Egg production forecasting: Determining efficient modeling approaches.

Area of Science:

• Poultry Science
• Artificial Intelligence
• Statistical Modeling

Background:

• Accurate forecasting of egg production is crucial for commercial layer management.
• Traditional statistical models have limitations in capturing complex production dynamics.

Purpose of the Study:

• To compare the efficiency of various artificial intelligence and statistical models for forecasting egg production in commercial layers.
• To identify the most accurate model for predicting weekly egg yields from week 22 to 36.

Main Methods:

• Developed and compared mathematical, statistical, and artificial intelligence models using data from a commercial layer trial.
• Generated simulated data for neural network training and testing.
• Evaluated three neural network architectures: back-propagation-3, Ward-5, and general regression neural network (GRNN).
• Compared GRNN predictions against original data, strain averages, linear regression, and the Gompertz model.

Main Results:

• The general regression neural network (GRNN) achieved the best fit (R² = 0.71), closely matching commercial egg production data.
• GRNN outperformed traditional models, including linear regression and the Gompertz nonlinear model, in predicting egg production curves.
• Neural network models generally require less data and are practical for farm management.

Conclusions:

• The general regression neural network is a highly efficient and accurate model for forecasting commercial layer egg production.
• While GRNN shows superiority, managing initial overprediction is necessary for optimal application.
• Neural network models offer practical advantages for poultry farm management due to their efficiency and data requirements.