Behavioral indicators of calf adaptation to automated milk feeders

Training calves to drink milk independently from an automated milk feeder (AMF) requires substantial labor, with calves taking several days to learn; however, it is unknown if feeding behavior can predict successfully trained calves. The objectives of this retrospective cohort study were to (1) evaluate the association between AMF feeding behaviors (milk intake, drinking speed, and rewarded visits) and their relative changes with training success, and (2) determine which feeding behaviors were useful for classifying calves as successfully trained to use AMF by d 4 in Angus × Holstein calves fed near ad libitum milk. Angus × Holstein calves (n = 461; 129 training success, 332 unsuccessful calves) were raised by one commercial calf raiser that offered near ad libitum milk replacer with an AMF (Foerster-Technik, Engen, Germany). Calves were classified as successfully trained when they independently visited the AMF >2 times/d and consumed >6 L/d of milk within 4 d. Day 4 was selected as the training success threshold (d 0) because the average training day was 3.5 ± 3.6 d (mean ± SD). Mixed linear regression models were used for the association of feeding behavior, and relative changes in each behavior with training success for d -3 to d 7 adjusting for the fixed effects of day, training success status, pen, the training status × day interaction, repeating by day, with calf nested within source farm as a random effect. Logistic regression models were used to identify the optimal feeding behaviors required to classify calf training success using area under the curve (AUC), Youden's index, sensitivity, specificity, accuracy, and precision as performance indicators and the random effect of source farm. We observed a training success status × day interaction for milk intake and relative changes in milk intake. Successfully trained calves had greater milk intake from d -2 to 7 and greater relative changes in milk intake from d 2 to 7 compared with unsuccessful calves. Moreover, there was a training success status × day interaction for rewarded visits and relative changes in rewarded visits. Compared with unsuccessful calves, successfully trained calves had greater rewarded visits from d -1 to d 7 and greater relative changes in rewarded visits on d -3. We found no association of training success status with drinking speed or relative changes in drinking speed. The best-performing prediction model used milk intake, rewarded visits, relative changes in milk intake, and relative changes in rewarded visits (AUC = 0.86, Youden's index = 0.63, sensitivity = 0.87, specificity = 0.76, accuracy = 0.79, and precision = 0.59). These findings suggest that farmers can monitor average milk intakes, rewarded visits, and relative changes in rewarded visits to identify which calves drink independently from an AMF.