Animal-Based Indicators for On-Farm Welfare Assessment in Goats

Investigating the use of machine learning algorithms to support risk-based animal welfare inspections of cattle and pig farms

In livestock production, animal-related data are often registered in specialised databases and are usually not interconnected, except for a common identifier. Analysis of combined datasets and the possible inclusion of third-party information can provide a more complete picture or reveal complex relationships. The aim of this study was to develop a risk index to predict farms with an increased likelihood for animal welfare violations, defined as non-compliance during on-farm welfare inspections. A data-driven approach was chosen for this purpose, focusing on the combination of existing Swiss government databases and registers. Individual animal-level data were aggregated at the herd level. Since data collection and availability were best for cattle and pigs, the focus was on these two livestock species. We present machine learning models that can be used as a tool to plan and optimise risk-based on-farm welfare inspections by proposing a consolidated list of priority holdings to be visited. The results of previous on-farm welfare inspections were used to calibrate a binary welfare index, which is the prediction goal. The risk index is based on proxy information, such as the participation in animal welfare programmes with structured housing and outdoor access, herd type and size, or animal movement data. Since transparency of the model is critical both for public acceptance of such a data-driven index and farm control planning, the Random Forest model, for which the decision process can be illustrated, was investigated in depth. Using historical inspection data with an overall low prevalence of violations of approximately 4% for both species, the developed index was able to predict violations with a sensitivity of 81.2 and 79.5% for cattle and pig farms, respectively. The study has shown that combining multiple and heterogeneous data sources improves the quality of the models. Furthermore, privacy-preserving methods are applied to a research environment to explore the available data before restricting the feature space to the most relevant. This study demonstrates that data-driven monitoring of livestock populations is already possible with the existing datasets and the models developed can be a useful tool to plan and conduct risk-based animal welfare inspection.

Development of a data-driven method for assessing health and welfare in the most common livestock species in Switzerland: The Smart Animal Health project

Improving animal health and welfare in livestock systems depends on reliable proxies for assessment and monitoring. The aim of this project was to develop a novel method that relies on animal-based indicators and data-driven metrics for assessing health and welfare at farm level for the most common livestock species in Switzerland. Method development followed a uniform multi-stage process for each species. Scientific literature was systematically reviewed to identify potential health and welfare indicators for cattle, sheep, goats, pigs and poultry. Suitable indicators were applied in the field and compared with outcomes of the Welfare Quality® scores of a given farm. To identify farms at risk for violations of animal welfare regulations, several agricultural and animal health databases were interconnected and various supervised machine-learning techniques were applied to model the status of farms. Literature reviews identified a variety of indicators, some of which are well established, while others lack reliability or practicability, or still need further validation. Data quality and availability strongly varied among animal species, with most data available for dairy cows and pigs. Data-based indicators were almost exclusively limited to the categories “Animal health” and “Husbandry and feeding”. The assessment of “Appropriate behavior” and “Freedom from pain, suffering, harm and anxiety” depended largely on indicators that had to be assessed and monitored on-farm. The different machine-learning techniques used to identify farms for risk-based animal welfare inspections reached similar classification performances with sensitivities above 80%. Features with the highest predictive weights were: Participation in federal ecological and animal welfare programs, farm demographics and farmers' notification discipline for animal movements. A common method with individual sets of indicators for each species was developed. The results show that, depending on data availability for the individual animal categories, models based on proxy data can achieve high correlations with animal health and welfare assessed on-farm. Nevertheless, for sufficient validity, a combination of data-based indicators and on-farm assessments is currently required. For a broad implementation of the methods, alternatives to extensive manual on-farm assessments are needed, whereby smart farming technologies have great potential to support the assessment if the specific monitoring goals are defined.

Lesions found at foot trimming of dairy goats: Baseline data for comparing lesions and locomotion scoring

Data on the foot lesions of housed dairy goats, especially those that are not lame, is sparse. In this study, visits were made to four farms for whole herd locomotion scoring after milking and to observe routine foot trimming. During trimming, the amount of trimming required for each hoof was recorded, alongside the presence of lesions. The process was repeated at the next routine foot trim on three of the four farms (on Farm 1 whole herd locomotion scoring after milking was not feasible). Across both visits, 3445 locomotion scores were recorded, alongside 3850 trimming and lesion records. There were large differences between farms and visits in the proportion of lame goats (6.7 to 25.5%) and in the proportion of goats which needed a substantial trim of all four feet (15.1-33.1%). Hoof lesions were observed in 65.5% of trimmed goats. We defined lesions by their site and presence of separation and/or granuloma, with white line separation being the most commonly recorded lesion (41.5% of goats). On all visits on the three farms which were locomotion scored, the number of goats recorded as lame was lower than the number observed with white line lesions, but the relative proportion ranged from 17% to 95.4%. Separations were much more common than granulomas, as granulomas were usually accompanied by separation. Toe lesions were much more likely to be associated with granulomas than white line lesions, with 91.5% of toe separations having granulomas but only 10.6% of white line separations. This study provides useful baseline data for comparing lesions and locomotion scoring between goat herds but also identifies areas where we need more data and more consensus.