Stacking ensemble machine learning and geospatial approach for sustainable development: an application to regional inequality in Ghana

Abstract

This paper introduces a novel methodological framework combining stacking ensemble machine learning with multi-source geospatial data to measure regional inequality in Ghana from 1994 to 2020. Our approach integrates diverse algorithms, such as Random Forest, Gradient Boosting Machines, and Support Vector Regression, to generate superior estimates of regional economic activity from satellite-derived environmental, health, and infrastructure variables. The analysis reveals that while regional disparities have modestly declined since the mid-1990s, this trend accelerated after 2014, coinciding with significant policy reforms. Nighttime lights intensity emerges as the most powerful predictor of regional economic performance, followed by population density and malaria prevalence, highlighting the critical importance of infrastructure development, demographic dynamics, and health conditions in shaping spatial inequality. Environmental variables including vegetation indices and temperature measures provide additional explanatory power, demonstrating the interlinked nature of infrastructure, demographic, health, and environmental conditions in determining regional development outcomes. We also document a persistent north–south divide and heterogeneous Kuznets relationships across regions. Our findings suggest that integrated policy approaches prioritizing industrialization and infrastructure development, alongside targeted interventions addressing demographic transitions, health improvements, and climate resilience, could significantly reduce regional disparities. Our methodology offers a scalable framework for high-frequency monitoring of sustainable development in contexts where conventional data collection faces significant constraints.