Comparative Analysis of Machine Learning Models to Predict Backfat Thickness in Hanwoo Cattle

Abstract Accurate prediction of carcass traits is essential for genetic improvement and value optimization in the beef industry. Backfatthickness is particularly important because it directly affects both market value and consumer preference. This study evaluatedthe predictive performance of machine learning (ML) models for estimating backfat thickness in Hanwoo cattle. A total of 386 Hanwoo carcass records were used, and 10 carcass features served as input variables, with backfat thickness being the prediction target. Model performance was assessed primarily by mean absolute error (MAE) and secondarily by the coefficient of determination (R2). Support vector regression (SVR) showed the best predictive performance, achieving the lowest MAE(2.796) and the highest R2 (0.375) after hyperparameter tuning. In contrast, other models showed higher MAE values, generally ranging from 3.0 to 4.0. Pearson correlation analysis identified intramuscular fat score (r = 0.37), quality grade (r = −0.33), andsex (r = 0.34) as the most influential predictors of backfat thickness. These findings support SVR as a robust approach for predicting carcass backfat thickness in Hanwoo cattle and provide a practical framework for ML-based precision phenotyping and breeding strategies.