Machine-learning-based calving prediction from activity, lying, and ruminating behaviors in dairy cattle.

The objective of this study was to use automated activity, lying, and rumination monitors to characterize prepartum behavior and predict calving in dairy cattle. Data were collected from 20 primiparous and 33 multiparous Holstein dairy cattle from September 2011 to May 2013 at the University of Kentucky Coldstream Dairy. The HR Tag (SCR Engineers Ltd., Netanya, Israel) automatically collected neck activity and rumination data in 2-h increments. The IceQube (IceRobotics Ltd., South Queensferry, United Kingdom) automatically collected number of steps, lying time, standing time, number of transitions from standing to lying (lying bouts), and total motion, summed in 15-min increments. IceQube data were summed in 2-h increments to match HR Tag data. All behavioral data were collected for 14 d before the predicted calving date. Retrospective data analysis was performed using mixed linear models to examine behavioral changes by day in the 14 d before calving. Bihourly behavioral differences from baseline values over the 14 d before calving were also evaluated using mixed linear models. Changes in daily rumination time, total motion, lying time, and lying bouts occurred in the 14 d before calving. In the bihourly analysis, extreme values for all behaviors occurred in the final 24 h, indicating that the monitored behaviors may be useful in calving prediction. To determine whether technologies were useful at predicting calving, random forest, linear discriminant analysis, and neural network machine-learning techniques were constructed and implemented using R version 3.1.0 (R Foundation for Statistical Computing, Vienna, Austria). These methods were used on variables from each technology and all combined variables from both technologies. A neural network analysis that combined variables from both technologies at the daily level yielded 100.0% sensitivity and 86.8% specificity. A neural network analysis that combined variables from both technologies in bihourly increments was used to identify 2-h periods in the 8 h before calving with 82.8% sensitivity and 80.4% specificity. Changes in behavior and machine-learning alerts indicate that commercially marketed behavioral monitors may have calving prediction potential.

A validation of technologies monitoring dairy cow feeding, ruminating, and lying behaviors.

The objective of this study was to evaluate commercially available precision dairy technologies against direct visual observations of feeding, rumination, and lying behaviors. Primiparous (n=24) and multiparous (n=24) lactating Holstein dairy cattle (mean ± standard deviation; 223.4±117.8 d in milk, producing 29.2±8.2kg of milk/d) were fitted with 6 different triaxial accelerometer technologies evaluating cow behaviors at or before freshening. The AfiAct Pedometer Plus (Afimilk, Kibbutz Afikim, Israel) was used to monitor lying time. The CowManager SensOor (Agis, Harmelen, Netherlands) monitored rumination and feeding time. The HOBO Data Logger (HOBO Pendant G Acceleration Data Logger, Onset Computer Corp., Pocasset, MA) monitored lying time. The CowAlert IceQube (IceRobotics Ltd., Edinburgh, Scotland) monitored lying time. The Smartbow (Smartbow GmbH, Jutogasse, Austria) monitored rumination time. The Track A Cow (ENGS, Rosh Pina, Israel) monitored lying time and time spent around feeding areas for the calculation of feeding time. Over 8 d, 6 cows per day were visually observed for feeding, rumination, and lying behaviors for 2 h after morning and evening milking. The time of day was recorded when each behavior began and ended. These times were used to generate the length of time behaviors were visually observed. Pearson correlations (r; calculated using the CORR procedure of SAS Version 9.3, SAS Institute Inc., Cary, NC), and concordance correlations (CCC; calculated using the epiR package of R version 3.1.0, R Foundation for Statistical Computing, Vienna, Austria) evaluated association between visual observations and technology-recorded behaviors. Visually recorded feeding behaviors were moderately correlated with the CowManager SensOor (r=0.88, CCC=0.82) and Track A Cow (r=0.93, CCC=0.79) monitors. Visually recorded rumination behaviors were strongly correlated with the Smartbow (r=0.97, CCC=0.96), and weakly correlated with the CowManager SensOor (r=0.69, CCC=0.59). Visually recorded lying behaviors were strongly correlated with the AfiAct Pedometer Plus (r >0.99, CCC >0.99), CowAlert IceQube (r >0.99, CCC >0.99), and Track A Cow (r >0.99, CCC >0.99). The HOBO Data Loggers were moderately correlated (r >0.83, CCC >0.81) with visual observations. Based on these results, the evaluated precision dairy monitoring technologies accurately monitored dairy cattle behavior.

Behavioral and physiological changes around estrus events identified using multiple automated monitoring technologies.

This study included 2 objectives. The first objective was to describe estrus-related changes in parameters automatically recorded by the CowManager SensOor (Agis Automatisering, Harmelen, the Netherlands), DVM bolus (DVM Systems LLC, Greeley, CO), HR Tag (SCR Engineers Ltd., Netanya, Israel), IceQube (IceRobotics Ltd., Edinburgh, UK), and Track a Cow (Animart Inc., Beaver Dam, WI). This objective was accomplished using 35 cows in 3 groups between January and June 2013 at the University of Kentucky Coldstream Dairy. We used a modified Ovsynch with G7G protocol to partially synchronize ovulation, ending after the last PGF2α injection (d 0) to allow estrus expression. Visual observation for standing estrus was conducted for four 30-min periods at 0330, 1000, 1430, and 2200h on d 2, 3, 4, and 5. Eighteen of the 35 cows stood to be mounted at least once during the observation period. These cows were used to compare differences between the 6h before and after the first standing event (estrus) and the 2wk preceding that period (nonestrus) for all technology parameters. Differences between estrus and nonestrus were observed for CowManager SensOor minutes feeding per hour, minutes of high ear activity per hour, and minutes ruminating per hour; twice daily DVM bolus reticulorumen temperature; HR Tag neck activity per 2h and minutes ruminating per 2h; IceQube lying bouts per hour, minutes lying per hour, and number of steps per hour; and Track a Cow leg activity per hour and minutes lying per hour. No difference between estrus and nonestrus was observed for CowManager SensOor ear surface temperature per hour. The second objective of this study was to explore the estrus detection potential of machine-learning techniques using automatically collected data. Three machine-learning techniques (random forest, linear discriminant analysis, and neural network) were applied to automatically collected parameter data from the 18 cows observed in standing estrus. Machine learning accuracy for all technologies ranged from 91.0 to 100.0%. When we compared visual observation with progesterone profiles of all 32 cows, we found 65.6% accuracy. Based on these results, machine-learning techniques have potential to be applied to automatically collected technology data for estrus detection.

Stall cleanliness and stall temperature of two different freestall bases.

The objective of this study was to describe the differences in freestall cleanliness and stall temperature between a barn with Dual Chamber Cow Waterbeds (DCCW; Advanced Comfort Technology, Reedsburg, WI) and a barn with rubber-filled mattresses at the University of Kentucky Coldstream Dairy Research Farm from January 18, 2012, to May 3, 2013. Stall cleanliness was measured twice weekly (n=134) by the same 2 observers using a 0.91 m×0.91 m wire grid containing 128 equally sized rectangles (10.16 cm×5.08 cm). This grid was centered at the rear portion of the stall; a rectangle that was visibly wet or had any amount of feces present was defined as a dirty rectangle. Weekly stall temperature (n=66) was measured by the same observer during a.m. milkings in the same predetermined stalls. Feces and wet sawdust were removed from the stalls before stall temperatures were acquired. Temperatures were obtained using a handheld thermometer at 30.48 cm above the stall base as determined via dual laser measurements. Stall temperature was measured on the front, middle, and back of the stall first with clean sawdust and then with the sawdust removed from the stall and wiped clean with a towel. Daily temperature-humidity index (THI) was calculated using Kentucky climate data calculated through the University of Kentucky College of Agriculture via a data logger, located 5.63 km from the Coldstream Dairy Farm. Stall cleanliness was not different between the DCCW barn (26.09±0.89 rectangles) and the rubber-filled mattress barn (23.70±0.89 rectangles). Mean THI throughout the study was 64.39±0.82. Stall temperature was different among THI categories. Temperature-humidity index categories 1 (coldest), 2, 3, and 4 (warmest) had THI ranges of 22.94 to 50.77, 50.77 to 64.88, 64.88 to 78.75, and 78.75 to 101.59, respectively. Stall temperatures (°C; least squares means±SE) were 2.26±0.30, 8.86±0.30, 15.52±0.30, and 20.95±0.30 for THI categories 1 to 4, respectively. Stalls with rubber-filled mattresses had a lower temperature (°C) than DCCW with least squares means±SE of 10.52±0.21°C and 13.29±0.21°C, respectively. The DCCW were probably significantly warmer because water holds heat well. The DCCW may have more of a heat-insulating effect compared with rubber-filled mattresses.