Predicting chronic wasting disease in white-tailed deer at the county scale using machine learning.

Area of Science:

• Wildlife ecology
• Disease ecology
• Conservation biology

Background:

• Chronic wasting disease (CWD) poses a significant threat to wild cervid populations, impacting conservation efforts and wildlife management.
• Current CWD risk factor investigations are primarily at the state level, limiting regional surveillance efficiency.
• Predictive modeling can enhance surveillance strategies for emerging CWD infections.

Purpose of the Study:

• To develop and evaluate a regional machine learning model for predicting CWD incidence in white-tailed deer.
• To identify counties at higher risk for new CWD infections to guide surveillance efforts.
• To assess the utility of a predictive web application for CWD surveillance planning.

Main Methods:

• Utilized CWD surveillance data from white-tailed deer in 16 eastern and midwestern US states.
• Employed four machine learning models, including Light Boosting Gradient, with five-fold cross-validation and grid search.
• Compared model predictions against subsequent year's surveillance data to validate performance.

Main Results:

• The Light Boosting Gradient model demonstrated the highest reliability for predicting CWD incidence at a regional scale.
• The predictive model can assist in identifying areas for targeted CWD surveillance.
• Discrepancies between model predictions and actual surveillance data highlight the need for a multi-faceted approach.

Conclusions:

• A regional predictive model, like the Light Boosting Gradient model, can significantly improve CWD surveillance efficiency.
• The CWD Prediction Web App offers a valuable tool for identifying potential CWD hotspots.
• Effective surveillance planning requires integrating predictive modeling with on-the-ground data and expert judgment from wildlife agencies.