@techreport{TR-IC-PFG-25-53,
   number = {TR-IC-PFG-25-53},
   author = {João Theophilo Morais Santos da Silva},
   title  =  {{Explainable Machine Learning for Mapping Informal Urban
                   Centers}},
   month = {December},
   year = {2025},
   institution = {Institute of Computing, University of Campinas},
   note = {In English, 10 pages.
    \par\selectlanguage{english}\textbf{Abstract}
       Informal  urban  centers  represent  a persistent challenge for
       urban planning in Brazil due to their heterogeneity, structural
       precariousness,    and    the    scarcity    of   standardized,  
       high-resolution   data.   In   this   study,   we   propose  an 
       interpretable  machine  learning  approach to classify informal
       urban  centers  across  six Brazilian hubs using an Explainable
       Boosting  Machine  (EBM).  A  spatially  structured dataset was
       built  by  intersecting  census-based  socioeconomic indicators
       with  the official mapping of informal urban centers, producing
       a  binary  classification framework. To evaluate the robustness
       of  the  model  under spatial heterogeneity, we implemented two
       training   strategies:   a   weighted   model  incorporating  a 
       city-propensity  adjustment  and  a  standard unweighted model,
       each  tested  on  a  held-out hub. Results for the Porto Alegre
       test hub show that the weighted model achieves an AUC of 0.826,
       outperforming  the  standard model at 0.818. Feature importance
       rankings  indicate consistent relevance of demographic density,
       income  indicators,  and  local  infrastructure.  The  findings 
       highlight  the  potential  of  interpretable  models to support
       policy-oriented  spatial  classification  tasks, providing both
       predictive  accuracy  and  transparent  insights that can guide
       targeted urban interventions.
  }
}