Improving the power of activity-based heat detection using additional automatically captured data

摘要

The purpose of this study was to determine the current performance of activity monitoring devices on New Zealand pasture-based farms and investigate the potential for improved detection using additional automatically captured data. The data consistedof heat events that were assigned from pregnancy diagnosis and mating records, and from milk progesterone levels collected twice weekly during mating period. Daily milk production, milking order, milk flow rate and milk conductivity records were also collected during the mating period. The best single predictor of oestrus was 24 hour milk yield difference (P <0.01) for Herd 1 and normalised milking order (P <0.01) for Herd 2. The normalised pedometer data was the next best predictor of oestrus in both herds (P <0.01). A linear logistic regression model was fitted within each herd. The best model included normalised milk production, milking order, milk flow rate and pedometer variables. Machine learning models with balanced bagging were also fitted to the data. The machine learning models provided a better fit than traditional statistical models. Pedometer data can aid in the detection of cow oestrus, however the power of detection improves significantly with the addition of milk yield, milk flow and milking order data.