Related Experiment Video
Updated: Sep 18, 2025
Assessing Species-specific Contributions To Craniofacial Development Using Quail-duck Chimeras
Published on: May 31, 2014
Predicting and explaining high dead-on-arrival outcomes in meat-type ducks using deep learning: A path to improved
Chalita Jainonthee1, Phutsadee Sanwisate2, Panneepa Sivapirunthep3
1Veterinary Academic Office, Faculty of Veterinary Medicine, Chiang Mai University, Chiang Mai 50100, Thailand; Veterinary Public Health and Food Safety Centre for Asia Pacific (VPHCAP), Faculty of Veterinary Medicine, Chiang Mai University, Chiang Mai 50100, Thailand; Research Center for Veterinary Biosciences and Veterinary Public Health, Faculty of Veterinary Medicine, Chiang Mai University, Chiang Mai 50100, Thailand.
High dead-on-arrival (DOA) rates in meat ducks are predicted by an explainable deep learning model. Key factors include flock size, lairage temperature, duck age, and transport time, enabling targeted welfare improvements.
Area of Science:
- Animal Science
- Agricultural Engineering
- Data Science
Background:
- Dead-on-arrival (DOA) rates are a significant welfare and economic issue in poultry production.
- Meat-type ducks, unlike broilers and layers, require more research regarding preslaughter stressor impacts.
- Existing machine learning models for DOA prediction often lack transparency, limiting practical use.
Purpose of the Study:
- To develop an explainable deep learning model for predicting high DOA outcomes in meat-type ducks.
- To identify key preslaughter management and environmental factors influencing DOA rates.
- To enhance the practical application of predictive models through interpretability.
Main Methods:
- Analysis of 8220 truckload entries of meat-type ducks from 2022-2023.
- Development of a deep learning model for DOA classification.
- Application of SHapley Additive exPlanations (SHAP) for model interpretability.
Main Results:
- The model achieved 80.29% accuracy, 79.25% precision, 80.29% recall, and 76.03% AUC-ROC.
- Significant predictors of high DOA included duck head count, lairage temperature, duck age, and transport duration.
- Higher density, lower lairage temperatures, younger ducks, and shorter transport times were associated with increased DOA risk.
Conclusions:
- Explainable deep learning models can accurately predict high DOA rates in meat ducks.
- SHAP analysis provides crucial insights into factors driving DOA, supporting targeted interventions.
- Findings enable optimization of handling, transport, and lairage to improve duck welfare and reduce mortality.
