Developing an algorithm to monitor fattening pig welfare at group level: the approach by the ClearFarm project

ClearFarm dedicates to constructing an online platform to monitor animal welfare throughout the value chain for dairy cattle and pigs. The platform is driven by machine learnt algorithms, which are fed with welfare information measured by sensors. Due to several technical and practical limitations, we propose one algorithm to monitor fattening pigs at group level. The algorithm is structured by Five Domains model of Mellor et al. (2020). Mental health domain is excluded from the platform because there is currently no sensor applicable to measure mental health. Score for each domain is presented in two forms: a numeric and a colour scores. Red (0-3) indicates a high probability of some welfare problem that needs an immediate intervention; yellow (3-7) indicates a non-negligible probability of some welfare problem that suggests being precautious; green (8-10) indicates low probability of some welfare problem that animals are in acceptable welfare status. Parameters measured by sensors are allocated to their relevant domain. For example, respiratory health index belongs to health domain; ammonia concentration belongs to environment domain. Thresholds for each parameter are determined by literature or legislation. The score for each domain is the aggregation among the parameters within. This algorithm continuously gives a score when sensor data is fed in, and flags early-warning signals, such that intervention can be made timely and precisely. This algorithm is not yet meaningful because of several knowledge gaps, one of which is limited validated sensor technology in pigs to monitor a sufficient number of welfare indicators.