Carcass condemnation prediction using artificial neural networks in poultry slaughterhouses

Abstract

Challenges related to carcass quality and high condemnation rates result in significant economic losses. Also, the rigorous inspection of broiler carcasses is essential to ensure sanitary standards and food safety in industrial poultry production. To enhance traditional inspection methods, this study evaluated the use of artificial neural networks (ANN) to predict conditions and condemnations in broiler chickens. Data from 3370 flocks inspected at a poultry company in southern Brazil between 2019 and 2022 were analyzed and 16 output variables were modeled using NeuroShell Predictor software. Nine models (56.25 %) were classified as good or excellent, with the models for “partial and total condemnation” presenting coefficients of determination (R²) > 0.93 and correlation coefficients close to 0.96, reflecting strong predictive performance and consistency between the predicted and actual values. Conditions involving more complex visual diagnoses such as cachexia, ascites, and cellulitis resulted in less accurate models. These findings suggest that ANNs can be useful tools to support postmortem inspection, reducing economic losses and improving the efficiency and sustainability of the poultry production chain. Nonetheless, because the data were sourced from a single production system, model applicability is limited to this context. Further studies using external datasets from other companies are recommended to assess model generalization under different production conditions.