AI-Driven Famine Forecasting: Predicting Food Crises with Machine Learning

Abstract

Famines currently affect over 282 million people across 59 countries, with approximately 9 million individuals dying annually from hunger and hunger-related diseases—surpassing deaths from AIDS, malaria, and tuberculosis combined. This humanitarian crisis is exacerbated by inadequate forecasting of affected populations, leading to insufficient funding and resource distribution by organizations such as the World Health Organization (WHO). Underestimating the scale of food insecurity perpetuates hunger, leaving vulnerable populations without critical aid.​ This research presents an AI-driven solution for forecasting famine severity using the Integrated Food Security Phase Classification (IPC) index—a globally recognized standard for assessing food insecurity. Our approach leverages autoencoders to reduce data dimensionality, enabling better feature generalization, followed by Generative Adversarial Networks (GANs) to effectively scale the data. Finally, a deep learning model predicts the distribution of IPC phases with high accuracy. This workflow enhances data representation and improves predictive performance, achieving a 91% accuracy rate in forecasting IPC distributions.​ The AI system not only generalized key factors contributing to famine but also enables precise forecasting of food insecurity levels across countries. By providing more accurate predictions, this model equips organizations such as the WHO and NATO with actionable insights for proactive resource allocation and strategic funding distribution, ultimately mitigating the impact of famines on vulnerable populations.