Abstract
Up-to-date poverty maps are an important tool for policy makers, but untilnow, have been prohibitively expensive to produce. We propose a generalizableprediction methodology to produce poverty maps at the village level usinggeospatial data and machine learning algorithms. We tested the proposed methodfor 25 Sub-Saharan African countries and validated them against survey data.The proposed method can increase the validity of both single country andcross-country estimations leading to higher precision in poverty maps of 44Sub-Saharan African countries than previously available. More importantly, ourcross-country estimation enables the creation of poverty maps when it is notpractical or cost-effective to field new national household surveys, as is thecase with many low- and middle-income countries.