Parametric and Hybrid Growth Modeling for Final-Week Broiler Live Weight Prediction

Abstract

Accurate forecasting of broiler live weight near market age is essential for effective production planning and decision-making. This study proposes a hybrid modeling framework for predicting daily mean broiler live weight during the final production week (days 36–42) using only early growth data (days 1–35). Three parametric growth models Gompertz, Richards, and Von Bertalanffy were first fitted to daily mean live weight observations to provide baseline forecasts. A residual-based hybrid extension was then developed by training a gradient boosting regressor on lagged residuals, day index, and daily bird counts to refine late-stage predictions. Among the parametric models, the Gompertz formulation achieved the highest accuracy (RMSE = 33.807 g, MAE = 22.552 g, R²=0.975), outperforming the Richards and Von Bertalanffy models. Hybridization yielded modest but consistent improvements for all models, with the greatest benefit observed for the Gompertz-based hybrid. Hyperparameter sensitivity analysis demonstrated that hybrid model performance was robust to tuning choices, indicating low risk of overfitting. Prediction uncertainty was quantified using a moving block bootstrap approach, which revealed narrow confidence intervals for the Gompertz-based models and wider intervals for the alternative formulations. Overall, the results indicate that biologically grounded growth models remain highly effective for late-stage broiler weight forecasting, and that residual-based hybrid modeling serves as a refinement mechanism rather than a substitute for mechanistic representations. The proposed framework offers an interpretable, robust, and uncertainty-aware approach for practical broiler live weight prediction near market age.