Predicting where climate-driven displacement will occur in Africa before it happens, so NGOs and humanitarian organizations can intervene early, protect vulnerable communities, and prevent refugee crises before they begin.
Every year across Africa, droughts dry up farmland, floods destroy homes, and communities are forced to flee. This cycle of climate shock leading to displacement and then to refugee crisis is not random. It is predictable.
Yet humanitarian response remains largely reactive: aid arrives after displacement has already occurred, after families have already been torn apart, after communities have already collapsed.
In 2024 alone, over 35 million people across Sub-Saharan Africa were forcibly displaced, with the majority driven by climate-related events compounded by food insecurity and conflict.
The question AfriShield asks is not "how do we respond?". It is:
"Where will the next crisis happen, and how do we get there first?"
AfriShield is a data science tool that combines historical climate disaster data, displacement records, and socioeconomic vulnerability indicators to:
- Profile vulnerability by identifying which communities are most at risk before a disaster strikes
- Predict displacement risk using machine learning to forecast which regions face the highest climate-driven displacement risk in the next 6 to 12 months
- Surface actionable insights that give NGOs and humanitarian organizations a clear, prioritized map of where to intervene now to prevent displacement
AfriShield is a vaccine, not a cure. It targets prevention at the source, breaking the chain between climate shock and forced displacement before it becomes a refugee crisis.
| Audience | How They Use AfriShield |
|---|---|
| NGO Program Officers | Identify priority intervention zones before crises emerge |
| Humanitarian Coordinators | Pre-position resources in high-risk regions |
| Policy Makers | Evidence base for climate adaptation and displacement prevention policy |
| Researchers & Academics | Reproducible pipeline for studying climate-displacement linkages |
| Faith & Community Leaders | Understand which communities need solidarity and support now |
Climate Data Displacement Data Socioeconomic Data
(EM-DAT droughts + (UNHCR refugees + (World Bank poverty,
& floods) & IDPs) water, food security)
│ │ │
└─────────────────────┴────────────────────────┘
│
┌─────────▼──────────┐
│ Data Preparation │
│ & Feature Eng. │
└─────────┬──────────┘
│
┌──────────────┴──────────────┐
│ │
┌──────────▼──────────┐ ┌──────────▼──────────┐
│ Vulnerability │ │ ML Risk Prediction │
│ Profiling │ │ Model (XGBoost) │
│ (Composite Score) │ │ 6-12 month forecast│
└──────────┬──────────┘ └──────────┬──────────┘
│ │
└──────────────┬──────────────┘
│
┌─────────▼──────────┐
│ AfriShield Dash │
│ Dashboard │
│ (NGO Interface) │
└────────────────────┘
- 🗺️ Risk Map: Pan-Africa choropleth showing predicted displacement risk score by country
- 📈 Trend Analysis: Historical climate events vs. displacement trends over time
- 🔍 Country Drill-Down: Detailed vulnerability profile per country with key indicators
- 🚨 Priority Alert List: Ranked list of countries requiring immediate preventive intervention
- 📉 Gap Tracker: Where humanitarian presence is lowest relative to risk
afrishield/
├── data/
│ ├── raw/ ← Source datasets (see data/README.md for download links)
│ └── processed/ ← Cleaned & merged data output by notebooks
├── notebooks/
│ ├── 01_data_preparation.ipynb ← Load, clean & merge all datasets
│ ├── 02_exploratory_analysis.ipynb ← EDA, distributions, correlations
│ ├── 03_vulnerability_profiling.ipynb ← Composite vulnerability scoring
│ └── 04_risk_model.ipynb ← ML model training & evaluation
├── app/
│ ├── app.py ← Main Dash application
│ └── assets/ ← CSS and static assets
├── models/
│ └── risk_model.pkl ← Trained model (generated by notebook 04)
├── data/README.md ← Dataset download instructions
├── requirements.txt
└── .gitignore
git clone https://github.com/Derekkarungani/afrishield.git
cd afrishieldpython -m venv venv
source venv/bin/activate # Mac/Linux
venv\Scripts\activate # Windowspip install -r requirements.txtSee data/README.md for full instructions on downloading all required datasets.
notebooks/01_data_preparation.ipynb
notebooks/02_exploratory_analysis.ipynb
notebooks/03_vulnerability_profiling.ipynb
notebooks/04_risk_model.ipynb
python app/app.pyThen open http://127.0.0.1:8050 in your browser.
| Dataset | Provider | Purpose |
|---|---|---|
| Climate Disaster Events | EM-DAT / CRED | Historical drought & flood data |
| Refugee & IDP Statistics | UNHCR | Displacement outcomes |
| Socioeconomic Indicators | World Bank | Vulnerability profiling |
This project was developed in the context of the Africa Regional Conference on Refugees, Migrants & Our Common Home, a convening of academic, faith-based, and community leaders sharing research and practical approaches to migration, displacement, and social inclusion across Africa.
AfriShield directly supports the conference's call for evidence-based, community-centered approaches to displacement prevention, arguing that data science, applied with urgency and care, can serve as a shield for the continent's most vulnerable communities.
Contributions, feedback, and collaboration are welcome. Please open an issue or submit a pull request.
MIT License. Free to use, adapt, and build upon with attribution.
Derek Karungani MIT Emerging Talent Certificate Program, Computer and Data Science