Mapping Africa's Infrastructure Potential with Geospatial Big Data and Causal ML

Abstract

Using rich geospatial data and causal machine learning (ML), this paper maps potential economic benefits from incremental investments in all major types of public and economic infrastructure across Africa. These 'infrastructure potential maps' cover all African populated areas at a spatial resolution of 9.7km. They show that the local returns to infrastructure are highly variable and context-specific. For example 'hard infrastructure' such as paved roads and communications is more beneficial in cities, whereas 'social infrastructure' such as education, health, public services and utilities is more critical in rural areas. Market access and agglomeration forces largely govern these returns. The open Africa Infrastructure Database built for this project provides granular data classified into 54 economic categories. Its exploration further reveals that Africa's infrastructure is concentrated and often inefficiently allocated. All notable findings are consistent with economic literature, highlighting causal ML's ability to extract insights from geospatial data and assist spatial planning.

Keywords: Africa, infrastructure, investment potential, geospatial big data, causal ML, explainable AI

JEL Classification: O18, R11, R40, C14