Predicting the Dietary Metabolizable Energy and Crude Protein Levels of 1- to 70-Day-Old Hy-Line Brown Male Chicks Using Response Surface Methodology and Artificial Neural Network Models

To determine dietary metabolizable energy (ME) and crude protein (CP) levels of Hy-Line Brown male chicks and compare response surface methodology (RSM) with artificial neural network (ANN) modeling, a total of 2160 birds were assigned to a 4 × 3 factorial design with three growth stages. Both RSM and ANN were fitted to predict body weight gain (BWG) and feed-to-gain ratio (F/G), with performance evaluated via LOOCV, leave-2-treatments-out validation, and bootstrap resampling. The 22–56-day stage yielded the best model performance overall. For F/G, both models achieved strong and comparable fit (in-sample R2 ≈ 0.86–0.88; LOOCV retention ≈ 97%), and both remained predictive under leave-2-treatments-out validation (RSM R2 = 0.72, ANN R2 = 0.45). For BWG, model performance was weaker, and ANN’s apparent in-sample advantage over RSM (R2 = 0.585 vs. 0.388 at 22–56 d) did not hold under leave-2-treatments-out validation (ANN R2 = −12.7 vs. RSM R2 = −0.24). Lastly, multi-objective optimization combining both models predicted stage-specific nutrient combinations. The results of this study provide empirical support and a methodological reference for establishing stage-specific nutritional standards for Hy-Line Brown male chicks.