Development of Machine Learning Models for Estimating Metabolizable Protein Supply from Feed in Lactating Dairy Cows.

Abstract

Accurate prediction of protein utilization in dairy cows is essential for optimizing nutrition and milk yield to achieve sustainable cattle production. This study aimed to develop novel machine learning models to predict rumen-undegradable protein (RUP) and duodenal microbial nitrogen (MicN) based on dietary protein intake. A dataset comprising 1779 observations from 436 scientific publications was used to train support vector regression (SVR) and random forest regression (RFR) models. Different predictor sets were identified for each model, including factors such as days in milk (DIM), dry matter intake (DMI), dietary fiber content, and crude protein fractions. Model performance was evaluated using statistical metrics, including the coefficient of determination (R²), root mean square error of prediction (RMSEP), and concordance correlation coefficient (CCC), with results compared to existing NASEM (2021) models. The RFR model provided the most precise and unbiased predictions for RUP (R² = 0.60, RMSEP = 0.326 kg/d, CCC = 0.71), while the SVR model was most effective for MicN (R² = 0.76, RMSEP = 42.4 g/d, CCC = 0.86). Both models outperformed traditional methods, demonstrating the potential of machine learning in improving protein utilization predictions. Future studies could explore hybrid approaches integrating conventional and AI-based models to enhance predictive accuracy.