Introduction:

The project will include work with pre-partum dairy cattle. The goal of the research is to use machine learning techniques to predict calving events. Behavioral monitors were attached to each dairy cow, and collected data through the birthing process.

Predicting impending calving using automatic activity and rumination measures in dairy cattle.

Commercially available behavioral monitors have long been available to dairy farmers to monitor their cattle. The potential exists for current behavioral monitors to be used in the prediction of calving events. Dairy cattle frequently experience difficulty at time of calving, and predicting these events would allow dairy farmers to alleviate some of these effects. The objective of this study was to monitor behavioral changes in prepartum dairy cattle and predict impending calvings through the automated activity and rumination observation. Data collection for 29 primiparous and 46 multiparous Holstein dairy cattle occurred from September 2011 through May 2013 at the University of Kentucky Coldstream Dairy. The HR Tag (SCR Engineers, Ltd., Israel) was used to automatically collect neck activity and rumination data. The IceQube (IceRobotics, Ltd., Scotland) collected step number, hours lying, hours standing, lying bouts, and total motion data. Data collection occurred for 7 weeks prepartum and retrospective data analysis was performed using SAS (Cary, NC). All data was summed into differing time blocks for use in calving prediction. Data summed into 2 h, 6 h, 12 h, and 24 h time blocks for each cow will be used to predict calving events.
All behavioral data will be assessed and used to predict calving events using machine-learning methods. These methods have been constructed and will be implemented using R version 3.1.0 (R Foundation for Statistical Computing, Vienna, Austria). Random forest, linear discriminant analysis, and neural network machine-learning techniques will be used to predict calving events. Machine-learning techniques will need to be applied to 21 d of prepartum behavioral data before calving events (n = 46). Establishing behavioral changes through least-squares means and creating alerts relative to calving using this same data may be useful in predicting impending calvings without need for additional technologies or parameters.

Software:

The methods to be used will be machine learning. Machine learning techniques utilizing a random forest analysis, linear discriminant analysis, and a neural network analysis will be applied to various data sets. Because of the complexity and computational power needed to perform these analyses, we are requesting you assistance.

These packages will be run in R, which the university currently offers.

Students

Publications

Grants