Risk factors explaining the animal welfare status of a dairy herd are currently being intensively investigated by our research team at McGill. Being able to predict actual dairy herds at low or high animal welfare status can make a major contribution to improve the animal welfare status on a large scale and change the status quo.
Daniel Warner’s research focuses on using new machine-learning techniques to predict whether a specific dairy herd may have animal welfare deficiencies. The routinely collected milking data from Québec’s dairy herds by Valacta come in handy as these data contain plenty of information that may explain the animal welfare status of a specific dairy herd. As it is typical for such big data, relationships and patterns in routine milking data are not easy to grasp.
Daniel has been, therefore, working on several machine-learning techniques, ranging from prediction models based on a simple decision tree to an entire (random) forest. Such models are often used as a decision support tool by big data companies around the globe. In our project, such models will help us to predict the actual animal welfare status of dairy herds based on routine milking data.
Preliminary results are promising. However, the false positive and false negative alerts produced by these models are a continuous cause of concern. While it is essential that our models detect a high percentage of dairy herds with animal welfare deficiencies, it is equally important for a dairy producer that its herd may not be erroneously classified as a low animal welfare herd when in reality it is not.
Daniel and his colleagues at Valacta and McGill intensified their research efforts to come up with sound prediction models. Based on our results, an intervention protocol will be developed to assist Québec’s dairy producers in improving the animal welfare status.
Commentaires