Predicting fibre digestibility in Holstein dairy cows fed dry-hay-based rations through machine learning

Area of Science:

• Animal Science
• Dairy Nutrition
• Ruminant Physiology

Background:

• Accurate rationing programs for dairy cattle rely on predicting feed digestibility.
• Developing reliable prediction equations typically requires extensive in vivo data.
• Existing methods for predicting fiber digestibility in lactating cows are limited, especially for hay-based rations.

Purpose of the Study:

• To develop a predictive equation for total-tract potentially digestible neutral detergent fiber digestibility (TTpdNDFD) in lactating cows fed hay-based rations.
• To utilize existing in vivo data to create a robust and accurate prediction model.
• To explore the application of such an equation in precision dairy farming.

Main Methods:

• Compiled individual data from 11 studies, encompassing 69 cows, 35 treatments, and 1,614 observations.
• Used diet characteristics (ash, starch, CP, NDF, ADF, ADL, undegradable NDF, potential degradable NDF) as predictors.
• Employed bidirectional stepwise selection in JMP software, with outlier removal, multicollinearity checks, and 10-fold cross-validation for model training.

Main Results:

• Achieved high model performance with R² values of 0.91 (training) and 0.90 (validation), and low Root Mean Square Error (RMSE) of 2.99 (training) and 3.26 (validation).
• Confirmed that TTpdNDFD decreases with increasing dietary starch and fast-fermentable fiber content.
• Demonstrated that the developed equation accurately predicts fiber digestibility using only diet characteristics.

Conclusions:

• An accurate equation for predicting TTpdNDFD in lactating cows fed hay-based rations can be developed from historical data.
• This prediction tool, when integrated with precision farming, can optimize feed planning, monitoring, and cost-efficiency for dairy herds.
• Further research with expanded datasets is recommended to enhance the scalability and applicability of these predictive models.