Mapping Africa’s Infrastructure Potential with Geospatial Big Data, Causal ML, and XAI

Using rich geospatial data analyzed with 'causal' and explainable machine learning methods, this paper studies Africa’s spatial economy and maps potential benefits from public infrastructure investments. Results imply that infrastructure in Africa is spatially concentrated and often inefficiently allocated relative to the location of population and economic activities. Important spatial predictors of wealth and economic activity are paved roads, travel time to cities and ports, population, power infrastructure, education, transport, and residential areas. These predictors interact heavily with one another. Marginal effects imply a 0-10% improvement in GDP and a 0-1 point increase in the International Wealth Index in response to 100% increases in different infrastructures, with transport investments having the highest average effect. Heterogeneous spatial effects suggest that transport and power have stronger effects in urban areas, whereas education, communications, and public services show stronger effects in semi-urban and rural areas. Simulating approximately equal cost investments to all locations suggests higher development impact in rural areas, but greater monetary returns in urban areas. Simultaneous (Big-Push-type) investment simulations provide evidence for localized infrastructure complementarities, especially in urban areas