Rumen Fermentation Parameters Prediction Model for Dairy Cows Using a Stacking Ensemble Learning Method.

Area of Science:

• Animal Science
• Machine Learning
• Rumen Microbiology

Background:

• Rumen fermentation produces volatile fatty acids (VFAs) and methane, crucial parameters for dairy cow health and productivity.
• Current in vitro and machine learning models for predicting these parameters often lack generalization due to small sample sizes.

Purpose of the Study:

• To develop a robust prediction model for rumen fermentation parameters (methane, acetic acid, propionic acid) in dairy cows.
• To utilize stacking ensemble learning for improved prediction accuracy with limited data.

Main Methods:

• Employed in vitro techniques to collect rumen fermentation data.
• Developed a stacking ensemble learning model using neutral detergent fiber (NDF), acid detergent fiber (ADF), crude protein (CP), and dry matter (DM) as inputs.
• Validated the model's robustness with an independent experiment using varying concentrate-to-forage (C:F) ratios.

Main Results:

• The stacking model demonstrated high prediction accuracy for methane (R2 = 0.928), acetic acid (R2 = 0.888), and propionic acid (R2 = 0.924).
• The model effectively simulated methane and VFA variations in response to dietary fiber content.
• Successfully predicted rumen fermentation type shifts and methane changes under different C:F ratios, confirming model robustness with small sample sizes.

Conclusions:

• The developed stacking ensemble model offers a reliable tool for predicting rumen fermentation parameters.
• This model can guide dairy cow diet optimization, methane emission reduction strategies, and enhance feed utilization and cow health.