Optimizing wild bird fecal surveillance sites for HPAI through risk estimation using machine learning algorithms

Wild birds are major reservoirs contributing to the spread of highly pathogenic avian influenza (HPAI). Accordingly, fecal-based surveillance of wild birds has been implemented as an early warning system for poultry in South Korea. However, the efficiency of the current system remains limited due to the absence of a quantitative basis. This study aimed to develop a risk-based strategy for optimizing fecal sampling sites, thereby strengthening early warning capacity and preventing the introduction of HPAI into poultry farms.

The study area comprised 110 designated surveillance areas divided into 1-km edge hexagonal grids. The study period covered October to February from 2021 to 2024. For each grid, we integrated wild bird fecal sampling data with various categories of factors. The dependent variable was the detection of avian influenza virus (AIV) in wild bird feces, while all other variables served as predictors. The risk of AIV detection was estimated using machine learning algorithms, including logistic regression, gradient boosting machine, extreme gradient boosting, and an ensemble model. Model performance was evaluated using the area under the curve (AUC), sensitivity, specificity, and accuracy.

The ensemble model achieved the best predictive performance. Climatic factors, sample size, population, and elevation were identified as the most influential predictors. Spatial inconsistency was observed between the current sampling sites and the proposed risk-based sampling sites.

The proposed risk-based framework enables prioritization of sampling sites, thereby optimizing fecal-based surveillance for HPAI. This approach is expected to enhance the efficiency of EWS and contribute to the prevention of HPAI spread to poultry.