Field Implementation of Forecasting Models for Predicting Nursery Mortality in a Midwestern US Swine Production...

Area of Science:

• Animal Science
• Agricultural Economics
• Veterinary Epidemiology

Background:

• Nursery mortality significantly impacts swine production economics.
• Accurate prediction of mortality is crucial for effective management strategies.

Purpose of the Study:

• To evaluate the performance of five forecasting models for predicting pig nursery mortality.
• To identify the most accurate model using pre-weaning and stocking variables.

Main Methods:

• Utilized a master dataset of 3242 pig groups (~13 million animals) with 42 pre-weaning and stocking variables.
• Employed cross-validation to train and test models, including Support Vector Machine (SVM).
• Validated the best performing model on a separate dataset of 72 new groups.

Main Results:

• The Support Vector Machine (SVM) model demonstrated superior performance with RMSE = 0.406, MAE = 0.284, and R² = 0.731.
• On a new dataset, the SVM model achieved 77.78% accuracy in predicting high or low nursery mortality groups.
• Despite a decrease in R² to 0.554 on the new data, the model maintained predictive capability.

Conclusions:

• Forecasting models, particularly SVM, can effectively predict pig nursery mortality.
• Pre-weaning and stocking condition variables are valuable predictors of nursery mortality.
• This approach supports proactive management decisions in swine farming.