Improving the estimation of the lean meat percentage in pig carcasses using Box-Cox transformation and additional
T Rombouts1, M Seynaeve1, S De Smet1
1Laboratory for Animal Nutrition and Animal Product Quality, Department of Animal Sciences and Aquatic Ecology, Ghent University, Coupure Links 653, Block F 1(st) floor, 9000 Ghent, Belgium.
Abstract:
The objective of this study was to evaluate and improve the prediction models of the AutoFOM III and FOM II apparatuses (Frontmatec Group, Denmark) used to estimate the lean meat percentage (LMP) of pig carcasses in Belgium, since the current models underestimate the pig carcasses with a LMP higher than 66 %. Non-linearity in the backfat thickness (BT) parameters was identified as the main reason for this bias in prediction. Box-Cox transformation of the parameters R2P10, R2P8 and R2P4 from AutoFOM III allowed to lower the root mean squared error of prediction (RMSEP) of the model from 1.72 to 1.59, while simultaneously removing the bias of the high LMP carcasses. For the FOM II apparatus, there was no effect of the transformation of the only BT parameter on the RMSEP (2.15 before and 2.14 after transformation) and on the bias. Next to the transformation, it was investigated whether adding other information about the carcasses could also improve the RMSEP of the prediction models. The parameters hot carcass weight, carcass length, ham width, ham angle and sex were added to the original models without transformation and lowered the RMSEP from AutoFOM III and FOM II to 1.55 and 1.83 respectively. Finally, the best results were found by combining the Box-Cox transformation and adding other carcass parameters, resulting in RMSEP values of 1.50 and 1.82 for AutoFOM III and FOM II respectively, on top of the removal of the high LMP bias. IMPLICATIONS: Accurate estimation of the lean meat percentage of pig carcasses is of great economic importance for the pig production and slaughtering sector, so every opportunity to increase precision should be seized. This study shows that the current linear prediction models can be improved by taking into account non-linearity, depending on the device. An even larger increase in precision can be achieved by adding carcass information that is currently not measured or not linked to the classification device but that is partly already available at the slaughterline.
