Machine Learning-Based Geospatial Risk Modeling of Global Avian Influenza Outbreaks.

Area of Science:

• Veterinary Epidemiology
• Disease Ecology
• Computational Biology

Background:

• The H5N1 avian influenza virus presents a significant global health concern, necessitating advanced spatiotemporal risk assessment strategies.
• Understanding the complex interplay of environmental, ecological, and anthropogenic factors is crucial for predicting H5N1 outbreaks.

Purpose of the Study:

• To develop and validate a global modeling framework for characterizing H5N1 avian influenza outbreak risk.
• To identify key environmental, ecological, and anthropogenic drivers influencing H5N1 spatiotemporal distribution.
• To visualize seasonal H5N1 outbreak risk patterns globally.

Main Methods:

• Utilized machine learning (ML) algorithms including logistic regression, SVM, random forest, LGBM, and XGB.
• Employed geospatial analysis with confirmed H5N1 presence data from WAHIS (2012-2023) and target-group background for pseudo-absences.
• Validated models using spatial block cross-validation and an independent temporal holdout dataset; applied seasonal MaxEnt for risk mapping.

Main Results:

• Tree-based ensemble ML models (RF, LGBM, XGB) demonstrated superior and stable performance in both spatial and temporal validations.
• Seasonal risk maps indicated heightened H5N1 risk during autumn and winter, moderate risk during spring migration, and lower risk in summer.
• Livestock density and anthropogenic variables emerged as the most significant predictors of H5N1 outbreaks in multivariate analyses.

Conclusions:

• The developed ML framework provides a robust tool for global H5N1 avian influenza risk assessment.
• Seasonal patterns and key drivers like livestock density and anthropogenic factors are critical for targeted surveillance and control strategies.
• Integrating diverse data sources and advanced modeling techniques is essential for managing transboundary animal diseases.