Phenotype-driven machine learning models for predicting average daily gain in Yorkshire pigs with SHAP interpretation

A regression-based ML framework accurately predicts average daily gain (ADG)in Yorkshire pigs. 2. CatBoost outperformed 14 models and showed strong external validation. 3. SHAP analysis identified key phenotypic predictors affecting ADG variation. 4. A web-based tool enables real-time and interpretable ADG prediction on farms. Average daily gain (ADG) is a key indicator of growth performance in swine production. Although genomic prediction tools such as genomic best linear unbiased prediction (GBLUP) and single-step genomic best linear unbiased prediction (ssGBLUP) are widely used in breeding programs, their application may be limited by cost and data availability. To provide a practical and cost-effective complement to genomic evaluation, we developed a machine learning–based phenotypic prediction framework for estimating ADG in Yorkshire pigs using routinely recorded early-life variables. Production records from 12,079 pigs raised under standardized conditions between February 2020 and April 2024 were curated, and after data cleaning, fifteen regression algorithms were trained and evaluated using the root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and the coefficient of determination ( R 2 ). Model interpretability was assessed using SHapley Additive exPlanations (SHAP), and an independent external cohort was used for validation. Results indicated that CatBoost delivered the highest predictive accuracy and demonstrated strong generalization in both internal and external validations. SHAP analysis identified biologically meaningful early-life predictors contributing to ADG variation. To promote practical adoption, we developed a user-friendly web application that enables real-time prediction and interpretation of ADG outcomes. Overall, this study demonstrates that routinely collected phenotypic and management data can effectively support accurate ADG prediction through machine learning, offering a data-driven tool to enhance decision-making and production efficiency in swine systems.