Machine learning predictive modeling for condemnation risk assessment in antibiotic-free raised broilers.

Area of Science:

• Poultry Science
• Machine Learning
• Agricultural Economics

Background:

• Broiler carcass condemnation causes significant financial losses and food waste in the poultry industry.
• Assessing condemnation risk in antibiotic-free broilers is crucial for economic viability.

Purpose of the Study:

• To evaluate and compare machine learning (ML) models for predicting high condemnation rates in antibiotic-free broilers.
• To identify key variables influencing broiler condemnation.

Main Methods:

• Utilized least absolute shrinkage and selection operator (LASSO), classification tree (CT), and random forests (RF) models.
• Applied sampling techniques (ROS, RUS, BOTH, ROSE) to address imbalanced datasets.
• Analyzed 23,959 truckloads with 14 independent variables from rearing to slaughter.

Main Results:

• Random Forests with random under sampling (RF with RUS) demonstrated superior performance in predicting high condemnation rates.
• Mean body weight, weight per crate, mortality/culling rates, and lairage time were identified as the most influential predictors.
• High condemnation rates (threshold ≥ 3.0%) were present in 8.05% of the dataset.

Conclusions:

• ML, particularly RF with RUS, offers a robust framework for predicting broiler condemnation risk.
• Identifying critical variables enables targeted farm management strategies to minimize economic losses.
• The adaptable ML approach can be applied across different broiler production systems.