Optimizing Feeding Schedule and Live-Weight Prediction for Native Chicken Based on Machine Learning.

Area of Science:

• Animal Science
• Agricultural Technology
• Machine Learning in Agriculture

Background:

• Accurate scheduling of processing dates is crucial for market supply, demand, and optimal pricing in poultry production.
• Local Taiwanese chicken breeds, including Guzao males, Huangjin females, and Red Junglefowl males, require tailored feeding programs.
• Current feeding practices may not fully leverage data for precise flock management.

Purpose of the Study:

• To develop a practical feeding program system for Taiwanese chicken breeds.
• To integrate daily body weight predictions to guide flocks toward target weights before processing.
• To evaluate machine learning models for predicting body weight and optimizing feed allocation.

Main Methods:

• Four prediction models (Random Forest, XGBoost, Extra Trees, Artificial Neural Network) were evaluated for body weight prediction.
• The best-performing model was embedded into a practical feeding system.
• An Extra Trees model was used for estimating remaining rations and updating daily feed allocations.
• A validation experiment was conducted on a batch of Guzao males over 54 days.

Main Results:

• The developed system integrates cage-level body weight predictions to guide flock growth.
• The Extra Trees model proved effective for feed estimation and allocation adjustments.
• A validation experiment showed preliminary support for the data-guided feeding program's feasibility.
• Comparison with conventional feeding indicated potential improvements in feed conversion ratio.

Conclusions:

• The data-guided feeding program system demonstrates potential agricultural application value for Taiwanese chicken breeds.
• Further validation with additional batches and cross-farm evaluations is necessary to confirm generalizability and operational performance.
• This approach offers a pathway to more precise and efficient poultry production management.