Diini03/somalia-displacement-classifier

# Somalia Displacement Severity Classifier A machine learning project that predicts whether an incoming displacement event in Somalia will be large (affecting more than 100 people) or small — based on early signals like the cause, region, month, and event duration. Built as the capstone project for the Goobo Labs DS/ML Bootcamp 2026. --- ## Problem Humanitarian organizations operating in Somalia — UNHCR, OCHA, Medair — need to pre-position supplies and staff before a displacement event peaks. Waiting for the full picture is already too late. This model gives field officers a severity prediction from just 4 inputs, creating a 1–3 week window to respond. --- ## Dataset - **Source:** Internal Displacement Monitoring Centre (IDMC) via HDX - **Link:** data.humdata.org - **Size:** 3,091 displacement events (2025–2026) - **Target:** `is_large_event` — 1 if more than 100 people displaced, 0 otherwise --- ## Models Trained | Model | Accuracy | Recall | F1-Score | | ------------------- | -------- | ------ | -------- | | Logistic Regression | — | — | — | | Random Forest | — | — | — | | XGBoost ✅ | — | — | — | > **Best model: XGBoost** — selected by highest Recall. In humanitarian response, missing a large event (false negative) is more costly than a false alarm. --- ## Project Structure ``` somalia-displacement-classifier/ ├── dataset/ │ ├── som_idmc_idu_events.csv ← raw data │ └── clean_dataset.csv ← generated by notebook 01 ├── notebooks/ │ ├── 01_eda_and_preprocessing.ipynb │ └── 02_modeling.ipynb ├── api/ │ ├── app.py ← FastAPI backend │ └── index.html ← UI ├── models/ │ ├── best_model.pkl ← XGBoost (best) │ ├── lr_model.pkl │ ├── rf_model.pkl │ ├── scaler.pkl │ └── label_encoders.pkl ├── README.md ├── requirements.txt └── project_paper.md ``` --- …