AUTHOR=Cantor Melissa C. , Casella Enrico , Silvestri Simone , Renaud David L. , Costa Joao H. C. 

TITLE=Using Machine Learning and Behavioral Patterns Observed by Automated Feeders and Accelerometers for the Early Indication of Clinical Bovine Respiratory Disease Status in Preweaned Dairy Calves

JOURNAL=Frontiers in Animal Science

VOLUME=Volume 3 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/animal-science/articles/10.3389/fanim.2022.852359

DOI=10.3389/fanim.2022.852359

ISSN=2673-6225

ABSTRACT=The objective of this retrospective cohort study was to evaluate a K-Nearest Neighbor (KNN) algorithm to classify and indicate clinical Bovine Respiratory Disease (BRD) status using behavioral patterns in preweaned dairy calves. Calves (N=106) were enrolled on this study from one facility for the preweaning period. Precision dairy technologies were used to record feeding behavior with an automated feeder and activity behavior with a pedometer (automated features). Daily, calves were manually health scored for Bovine Respiratory Disease (BRD; Wisconsin scoring system, WI, USA) and weights were taken twice weekly (manual features). Calves were also scored for ultrasonographic lung consolidation twice weekly. A BRD bout (day 0) was classified as 2 scores classified as abnormal on the Wisconsin scoring system and an area of consolidated lung ≥ 3.0 cm2. There were 54 calves diagnosed with a BRD bout. Two scenarios were considered for KNN inference. In the first scenario (diagnosis scenario), the KNN algorithm classified calves as BRD positive or as negative for respiratory infection. For the second scenario (pre-BRD bout scenario) the 14 days before a BRD bout were evaluated to determine if behavioral changes were indicative of calves destined for disease. Both scenarios investigated the use of automated features, manual features, or both. For the diagnosis scenario, manual features had negligible improvements compared to automated features, with an accuracy of (0.95 ± 0.02) and (0.94 ± 0:02), respectively, for classifying calves as negative for respiratory infection. There was an equal accuracy of (0.98 ± 0:01) for classifying calves as sick using automated and manual features. For the pre-BRD bout scenario, automated features were highly accurate 6 days prior to diagnosis (0.90 ± 0:02), while manual features had low accuracy at 6 days (0.52 ± 0.03). Automated features were near perfectly accurate 1 day before BRD diagnosis compared to the high accuracy of manual features (0.86  ±  0.03). This research indicates that machine learning algorithms accurately predict BRD status at up to 6 days using a myriad of 28 feeding behaviors and activity levels in calves. Precision dairy technologies hold the potential to indicate BRD status in preweaned calves.