This repository contains an end-to-end pipeline and dashboard for exploring and forecasting staple food prices across regions in Ethiopia.
The project:
- Cleans and harmonises monthly price data from multiple sources.
- Builds an analysis-ready “Tier A” panel of staple food prices.
- Benchmarks several time-series models (Naive, ARIMA/SARIMA, XGBoost, hybrids).
- Serves an interactive Streamlit dashboard for exploration & 3-month forecasts.
Goal:
Provide a lightweight, transparent forecasting tool for staple food prices in Ethiopia, focused on a short-term 3-month planning horizon for humanitarian / food security use cases.
Key design choices:
- Unit of analysis:
(admin_1, product)(region–product pairs). - Temporal unit: monthly data.
- Target:
value_imputed(retail prices per standard unit, cleaned & imputed). - Operational forecast model: Naive (last observed value) with horizon = 3 months.
A full exploration + modelling notebook exists (model_comparison_and_export.py logic), but the dashboard only uses the final chosen model + comparison table.