MastaT002/Drought-Food-Security-Early-Warning-System

ML-powered early warning system for predicting high food insecurity risk across Kenya ASAL counties. Combines IPC food security outcomes, rainfall, NDVI vegetation, and food price indicators to support early risk monitoring and humanitarian planning. #DataScience #FoodSecurity #Kenya # **Drought & Food Security Early Warning System** A data analytics and machine learning portfolio project focused on building an early-warning system for food insecurity risk across Kenya's Arid and Semi-Arid Lands (ASAL). **Data sources:** IPC food insecurity outcomes, CHIRPS rainfall data, MODIS NDVI vegetation indicators, and WFP Kenya food price data. **Current stage:** Baseline model completed, food price features added, enhanced model tested, and Streamlit dashboard being updated to compare baseline vs food price model predictions. **Latest enhanced model result:** XGBoost with rainfall, NDVI, and food price features achieved the strongest overall performance after adding market indicators. **Dashboard output:** County-level prediction risk tables and an interactive map dashboard were created for the baseline XGBoost model and the enhanced food price XGBoost model. **Live dashboard:** Kenya Drought & Food Security Risk Dashboard **Next stage:** Improve the dashboard comparison view, test longer time-lag features, tune the enhanced model, and improve validation across future IPC periods. --- ## The Problem * **3.7 million Kenyans** are projected to face IPC Phase 3 "Crisis" level food insecurity between April–June 2026 * **23 ASAL counties** are particularly vulnerable to drought-driven food crises * Humanitarian agencies often react to crises **after** they peak, not before * There is limited publicly available county-level modeling that links food insecurity outcomes with rainfall and vegetation indicators for early-warning analysis ## Objective Build an interpretable machine learning early-warning model that predicts high-risk food insecurity cases across Kenya ASAL counties using publicly available food security, climate, vegetation, and market price data. In this project, a county-period is classified as **high risk** if **20% or more of the county population is in IPC Phase 3 or worse**. The baseline model uses rainfall and NDVI indicators, …