PEC Framework: Madagascar Food Security Dataset for Predict-Explain-Certify Certification

This dataset supports the Predict-Explain-Certify (PEC) framework for food security prediction in Madagascar. It contains 242 observations across 22 regions, 139 variables from 13 institutional sources, and 11 years (2012–2022). Three nutritional targets are included: Insuffisance Pondérale  (Underweight, IP), Malnutrition Chronique (Stunting, MC), and Malnutrition Aiguë (Wasting, MA).

The PEC framework achieved Grade A certification scores of 94.5 (IP), 95.8 (MC), and 98.0 (MA)  out of 100, using an interpretable Ridge regression model that outperforms all black-box  alternatives (R²=0.91 vs. CatBoost R²=0.68 for IP). Multi-method triangulation with SHAP, LIME, and Permutation Importance yields Kendall τ > 0.8 across all targets. Temporal features (MA3, Lag1, Delta) dominate SHAP importance (35–42%), followed by poverty (19–25%), water/sanitation (13–15%), and climate (8–15%).                                                                                                                                                               
 The dataset includes: raw observations, engineered features (MA3, Lag1, Delta), regional metadata, SHAP values for all 22 regions, What-If policy simulation results (8 scenarios, 2026–2035), and complete PEC certification scores. Operational web platforms: https://pec-mada.streamlit.…, https://pec-aide.streamlit.…, https://pec-what.streamlit.…