracy of results and improve knowledge required for practical implication of the model.