Activity and synchrony patterns obtained by a tracking-by-detection algorithm as potential predictors of tail biting at pen and individual level in pigs

Tail biting is a major welfare concern in pigs. Timely interventions may help to reduce its spread in an outbreak. However, tail biting is usually detected only at an injury stage. Therefore, early indicators are needed to predict outbreaks. While previous studies have suggested that pen-level activity increases before a tail-biting outbreak, early identification of individual biters and victims could allow more targeted interventions, such as removing the biter or the victim. However, the link between individual activity and tail biting remains unclear, and identifying biters and victims based on activity patterns requires long-term individual monitoring, feasible only through automated methods. This study aimed to examine whether a change in activity and synchrony patterns obtained by a tracking-by-detection algorithm is predictive of tail damage at pen level, and to investigate whether activity measurements are related to performing and receiving tail biting at the individual level. Nine pens with tail damage (i.e., damage pens) were selected across three batches of growing pigs (n = 30 pens in total), in which at least one pig had a tail wound on two consecutive observation days (where D₀ refers to the first observation day). Each damage pen was matched with a control pen showing no pigs with tail wounds on the same days. Tail biting, including the identities of the biter and victim, was scored from videos of damage pens on D_-5, D_-2, and D₊₁. A tracking algorithm was applied on several days around the detected tail damage (D_-9, D_-5, D_-2, and D₊₁) to calculate the distance moved by each individual. The average daily distance moved in the pen did not affect the probability of a pen to become a damage pen (P > 0.05). However, in damage pens, positive relationships between the frequency of tail biting and individual activity on some days preceding the damage were found. For example, higher individual distances moved on D_-9 were associated with increased tail-biting frequencies on D_-5 (P < 0.001), and those on D_-5 with increased tail-biting frequencies on D_-2 (P < 0.05). No relationships between activity synchrony on the days preceding tail damage and the probability of a pen becoming a damage pen or the frequency of biting were found. This study suggests that individual activity, measured with tracking data, may serve as a potential predictor for identifying biters. However, the heterogeneity of tail biting may limit the development of a simple and generalised method for predicting tail biting based on automatic activity monitoring.