Development of Machine Learning-Based Sub-Models for Predicting Net Protein Requirements in Lactating Dairy Cows

Area of Science:

• Animal Nutrition and Metabolism
• Machine Learning in Agriculture
• Dairy Science

Background:

• Accurate estimation of lactating dairy cow protein requirements is crucial for diet formulation, feed efficiency, and reduced nitrogen excretion.
• Existing methods for calculating protein needs can be complex and data-intensive.

Purpose of the Study:

• To develop and evaluate machine learning models, specifically Random Forest Regression (RFR) and Support Vector Regression (SVR), for predicting net protein requirements for maintenance (NPm) and lactation (NPl) in dairy cows.
• To assess the predictive performance of these models using farm-ready input variables.

Main Methods:

• Compiled a dataset of 1779 observations from 436 publications and databases.
• Utilized predictor variables including milk yield, dry matter intake, days in milk, body weight, and dietary crude protein.
• Estimated NPm using National Academies of Sciences, Engineering, and Medicine (NASEM, 2021) equations and NPl from milk true protein yield.
• Employed 10-fold cross-validation to evaluate model adequacy.

Main Results:

• The RFR model achieved superior predictive performance for both NPm (R² = 0.82, RMSEP = 22.38 g/d, CCC = 0.89) and NPl (R² = 0.82, RMSEP = 95.17 g/d, CCC = 0.89) compared to SVR.
• RFR's effectiveness highlights its ability to capture the rule-based nature of the NASEM equations.
• The models demonstrated the potential for accurate protein requirement estimation using a reduced set of input variables.

Conclusions:

• Random Forest Regression offers a robust and potentially simpler machine learning approach for estimating dairy cow protein requirements.
• These models can aid in formulating more precise diets, enhancing feed utilization, and minimizing environmental nitrogen impact.
• Future research should validate these models in field settings and explore hybrid mechanistic-machine learning frameworks.