Feature-driven optimization for growth and mortality prevention in poultry farms

Area of Science:

• Precision Agriculture
• Machine Learning in Animal Science
• Poultry Production Systems

Background:

• The poultry industry faces significant challenges in minimizing mortality rates and optimizing growth performance.
• Environmental and operational variables introduce variability impacting production outcomes.
• Effective prediction models are crucial for informed decision-making in poultry management.

Purpose of the Study:

• To develop and implement a feature-driven optimization model for predicting poultry mortality and average weight.
• To evaluate the efficacy of various machine learning models for this predictive task.
• To provide a practical, intelligent decision support tool for poultry managers.

Main Methods:

• Utilized an 88-day dataset from over 20,000 Taiwan native broilers.
• Preprocessed data included outlier removal, normalization, and interpolation.
• Developed and refined a Neural Network (NN) model, specifically an Ensemble NN, for multi-output prediction.

Main Results:

• The Ensemble NN achieved low Root Mean Square Errors (RMSE) of 0.45 for Mortality and 0.02 for AvgWeight.
• Feature importance analysis identified 'Day' as the primary predictor for mortality, followed by feed and water consumption.
• The system was integrated into a MATLAB-based application for practical use.

Conclusions:

• The developed Ensemble NN model offers a scalable and accurate tool for precision agriculture in poultry.
• The intelligent system provides valuable decision support for optimizing poultry production.
• Bridging deep learning with livestock management enhances efficiency and reduces losses.