Adsa Logo White Adsa Title White

Estrus prediction of cows and heifers with an activity and rumination monitoring system in an organic grazing and a low-input conventional dairy herd.

B. J. Heins

Events

06-24-2020

Abstract:

305
Estrus prediction of cows and heifers with an activity and rumination monitoring system in an organic grazing and a low-input conventional dairy herd.
B. J. Heins*, K. Minegishi. University of Minnesota St. Paul, MN.

The objective of this study was to evaluate estrus detection with an activity and rumination system (ARS) in a seasonal calving organic grazing (ORG) and a low-input conventional (CONV) dairy herd. Data provided by the ARS was used with machine learning techniques to create estrus prediction models for heifers and cows. The study was conducted at the University of Minnesota West Central Research and Outreach Center, Morris, MN from January 2016 to August 2019. Cows calve seasonally on this research farm. Cows that calved in the spring were bred during the summer and cows that calved in the autumn were bred during the winter. The study had 4 winter breeding seasons (December to February) and 4 summer breeding seasons (June to August). During each breeding season, activity and rumination were monitored electronically using an ear-tag accelerometer sensor (CowManager SensOor, Agis Automatisering BV, Harmelen, the Netherlands). Estrus alerts of individual cows provided by the activity and rumination monitoring system were used to determine agreement with the breeding date of a cow. The study included 1,671 breeding dates from 917 cows and 180 breeding dates from 126 heifers (HEF). The estrus prediction analysis focused on 8 machine learning algorithms with R statistical version 3.51 (R Foundation, Vienna, Austria). Model prediction was assessed by receiver operation characteristic (ROC) curves. For the winter breeding season, the ROC curves had 80% sensitivity (SN) with 97% specificity (SP). The ROC curves for the summer breeding season were lower (72 to 77% SN with fixed 97% SP) for cows in CONV herd, for cows in ORG herd (50% SN), and for HEF (63 to 71% SN). With a lower sensitivity and a higher positive predictive value (PPV), summer prediction results were 76 to 81% SN and 54 to 70% PPV for CONV, 52 to 64% SN and 42 to 50% PPV for ORG, and 69 to 76% SN with 51 to 68% PPV for HEF. The custom models developed using the raw ARS data showed the potential range of sensitivity and specificity that can be achieved with these data.

Keywords: automated estrus detection, grazing, low-input dairy.