Identification of real estate submarkets by detecting communities in graphs

Resumo

The identification of submarkets plays an important role in the context of real estate mass appraisal, to
eliminate spatial autocorrelation, or the heteroscedasticity of residues in linear models, making statistical inference
more reliable and allowing a better interpretation of the formation of property prices. Traditionally carried out
empirically, by grouping techniques or by spatial residual modelling applied to hedonic data from properties in a
specific region, studies indicate that the use of data mining models has obtained promising results to carry out this
task automatically. In this work, considering a regionalized and not individual view of the real estate unit, it is
proposed to discover submarkets based on the detection of communities in graphs formed by the neighborhoods
of a city. From the socioeconomic and locational information of the city of Rio de Janeiro, a complex network of
neighborhoods is formed, and communities are identified based on this network's modularity. The hierarchical
approach to obtaining the community structure used here allows different scenarios for understanding the
submarkets, an important aspect for decision makers in an area of knowledge so susceptible to uncertainty.