yenguandroid/Improved-Forecasting-Financial-Inclusion-in-Ethiopia

# Ethiopia Financial Inclusion Forecast A data pipeline, event-impact model, and interactive dashboard that forecasts Ethiopia's Account Ownership and Digital Payment Usage rates through 2027 -- built on a transparently-sourced dataset (every figure traceable to a real citation) and explained with SHAP so the forecast's drivers are auditable, not a black box. ## Business Problem Ethiopia's National Financial Inclusion Strategy II (NFIS-II) set a target of 70% account ownership by 2025. The only independent benchmark for that number -- the Global Findex survey -- updates roughly every three years, leaving a long gap in which policymakers, NGOs, and mobile money operators have no defensible way to know whether the country is on track, or how much of any shortfall is attributable to specific interventions (Telebirr, M-Pesa, the Fayda digital ID, NBE interoperability mandates) already underway. This project builds that missing near-term view: a forecast that explicitly incorporates known events and their estimated effects, with honestly quantified uncertainty, rather than a naive trend line or an opaque model nobody can interrogate. ## Solution Overview A unified dataset of observations, events, targets, and modeled event-to-indicator "impact links" is enriched from a starter file, validated against a shared schema, and fed through a small pipeline: trend regression (linear and logarithmic, compared directly) plus an event-impact model that ramps each event's effect in over time and combines multiple events on the same indicator (additively by default, with a documented shared-ceiling interaction term for the one case where two products genuinely compete for the same adopters). The result is surfaced three ways: a Jupyter notebook chain for the full analysis, an interactive Streamlit dashboard for exploring it, and a SHAP explainability layer that answers *which events matter most*, *why the model predicted a specific value*, and *whether any pattern looks concer …