An Investigation of Surface Temperature Effect on Estrus Detection of Dairy Cows using Supervised Learning

This paper proposes an investigation of the surface temperature effect on the estrus detection of dairy cows using the supervised learning. The neck collars with temperature and motion sensors were attached to the tested dairy cows. Four IP cameras were installed in the cow shed to monitor dairy cow behaviors and for data labeling. Neck temperature and motion data were collected and classified for behaviors and estrus prediction using 3 different supervised learning techniques: artificial neural network (ANN), Decision Tree (DT), and Random Forest (RF). By incorporating the neck temperature, the validation accuracies significantly improved over 25% compared to those of the control set, which comprised only motion data. Other performance metrics also improved in the same manner. The estrus prediction accuracy was 100% for all 3 techniques, around 11% higher than that of the control set. It was later discovered that the temperature feature alone could provide relatively the same validation accuracy, indicating the neck temperature as an important feature for estrus prediction.