Predicting the Landscape Epidemiology of Foot-and-Mouth Disease in Endemic Regions: An Interpretable Machine Learning Approach

Foot-and-mouth disease (FMD) remains a devastating threat to livestock health and food security in the Middle East and North Africa (MENA), where complex interactions among host, environmental, and anthropogenic factors constitute an optimal endemic landscape for virus circulation. Here, we applied an interpretable machine learning (ML) statistical framework to model the epidemiological landscape of FMD between 2005 and 2025. Furthermore, we compared the ecological niche of serotypes O and A in the MENA region. Our ML algorithms demonstrated high predictive performance (accuracies > 85%) in identifying the geographical extent of high-risk areas, including under-reported regions such as the Southern and Northeastern Arabian Peninsula. Sheep density emerged as the dominant predictor for all FMD outbreaks and serotype O, with significant non-linear relationships with wind, temperature, and human population density. In contrast, serotype A risk was primarily influenced by buffalo density and proximity to roads and cropland. Our in-depth interaction and Shapley value analyses provided fine-scale interpretability by interrogating the threshold effects of each feature in shaping the spatial risk of FMD. Further implementation of our analytical pipeline to guide risk-based surveillance programs and intervention efforts will help reduce the economic and public health impacts of this devastating animal pathogen.