emmyCode4495/macro-vulnerability-africa

Macroeconomic Vulnerability Forecasting for Sub-Saharan Africa # Macroeconomic Vulnerability Forecasting — Sub-Saharan Africa An end-to-end data science / ML project that forecasts macroeconomic vulnerability (inflation shocks, debt distress, unemployment stress) across 20 Sub-Saharan African economies using open World Bank data, and maps how countries' economic indicators co-move using network analysis. This project extends prior published research on youth unemployment, public debt sustainability, and inflation dynamics in the region, moving from literature review to a live, reproducible ML pipeline. **Full report:** `reports/final_report.pdf` **Live demo:** _(add your Streamlit Cloud URL here after deploying — see below)_ ## Project status - [x] Phase 1 — Data collection pipeline - [x] Phase 2 — EDA & feature engineering - [x] Phase 3 — Modeling (classical ML, boosting, one-year-ahead early warning) + network analysis - [x] Phase 4 — Dashboard + report ## Headline results - **One-year-ahead early-warning classifier: ROC-AUC 0.85**, evaluated on 4 countries held out entirely from training (not just unseen years — unseen *economies*). - The composite vulnerability index independently recovers real, documented crises without being told about any of them: Angola's post-civil-war hyperinflation (2001-2003), Zambia's early-2000s debt distress, Malawi's 2022-2023 crisis, and **Mozambique 2016 — the exact year its hidden sovereign-debt scandal broke.** - SHAP interpretability confirms the model's predictions align with standard macroeconomic theory (low reserves, current account deficits, high inflation and lending rates all push toward "high vulnerability"). - Network analysis identifies Cameroon and Mozambique as the most systemically connected economies in the sample by eigenvector centrality. See `reports/final_report.pdf` for full methodology, figures, and a discussion of limitations. ## Data currency The pipeline always requests data through the current year — `END_YEAR` in `src/config.py` is set dynamically (`dateti …