declerke/Kenya-Crop-Forecast

# 🌾 Kenya Crop Yield Forecaster **A production-grade agricultural intelligence pipeline that fetches 24 years of World Bank open data across 5 East African countries and 6 agricultural indicators, trains 29 Facebook Prophet time-series models with 3-year holdout evaluation, and serves a fully interactive Streamlit dashboard with 5-year forecasts, confidence intervals, drought event annotations, and one-click Excel export — running entirely free at $0 cost.** | Metric | Value | |---|---| | Dataset rows | 720 | | Countries | Kenya, Uganda, Tanzania, Ethiopia, Nigeria | | Indicators | 6 | | Years covered | 2000–2023 | | Forecast horizon | 5 years (2024–2028) | | Prophet models trained | 29 | | Models skipped (sparse data) | 1 (Ethiopia / Malnutrition — 6 obs) | | Dashboard tabs | 4 | | Lines of code (app.py) | 625 | | Data source | World Bank REST API v2 | | Infrastructure cost | $0 | --- ## 🎯 Project Goal Kenya's agricultural sector faces compounding pressures from climate variability, land-use change, and food security challenges. National planners, agribusiness analysts, and development practitioners need forward-looking data on crop yields, food production, and land utilisation — but publicly available forecasting tools are either paywalled, too coarse for country-level analysis, or reliant on static datasets. This project builds a fully automated pipeline that: 1. Pulls live annual data from the World Bank Open Data API for Kenya and four regional peer countries. 2. Trains individual Prophet time-series models per country and indicator combination, validated against a 3-year holdout. 3. Surfaces all results in a dark-theme Streamlit dashboard with adjustable forecast horizons, statistical confidence bands, and historical drought annotations. 4. Allows analysts to download a structured Excel workbook covering forecasts, raw historical values, and model performance metrics for offline reporting. The project is benchmarked against Nigeria — Africa's largest a …