Food insecurity — a condition of limited or uncertain access to adequate food — is a major problem worldwide: around 29.3% of the global population was moderately or severely food insecure in 2021. To better assess and tackle the situation, face-to-face and remote surveys are often performed in various communities. However, these surveys require high costs in both monetary and human resources, which hampers a more frequent data collection that would be ideal for making more timely and informed decisions on policies that are oriented towards eliminating malnourishment and hunger. But what if we could accurately estimate and predict food insecurity from other, easier-to-obtain datasets? Elisa Omodei and colleagues recently investigated this question, proposing models that can nowcast the food security situation in near real time without the need for expensive survey data.
Given that food security has multiple drivers, the authors took into account and collected an extensive amount of data from existing sources — spanning 15 years and more than 40 countries — that cover different dimensions, such as conflict-related fatalities, economic information (for instance, food inflation and gross domestic product (GDP) per capita) and weather-related features (for instance, rainfall and vegetation). Gradient boosted regression trees were then used as models to predict two existing indicators that characterize household food insecurity levels, one capturing quantity and diversity of dietary intake and the other capturing the consequences of constrained access to food. With these models, the authors were able to not only predict these indicators with coefficients of determination (R2) > 0.6 but also obtain adequate near-real-time estimates that can complement other near-real-time efforts currently available. The authors also extended the SHAP (SHapley Additive exPlanations) framework to better understand which factors most influence changes in predicted trends (for instance, a decay in food consumption over time) per country, which is key to support a more informed decision-making. The proposed models are promising tools in the fight against food insecurity, and a prime example of how combining the right set of computational tools with the right data holds promise to solve the world’s biggest challenges.
This is a preview of subscription content, access via your institution