Soft-Sensing Approach for Predicting Bovine Respiratory Disease Severity

Bovine respiratory disease (BRD) is the most common cause of morbidity and mortalityin cattle around the world causing important health problems (e.g., mortality, growthproblems, and low productivity) in all cattle husbandry systems. BRD is a multifactorialsyndrome, with various predisposing factors (stressors) being necessary to inducedisease and affect disease severity. Predisposing factors include external stressors suchas inadequate vaccination protocols, comingling with other cattle, dusty environment,sudden and extreme weather changes, and dehydration. Additionally, internal factorssuch acute metabolic disturbances and inadequate transfer of passive immunity areinvolved as well. BRD is one of the most extensively studied diseases since the beginningof 19th century until today. However, it is still a challenge to predict whichspecific groups of cattle are at the highest risk of BRD progression (i.e., severity andmorbidity) and other BRD-related outcomes (e.g., mortality and disease severity). Inthe present paper, a conceptual framework for streaming adaptive machine-learningalgorithm for farm-based prediction of the BRD progression outcomes (mortality andseverity) is introduced. The introduced approach is based on combining farm (activity,animal-to-animal contact and feeding behaviour) and (sub)-clinical data (body temperature,nasal and ocular discharges and respiration rate) measurements togetherwith a machine-learning-based prediction model. A hybrid machine learning approachbased on localised least-squares support vector machines (LS-SVM) and k-nearestneighbours (kNN) is proposed. The developed adaptive model is suitable for onlineprediction of BRD-related mortality and severity for clinical decision support systems.