Classifying the posture and activity of ewes and lambs using accelerometers and machine learning on a commercial flock

Early decision making in commercial livestock systems is key to maximising animal welfare and production. Detailed information on an animal’s phenotype needed facilitate this, but can be difficult obtain a setting. Research into the use of bio-logging sheep continuously monitor individual behaviour indirectly inform health production has seen rapid growth recent years. Much this research, however, been conducted small numbers animals experimental setting over limited time periods. Previous studies have also focused ewes there little research potential for collecting behavioural data lambs, despite clear relevance The present study aimed test feasibility deploying accelerometers flock at point annual cycle (lambing), validate viability automated monitoring Also, we develop robust machine learning algorithms that classify both posture physical activity adult lambs. We used Random Forest algorithm predict: two mutually exclusive postures lambs (standing lying), achieving average accuracies 83.7% 85.9% respectively; four activities (grazing, ruminating, inactive walking), accuracy 70.9%; three (inactive, suckling, 80.8%. These performance large individuals afford opportunity provide detailed understanding daily budget Monitoring changes patterns across while capturing environmental conditions such as weather, day length, terrain management could reveal indicator metrics may early warning issues flocks.