AHV has positive contribution on milk production predictability

Predicting the milk yield of cows is essential for running a dairy farm optimally. It is not only important to be able to optimise your farm’s income and animal health in this way, but also to anticipate the expected milk yield. This is important for both dairy farmer and milk supplier.

What was researched?

The main objective of the research is to develop a (real-time) tool that predicts milk yield for different dairy farms using a Machine Learning method.

This study used the Gradient Boosting method to predict milk yield. This study uses machine learning technique XGBoost to predict milk yield for each dairy farm. This model was applied and evaluated based on expected and delivered milk yield. Our results show that our model makes accurate predictions, with the worst-case scenario showing a deviation of 2.2% in the 2022 validation set.

Milk production predictability

We also used the ‘feature selection’ technique SHAP (SHapley Additive exPlanations) to provide insight into the XGBoost model. This method allows us to assign a value to certain features or products used on a farm. This provides insight into which features and/or products are critical for milk production and how management practices can be improved. Through this method, administering ‘AHV Quick Tablet’ and ‘AHV Extra Tablet’ among the herd has been shown to play a more important role in predicting milk production compared to a herd without AHV products.

Fig. Waterfall of feature importance.

Although the model was developed specifically for Dutch livestock farmers, we believe that the method used in this study can easily be applied to other livestock farmers outside The Netherlands. If you are curious about the entire article, click here!

Contributing to increased milk production

Besides the predictability of milk production, it is also interesting to know to what extent the products contribute to increased milk production. To gain a better insight into this, a second article will be written, more on this later.