Explainable Panel Econometrics for Maternal Health Policy Intelligence: Evidence from 179 Countries, 2000–2022

Maternal mortality remains one of the starkest indicators of global health inequity, with Sub-Saharan Africa accounting for roughly 70% of all maternal deaths worldwide. This paper presents an integrated explainable AI framework combining panel econometrics and machine learning to quantify the determinants of maternal mortality and generate actionable policy counterfactuals across 179 countries from 2000 to 2022. Using World Bank World Development Indicators data, we estimate Two-Way Fixed Effects (TWFE) models with country-clustered standard errors to formally quantify selection bias in health expenditure elasticities. A Mundlak–Chamberlain Hausman test (F=126.40, p<0.001) confirms fixed effects over random effects. The cross-sectional health expenditure coefficient attenuates substantially under within-country identification (Δβ=+0.173), while fertility and female education remain robust. Random Forest and Gradient Boosting ensembles are evaluated on a temporal holdout (train 2000–2015; test 2016–2022), achieving holdout R²=0.770 and 0.810 respectively. We demonstrate that random cross-validation on panel data inflates apparent performance by 7.9 percentage points due to country-level leakage. SHAP attribution identifies skilled birth attendance as the dominant predictive feature across all four importance metrics, with female secondary education second. A Ghana case study reveals a 31% prediction gap, highlighting equity risks in globally-aggregated models. Policy simulations show no single macro-level lever breaches the SDG 3.1 threshold of 70 deaths per 100,000, motivating coordinated multi-lever approaches. An open-source Streamlit dashboard operationalises findings for policy audiences at who-mch-analytics.streamlit… Keep Visibility as Public, then click the green Publish button.