Urban Vulnerability and Planned Relocation in Belo Horizonte, Brazil: A Machine Learning Approach

Climate change, global warming, incomplete urbanization processes, and inequality in access to adequate housing are factors that often intertwine in major Brazilian cities. In this context, in areas affected by disasters, planned relocation emerges as one of the main strategies for adaptation and for ensuring the safety of the most vulnerable populations. This study presents the case of Belo Horizonte, one of Brazil’s largest cities, which not only has around 16% of its population living in areas at risk of hydrogeological disaster such as floods and landslides but has also maintained a planned relocation policy since the 1990s. To understand how demographic characteristics influence the planned relocation process, the Partitioning Around Medoids method, an unsupervised machine learning technique used for clustering, was applied considering socioeconomic and demographic variables. The analysis aimed to identify the profiles of families relocated through this policy between 2010 and 2019, based on datasets provided by the Municipal Government. The results indicate that the presence of children is recurrent among relocated households, suggesting greater risk aversion among families with children. Relocations were also found to be concentrated in the early stages of the family life cycle, whereas profiles such as older adults or extended families tend to remain exposed to risk due to the absence of relocation. Moreover, evidence of selectivity in the relocation of certain profiles was identified when compared to the general population living in officially classified risk areas, pointing to the existence of groups potentially “trapped” in situations of vulnerability.