Explainable artificial intelligence with data augmentation for available energy prediction in corn and soybean meal towards precision swine feeding

This is the authors' abstract. We don't add key points for this paper.

Accurate evaluation of available energy in corn and soybean meal (SBM) is essential for precision swine nutrition. Traditional linear models have limited capacity to capture the complex non-linear relationships among nutritional components. This study developed high-performing machine learning models to predict the available energy of corn and SBM for swine and to generate interpretable, model-derived insights into key nutritional determinants. Global datasets were integrated and robustly expanded using advanced data augmentation models. After data augmentation, decision tree, random forest, gradient boosting, and extreme gradient boosting (XGBoost) were developed. Predictive mechanisms were interpreted using permutation feature importance, feature interaction, partial dependence plots (PDP), Shapley Additive Explanations (SHAP) summary plots, and local interpretable model-agnostic explanations (LIME). Results indicated that models trained on augmented datasets showed improved performance, with XGBoost achieving R 2 values of 0.848 and 0.892 for corn and SBM net energy on the test sets. Starch and neutral detergent fiber (NDF) were identified as the two most important predictors contributing to corn available energy prediction. For SBM, lysine (Lys) and crude fiber (CF) emerged as the leading predictors. Furthermore, several model-derived response patterns were identified, which require validation with independent experimental datasets. Overall, the integration of data augmentation and explainable artificial intelligence (XAI) provides a promising computational framework for rapid ingredient evaluation, precision diet formulation, and future intelligent feeding decision-support systems.