Answers are grounded in this paper and cite their sources.
Bovine respiratory disease is highly prevalent, imposes significant economic costs on the livestock industry, and has long-term impacts on animal performance. Current detection methods are often labour-intensive and lack precision. In this study, data from a newly developed, calf-specific, management ear tag were combined in different ways with data from an automatic feeder, daily environmental measurements, and a pedometer to develop a bovine respiratory disease detection model for pre-weaned dairy-beef calves. A total of 118 calves were monitored during their pre-weaning period (9 ± 0.2 days on entry to the trial, 54 ± 0.3 days at weaning), with 100 calves included in the dataset. Using the random forest modelling technique, 40 different models were developed by incorporating different combinations of variables from the different technologies. All models were evaluated against an established clinical health scoring system for individual calves (daily Wisconsin score). To assess the efficacy of the models in detecting bovine respiratory disease, two metrics were used: accuracy of predictions (alerts generated from the model) and accuracy of disease detection (correctly identified disease events). Based on these metrics, the best performance was obtained from a model based on the automatic feeder data using averaged values across the previous day (model Bm1), which had the best balance between the accuracy of alerts (56.0%) and the accuracy of disease predictions (69.1%), providing some of the highest numbers of true positive and true negative results. Higher accuracy of alerts was obtained using all daily data from the different technologies (model FTIEB), yielding 67.5% accurate alerts. Meanwhile, higher accuracy of disease detection was obtained by using feeder-based average values across the two previous days (model Bm2), yielding 69.9% accurate disease detection. This analysis emphasised the power of integrating several technologies into predictive modelling for BRD in calves, as well as the importance of incorporating averaged values across previous days obtained from the automatic feeder into the prediction models. This study shows that using the Wisconsin score as a classification and integrating multiple data sources make early detection of BRD viable, achieving up to 70% accuracy, and should be investigated further.
No takes yet. Share an insight, caveat, or question.
Brans et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: