Urban morphology as a proxy for housing and infrastructure inequality: A machine learning approach using open building footprint data

Mapping spatial inequalities remains a major challenge, particularly in rapidly urbanizing regions. Although urban morphology offers valuable insights into the built environment, the extent to which it can serve as a proxy for urban inequalities remains underexplored. This study evaluates the potential of morphological indicators to reflect dimensions of urban precarity, which in this study refer exclusively to housing and infrastructure conditions. Using São Sebastião, Brazil, as a case study, we trained a random forest model on officially delineated slum locations, using indicators derived from Google Open Buildings, an open-access building footprint dataset, as predictors. The model achieved high accuracy in distinguishing slums from non-slums (AUC of 0.89), with over 90% of slum cells classified as either highly or very highly precarious. Validation with field observations and census data confirmed that the mapped precarity classes consistently correspond to observed conditions. Urban cells classified as more precarious are associated with smaller buildings, narrower and unpaved streets, less durable roof materials, and reduced access to basic infrastructure such as piped water, sewage, and garbage collection. These consistent gradients across precarity levels suggest that urban form is, to a significant extent, associated with these housing and infrastructure conditions. However, despite the scalability and reproducibility of the proposed approach, limitations persist, particularly in morphologically complex urban environments, where local knowledge and more advanced datasets may be necessary. Overall, this study provides evidence that urban morphological indicators can approximate key dimensions of urban precarity, especially those related to housing and infrastructure, even if they do not directly measure them. • Open building footprint data predicts patterns of urban inequalities. • Model achieves 89% accuracy in distinguishing slums from non-slums. • Field and census data confirm links between urban morphology and precarity. • Proposed methodology is scalable for cities worlwide.