Comparison of Regression Models for Predicting Major Primal-Cut Weights across Carcass-Weight Categories in Landrace × Yorkshire × Duroc Pigs

Objective: This study aimed to identify the best-performing regression algorithm for predicting the weights of major primal cuts within each pig carcass weight category, in order to improve the accuracy of meat yield estimation.Methods: Data from 2,676,916 Landrace × Yorkshire × Duroc (LYD) pigs slaughtered between 2022 and 2024 were utilized.Carcass weights were categorized into five intervals: ≤70, 71-80, 81-90, 91-100, and >100 kg.Five types of regression models (simple linear regression, multiple linear regression, Elastic Net, partial least squares regression, and random forest) were applied to seven primal cuts (shoulder blade, shoulder picnic, loin, belly, ribs, tenderloin, and ham).Model robustness was further assessed via 5-fold cross-validation, nested hyperparameter tuning, external temporal validation, VIFbased multicollinearity screening, and bootstrap-based uncertainty quantification of R² and RMSE.Results: The SLR models showed low explanatory power with coefficients of determination (R 2 ) ranging from 0.04 to 0.66.In contrast, MLR and Elastic Net demonstrated high predictive performance for most primal cuts and weight intervals, with R 2 values between 0.56 and 0.99.RF exhibited particularly high explanatory power for the belly.Carcass weight was confirmed as the most critical variable across all cuts, showing a strong positive correlation.Lean meat percentage generally had a positive influence on cut weights, whereas fat percentage had a negative influence, except in the belly, where fat-related variables were highly significant.Backfat thickness, sex, and season had limited impact on prediction performance. Conclusion:The results of this study confirmed the differences in primal cut growth patterns by carcass weight categories and demonstrated that advanced regression models can accurately predict yields.This is expected to provide fundamental data for precise yield estimation and the establishment of data driven production strategies, helping to enhance the efficiency of slaughter and meat processing management.