Fuzzy-augmented machine learning models for the prediction of feed efficiency and disease status in livestock: A mink case study

• A fuzzy-augmented machine-learning framework is proposed for livestock trait prediction. • Growth, feed efficiency, and health status are predicted from farm-acquirable measurements. • Fuzzy random forest achieved the most consistent and balanced model performance. • Fuzzification enhanced model performance for selected traits and learning algorithms. • The framework supports data-driven livestock management and smart decision making. This study applies a fuzzy-augmented machine learning framework, in which fuzzy logic is integrated into the modeling pipeline, to predict feed efficiency traits and Aleutian disease status in American mink ( Neogale vison ) as a model species. Feed efficiency and growth were quantified through feed conversion ratio (FCR), residual feed intake (RFI), and average daily gain (ADG), while Aleutian disease status was evaluated using counterimmunoelectrophoresis response. Relatively non-invasive and farm-acquirable phenotypic traits, including sex, color, body weight, body length, and birth year, were collected and used as input data for six machine learning models: XGBoost, gradient boosting, AdaBoost.R2, random forest, k-nearest neighbors, and a deep neural network. Each model was evaluated in both non-fuzzy and fuzzy-enhanced configurations. Fuzzification was performed using Gaussian membership functions to generate overlapping representations and a fuzzy inference scalar, enriching the models’ input features. Across most trait-algorithm combinations, fuzzification improved predictive accuracy by refining nonlinear feature interactions and reducing decision boundary ambiguity. The fuzzy random forest achieved the most balanced performance, with R² values of 0.79 for FCR, 0.82 for RFI, and 0.95 for ADG, corresponding to RMSE values of 5.47, 14.65 g day⁻¹, and 0.83 g day⁻¹, respectively, along with reliable health classification accuracy of approximately 0.79 and an area under the curve of approximately 0.74. The results demonstrate that the proposed framework enhances predictive precision and robustness in data-driven livestock systems and provides a practical tool for livestock management and the digitalization in animal agriculture.