Operational and actionable Acute Food Insecurity modelling 

Authors:

Abstract
The growing complexity of global food security, exacerbated by climate change and socio-economic disparities, calls for a multi-hazard approach to risk evaluation and management. Recognizing the lack of a universally accepted measure for food insecurity covering all dimensions, we first review target variables and input features in existing ML modeling efforts, providing an assessment of current data availability, accessibility, and fragmentation, and improving the understanding of possibilities and limitations of ML for the food security community. We further consolidate a comprehensive dataset, with an operational design for continuous enrichment, that includes various indicators and precursors, updated monthly on a subnational level across over one hundred countries. We apply innovative explainable artificial intelligence (XAI) methods to unravel the intricate relationships between food insecurity, drought, and conflict-related fatalities. Our models forecast food crises with different lead times, revealing the nuanced patterns recognized by machine learning algorithms over various time frames. Our analysis also shows that the relative importance of drivers can shift depending on the food security metric used, indicating that distinct processes are at play in its many dimensions. This study not only exposes the complex drivers of food security but also provides policymakers with an operational multi-risk forecasting tool, improving the ability to foresee and strategically manage food crises.