Behaviour Identification in Farmed Red Deer ( Cervus elaphus ) Using Wearable Activity Collars

Wearable sensors such as accelerometers have shown promise in identifying a range of behaviours in ruminant livestock for welfare and production purposes. The research and production of wearables have primarily focused on delivering to the dairy and beef cattle industries. As technologies have evolved, opportunities have arisen in other livestock sectors for the validation of wearable sensors to improve farm productivity and animal welfare. This research focuses on the validation of wearable activity collars to define behaviours in farmed red deer for potential use in welfare and production research. Due to the lack of off‐the‐shelf products for small ruminants, 24 wearable activity collars were custom‐built and fitted to 25 thirteen‐month‐old red deer hinds. Instantaneous sampling was used to observe the deer at 30 s intervals during 56 scheduled 2‐hour sessions by four observers over 5 days. Approximately eleven thousand behavioural records were used with 0.5 Hz collar accelerometer data to train and validate a classification algorithm. To ensure interpretability, the classifier used decision trees applied to a single feature, along with a post‐processing step to exploit temporal autocorrelation. The method achieved test set accuracies of 80.2 (73.2, 87.3)% for Grazing/Ruminating/Other; 93.7 (89.7, 97.7)% for Standing/Lying; and 90.7 (88.5, 92.8)% for Head Up/Head Down. These accuracies demonstrate that these activity collars can reliably distinguish between a range of behaviours in deer, making them a useful tool for deployment in behaviour and welfare research in red deer.