Multi-algorithm Machine Learning Uncovers Complex Interactions between Ecological and Host Factors that Drive Patterns of Avian Influenza Spatial Risk Across Eurasia and North Africa

Avian influenza viruses (AIVs) are highly adaptive pathogens capable of rapid mutation and cross-species transmission, posing a persistent threat to global health. AIVs’ uneven geographic distribution suggests that outbreak patterns are shaped by complex, non-linear relationships with ecological and host-related drivers, which may differ by viral pathogenicity and host type. Most machine learning statistical approaches primarily focus on predicting the spatial risk of AIVs and identifying their most important predictors, while overlooking the interrelationships between these predictors in shaping the underlying risk of outbreaks. In this study, we employed a robust multi-algorithm machine learning (ML) ensemble to analyze over 50,000 reported AIV outbreaks from 2004 to 2024 across Eurasia and Northern Africa. Using 34 environmental and host-related predictors, we modeled the risk of outbreaks for all AIVs collectively, as well as for high-pathogenicity (HPAI), low-pathogenicity (LPAI), and domestic versus wild bird subsets. Our ML models achieved high predictive accuracy (≥84%) across all categories. The ecological niche of HPAI closely mirrored that of overall AIV risk, while distinct spatial patterns emerged for LPAI and host-specific models. We identified notable high-risk areas with suitable ecological conditions for the circulation of AIVs in underreporting countries located in the Middle East and Northern Africa. Proximity to wetlands and vegetation indices proved stronger predictors than climatic variables or poultry density. However, non-linear interactions, particularly between poultry density, land cover, and climatic variability, were key in shaping the underlying spatial risk of almost all outbreaks. Notably, we identified high-risk areas in underreported regions such as Iran and Algeria, highlighting critical gaps in global surveillance. Our study demonstrates the necessity of interrogating ecological and host interactions to disentangle the complex drivers of AIV emergence and spread, providing an interpretable and scalable tool for targeted surveillance and risk-based policy planning for both human and animal health.