Predicting Body Weight in Japanese Quail Using Early-Life Growth, Feed Intake, and Egg Traits: A SHAP-LightGBM Modeling Framework

Accurate prediction of growth traits is crucial for enhancing genetics and improving the accuracy of selection in poultry breeding programs. Traditional phenotypic selection methods often struggle to model the complex, non-linear interactions that affect body weight development. In this study, we used an interpretable machine learning approach, Shapley Additive Explanations (SHAP)-enhanced Light Gradient Boosting Machine (LightGBM), to forecast body weight at 10 weeks (Bw_10wk) in Japanese quail. The dataset included 150 birds with recorded early-life body weights (Bw_2wk to Bw_8wk), feed intake measures (FI_2wk to FI_10wk), and egg quality traits (egg weight, length, and width). Model training employed an 80:20 train-test split, with hyperparameter tuning conducted for both LightGBM and XGBoost models. LightGBM outperformed XGBoost, achieving a test R 2 of 0.981, an RMSE of 7.24, and an MAE of 5.86, demonstrating a strong generalization ability. SHAP analysis showed that feed intake at 8 weeks (FI_8wk) was the top predictor of final body weight, followed by FI_2wk, intermediate body weights (Bw_6wk, Bw_8wk), and egg traits, especially egg width and egg weight. Including egg traits and early feed intake highlights a broader physiological context, suggesting that reproductive and behavioral indicators can serve as reasonable proxies for growth potential. These findings emphasize the importance of biologically interpretable machine learning models in identifying practical predictors for breeding programs. The SHAP-LightGBM framework thus serves as a valuable tool for combining multiple phenotypic traits to support precise and welfare-focused breeding strategies in Japanese quail, with possible uses in other poultry species.