Predicting Food-Security Crises in the Horn of Africa Using Machine Learning

Food insecurity is a global concern resulting from various complex processes and a diverse range of drivers. Due to its complexity, it is one of the most challenging drought impacts to predict. In this study, we introduce a novel machine learning model designed to forecast food crises in the Horn of Africa up to 12 months in advance. We trained an “XGBoost” model using more than 20 different input datasets to capture key food security drivers such as drought, economic shocks, conflicts and livelihood vulnerability. The model shows a promising ability to predict food security dynamics several months in advance (R²>0.6, three months in advance). Notably, it accurately predicted 20% of crisis onsets in pastoral regions (n = 84) and 40% of crisis onsets in agro-pastoral regions (n = 23) with a 3-month lead time. We compared these results to the established FEWS NET early warning system, and found a similar performance over these regions. However, our model is clearly less skilled in predicting food security for crop-farming regions than FEWS NET. This study underscores the importance of integrating machine learning into operational early-warning systems like FEWS NET and expanding these techniques to the continental or global-scale.