On the forecastability of food insecurity

Area of Science:

• Food security and sustainable development
• Applied machine learning and data science
• Global public health and nutrition

Background:

• Food insecurity is a major global challenge hindering the 2030 Agenda for Sustainable Development.
• Near real-time data from organizations like the World Food Programme is vital for monitoring food consumption trends.
• Predictive modeling can enhance early warning systems for food crises.

Purpose of the Study:

• To develop and evaluate a forecasting model for predicting insufficient food consumption trends.
• To assess the impact of historical data availability on model performance.
• To provide decision-makers with a tool for near-future food insecurity assessments.

Main Methods:

• Utilized gradient boosted regression trees for the forecasting model.
• Combined food consumption observations with secondary data on conflict, extreme weather, and economic shocks.
• Applied the model to 6 countries: Burkina Faso, Cameroon, Mali, Nigeria, Syria, and Yemen.

Main Results:

• Model performance is significantly influenced by the number of available historical observations.
• Accurate 30-day forecasts of insufficient food consumption were achieved for Syria and Yemen, outperforming naive approaches.
• The framework demonstrates the added value of continuous, sub-national, near real-time data collection.

Conclusions:

• Machine learning models can effectively forecast food insecurity trends with sufficient historical data.
• Near real-time, sub-national data collection is crucial for accurate and timely food security monitoring and prediction.
• The developed framework offers a valuable tool for proactive decision-making in regions vulnerable to food insecurity.