Food Insecurity Early Warning System
MSc Dissertation — Top 5 Best Project Award | Two-stage cascade ML rescuing 249 missed crises across 18 African nations using GDELT news signals
Technology Stack
XGBoostSHAPscikit-learnHidden Markov ModelsDynamic Mode DecompositiongeopandasstatsmodelsPythonJupyter
ML Frameworks
Deep Learning
NLP
Deployment
ROI
17.4% rescue rate
Accuracy
90.7% AUC (AR Baseline)
Key Performance Indicators
Countries Covered
18 African nationsDistricts Analysed
1,920 unique districtsObservations
20,722 district-periodsAward
Top 5 Best Project — MSc Data ScienceBusiness Impact Summary
| Return on Investment | 17.4% rescue rate |
| Model Accuracy | 90.7% AUC (AR Baseline) |
| Countries Covered | 18 African nations |
| Districts Analysed | 1,920 unique districts |
| Observations | 20,722 district-periods |
| Award | Top 5 Best Project — MSc Data Science |
Overview
MSc Dissertation achieving Distinction and Top 5 Best Project Award at Middlesex University London. Developed a two-stage cascade early warning framework for food crises across 18 Sub-Saharan African countries. Combines an autoregressive (AR) logistic regression baseline with an XGBoost news-signal model to rescue 249 crises missed by the AR model — particularly in conflict-affected regions like Zimbabwe, Sudan, and DR Congo.