Abstract
This study aimed to determine the available energy values of sorghum for Arbor Acres (AA) broilers and to develop and compare prediction equations using multivariate linear stepwise regression (MLSR) and machine learning-based linear regression (LR). It is important to clarify that the LR used in this study is fundamentally a linear regression model. The primary difference from MLSR lies in the variable selection process and optimization algorithm. Ten sorghum samples with varied chemical compositions were used. Test diets were formulated by replacing 40% of the energy-yielding components of a reference diet with individual sorghum. A total of 7929-day-old and 39623-day-old male AA broilers were used in two balance trials. Apparent metabolizable energy (AME), nitrogen-corrected AME (AMEn), and net energy (NE) were determined using the substitution method and respiratory calorimetry. Correlation analyses between energy values and chemical components were performed. Prediction equations for AME and NE were developed using both MLSR (in SPSS) and LR (in Python scikit-learn). The AME, AMEn, and NE values for sorghum were significantly higher in 26-28-day-old broilers compared to 12-14-day-old broilers (P < 0.05). Tannin (TN) content showed significant negative correlations with AME and AMEn. Accurate prediction equations for AME and NE were established for both age stages using key predictors such as acid detergent fiber (ADF), TN, phytic acid (PA), dry matter (DM), and crude protein (CP). Equations developed with both MLSR and LR methods demonstrated a good fit to the development data. Validation results indicated that the predictive accuracy and performance of the two methods were comparable. This study provides definitive available energy values of sorghum for broilers at different ages. The developed prediction equations, whether based on MLSR or LR, are effective tools for estimating the energy content of sorghum. The findings support the use of sorghum as a viable alternative energy source in broiler diets and provide a scientific basis for its precise application in feed formulation.