Urban Zoning Modeling from Territorial Data: A Machine Learning Approach

This study proposes an urban zoning modeling approach based on machine learning techniques, such as data-driven analytical representation and aimed at decision support in urban planning. The methodology integrates socioeconomic, morphological and road structure variables in a multidimensional basis associated with urban blocks. Supervised classification algorithms were evaluated, with emphasis on the Random Forest, which presented the best performance. The results indicate that zoning categories can be associated with distinguishable territorial patterns, evidencing the model's ability to represent the complexity of urban structure. The comparison between the current and modeled zoning allows an alternative reading of the territory based on the integration of multiple attributes. It is concluded that the model is a complementary instrument to urban planning, contributing to territorial analysis and decision support in urban review processes.