Prediction of Pork Belly Fat Percentage Based on Backfat Thickness Measurements in Pork Carcasses

This study developed and internally evaluated a two-input candidate model for estimating VCS2000-derived belly fat percentage (BFP) in slaughterhouses without automated carcass evaluation equipment. Data from 1,753 LYD (Landrace × Yorkshire × Duroc) pigs slaughtered in Korea were analyzed. Hot carcass (HC) traits were obtained using VCS2000, whereas corresponding chilled carcass (CC) backfat traits were measured manually. Multiple linear regression models were assessed using influence diagnostics, sensitivity analyses, cross-validation, and comparisons with machine learning algorithms. In HC measurements, BFP was strongly associated with multifidus muscle middle backfat thickness (MBT) and backfat thickness at the first thoracic vertebra (FTBT). Under the CC/manual condition, the MBT association remained relatively strong, whereas the FTBT association was weaker. The HC full and reduced models showed no substantial difference in explanatory or cross-validated predictive performance, whereas the CC/manual reduced model showed lower performance. The evaluated machine learning methods did not meaningfully outperform multiple linear regression. Measurement concordance was lower for FTBT than for MBT and grading backfat thickness, although the effects of carcass state and measurement procedure could not be separated. These findings provide a basis for further development of the two-input model as a candidate approach for BFP estimation.