A Comprehensible Machine Learning Method for Precise Milk Yield Forecasting in Dairy Production Systems

The profitability of dairy farms and sustainable agricultural management depend heavily on milk output. However, reliable forecasting is difficult since a variety of factors, including breed traits, feed composition, nutritional supplements, and milk quality measures, affect daily milk yield. Using structured farm-level data, this study suggests an interpretable machine learning approach for forecasting milk yield in dairy farming systems. Features including concentrate intake, consumption of green and dry fodder, vitamin supplementation, azolla quantity, usage of calcium and mineral mixtures, milk fat percentage, SNF percentage, and breed information are all included in the dataset. MAE, RMSE, and R² score were used to build and assess multiple regression-based methods, such as Linear Regression, Polynomial Regression, Ridge, Lasso, Support Vector Regression (SVR), K-Nearest Neighbours (KNN), Random Forest, Gradient Boosting, and XGBoost. With an R2 score of 0.92, which indicates great predictive capability and model reliability, the experimental findings show that Linear Regression, Polynomial Regression, and Ridge Regression performed the best. The results indicate that interpretable regression models are useful for predicting milk yield and can help farmers make data-driven decisions in precision dairy farming, optimise feed planning, and increase productivity.