With the continuous development of animal husbandry, accurately predicting the energy content of feed is crucial for optimizing formulas and improving production efficiency. This study applied AutoGluon, an automated machine learning framework, to predict digestible energy and metabolizable energy in swine feed, and compared its performance with artificial neural networks and support vector machines. Using 1341 records from the China feed database and 45 records for pregnant sows, AutoGluon achieved superior accuracy, with R2 values of 0.940 (digestible energy) and 0.938 (metabolizable energy) on the general test data set. On the pregnant sow test set, it performed best for metabolizable energy prediction (R2 = 0.939), while artificial neural networks showed a slight advantage for digestible energy (R2 = 0.920). An AutoGluon-based predictor called pig energy predictor was developed for practical application, enabling rapid and precise feed energy estimation. This study demonstrates the potential of pig energy predictor for practical applications in swine nutrition and provides a tailored model for general pig populations, supporting precision feed formulation and data-driven decision-making in animal husbandry.
Yu et al. (Sun,) studied this question.