Changes in ingestive-related behaviors (e.g., feeding and ruminating) are key indicators for assessing health and well-being in cattle. The aim of this study was to develop a decision-tree (DT) algorithm to classify neck-mounted accelerometer data for the measurement of feeding and ruminating behaviors in dairy cattle. The performances of the DT were compared with those of a support vector machine (SVM) algorithm. Ten multiparous dairy cows were used in this study. The cows were housed in an area of 36x13 m2 with individual cubicles and concrete slatted floor. The cows were fed roughage ad libitum. Drinking water was available ad libitum. Collar-mounted accelerometers were used to distinguish between three behavioral categories: feeding, ruminating and other activity (non-ingestive). The accelerometer (sampling at 10 Hz) was attached on the cow’s collar. Direct observations (i.e., reference) were made from 09:00 AM to 03:00 PM (6 hours per cow). To classify the data of the three behaviors, a new DT algorithm was developed. The decision-tree algorithm was selected for its low computational costs, which makes it implementable on the on-cow nodes. The performances were calculated using leave-one-out cross-validation. Results showed that the DT algorithm nearly matched the performances of computationally intensive algorithms such as SVM (i.e., overall accuracy of 89 % for the DT and 92 % for SVM). The precision, sensitivity, and specificity measures were between 77 % and 94 % for the DT, and between 82 % and 96 % for the SVM. These preliminary findings illustrate the potential of the collar-mounted accelerometer to classify feeding and ruminating behaviors with a simple DT method. This would optimize the power consumption of the sensors by transmitting just the behavior of the cow instead of all the raw data to the backend system. Measurements are being continued in order to validate the reported results.
|Titel||EurAgEng conference, 8 – 12 July 2018, Wageningen, the Netherlands|
|Status||Gepubliceerd - jul-2018|