Targeting urban poverty and food insecurity: A community-informed spatial analysis and machine learning approach

Recent advances in poverty prediction at a national scale employ new data sources and machine learning (ML) techniques. However, evidence related to the performance of these models for households experiencing poverty, food insecurity, and nutritional deficiency in urban areas is scarce. This research explores how geospatial indicators, particularly those informed by community insights, improve poverty prediction and household targeting for social transfers in urban settings. To study this issue, we combine data from a household survey collected in Lusaka, Zambia, with geospatial features such as neighborhood structure, access to infrastructure and services, satellite images, and connectivity metrics. By leveraging artificial intelligence (AI)/ML techniques, we use these features to predict welfare outcomes, focusing on asset-based wealth and food security. We find that algorithms that combine community-informed spatial features with household-level information are effective in predicting asset-based wealth scores, explaining 67.1% of the variance. While predicting food insecurity and iron deficiency is more challenging, incorporating spatial features still enhances prediction accuracy. As a result, the integration of community-informed spatial features leads to substantial gains in simulated poverty reduction, while also enhancing the transparency of targeting algorithms and addressing potential legitimacy concerns. Our process of aligning community insights with spatial data is adaptable to other urban settings. Policymakers can apply this methodology to build scalable and shock-responsive social safety net systems to address poverty and geographic inequality in urban areas.