Accurate and interpretable prediction of poor health in small ruminants with accelerometers and machine learning

摘要: Accurate assessment of the health status individual animals is a key step in timely and targeted treatment infections, which critical fight against anthelmintic antimicrobial resistance. The FAMACHA scoring system has been used successfully to detect levels anaemia caused by infection with parasitic nematode Haemonchus contortus small ruminants an effective way identify individuals need treatment. However, assessing labour-intensive costly as must be manually examined at frequent intervals over season. Here, we show that accelerometers can measure activity extensively grazing subject natural worm southern Africa long time-scales, when combined machine learning, predict smallest pre-clinical increases score well those respond treatment, all high precision (>95%). We demonstrate these classifiers remain robust time, remarkably, generalise without retraining across goats sheep different regions types farming enterprise. Interpretation trained reveal effect haemonchosis increases, both exhibit similar reduction fine-grained variation their levels. Our study thus reveals common behavioural patterns ruminant species, low-cost biologgers exploit subtle changes animal enable intervention. This real potential improve economic outcomes welfare limit use drugs hence diminish pressures on resistance under conditions commercial resource-poor communal farming. Significance Statement Increasing availability make learning viable solutions current challenges global livestock pipeline accurately predicts earliest signs disease ruminants. With exemplar, illustrate predictive model generalises time even species retraining. prediction driven poor health. findings suggest monitored remotely, reducing labour costs, improving welfare, allowing for selective contrasting conditions. will decrease loss, maximise outcomes, reduce drug