How AI Supports Anticipatory Action in Forced Displacement
Refugee crises are often treated as sudden emergencies. Aid typically arrives after people have fled conflict, focusing on the communities that receive them. But what if refugee movements could be forecasted in advance? What if countries hosting refugees had time to prepare for large inflows? With over 122 million forcibly displaced people worldwide—double the number from a decade ago—these questions have become urgent.
Humanitarian and development budgets are stretched thin, so forecasting tools can improve the efficiency and impact of responses. The World Bank is exploring artificial intelligence (AI) models to help countries anticipate crises and prepare ahead. This effort aligns with working alongside UNHCR to develop responses that integrate refugees into national systems, improving social services and economic outcomes for both refugees and host communities.
Uganda’s Approach and AI Pilot Projects
Uganda has progressive policies that allow refugees to access employment, land, and public services alongside host communities. The World Bank’s Social Development and Prosperity teams have piloted AI-based methods to strengthen preparedness within the Development Response to Displacement Impacts Project (DRDIP). DRDIP invests in social service infrastructure and livelihoods in refugee-hosting districts.
DRDIP includes a contingency fund called the Displacement Crisis Response Mechanism (DCRM), which releases rapid funding to districts facing pressure from large refugee inflows. However, funds were typically disbursed only after large inflows occurred and the host communities felt the impact.
At Uganda’s request, the World Bank studied how AI could predict refugee inflows, allowing contingency funds to be triggered earlier. This would enable host districts to expand public services before refugees arrive, reducing strain and improving outcomes.
AI and Machine Learning to Predict Refugee Inflows
The World Bank developed an AI-driven model to forecast refugee flows into Uganda from the Democratic Republic of the Congo (DRC) and South Sudan. The model uses diverse data sources, including conflict incidents, climate patterns, vegetation, built infrastructure, economic indicators, and online language sentiment. This reflects the idea that human behavior depends not only on measurable changes but also on perceptions of those changes.
The model analyzes over 90 independent variables with machine learning to predict daily refugee arrivals. It was tested on data not used during training and achieved over 80% accuracy in forecasting changes in refugee inflows from the DRC and South Sudan.
Key Drivers of Refugee Movements
- Armed conflict
- Economic activity
- Climate conditions
- Food prices
- Volume and sentiment of language about these factors
Beyond forecasting, the model identifies which factors most strongly influence refugee flows. This insight helps governments and humanitarian actors anticipate movements and better address displacement drivers. In refugee-origin areas, this evidence supports more effective interventions. It also informs World Bank country strategies, project design, and policy discussions.
Transforming How DCRM Works
Integrating the AI model allows the DCRM to begin scaling up public services 4 to 6 months before refugees arrive. This early action enables timely construction of schools, health facilities, and water points. By responding proactively, the World Bank can provide social infrastructure efficiently while supporting integration of refugees into national services and local planning.
This approach helps Uganda allocate limited resources more effectively, reduces tension between communities, and improves operational planning.
For those interested in AI applications in development and government sectors, exploring AI forecasting tools and training can provide valuable skills. Learn more about practical AI courses at Complete AI Training.
Your membership also unlocks:
