Análise espaço-temporal dos municípios brasileiros: aplicações de aprendizado de máquina na integração de dados territoriais e populacionais

Abstract

This study sought to explore the territorial, demographic, and environmental dynamics of Brazilian municipalities through an integrated data approach based on machine learning. The research is justified by the need to understand how different municipal characteristics relate to each other and evolve over time, considering aspects such as changes in territorial configuration, land use and land cover patterns, and demographic transformations. The methodology adopted involves the collection and processing of data from official databases, such as IBGE and MapBiomas, allowing the construction of a structured dataset suitable for the application of statistical and artificial intelligence techniques. The first step consists of defining the municipality as the unit of analysis, considering its institutionalization and evolution over time. A special focus is given to the period from 2003 to 2023, when there was a stabilization in the emancipatory processes. In addition, regional differences in municipal formation are analyzed, highlighting how each Brazilian macro-region presents distinct patterns of creation and expansion of municipalities. Next, the territorial characteristics of the municipalities were investigated, covering aspects such as territorial extension, spatial variations, and environmental factors that influence their configuration. From this, statistical inferences were drawn to assess significant differences between municipalities, using methods such as ANOVA to understand variations between regional groups. The research advances with the application of supervised and unsupervised machine learning techniques. Finally, Explainable Artificial Intelligence (XAI) techniques were used to interpret the generated models and understand the importance of each variable in municipal behavior. The identification that the formation of clusters of municipalities, based on territorial, population, and environmental characteristics, correspond more to the natural biomes of Brazil than to the traditional divisions by state or region. Thus, municipalities with similar environmental and land use characteristics, even if belonging to different states, were grouped into highly homogeneous clusters, demonstrating the strength of data-based and machine learning approaches for identifying spatial patterns that escape political-administrative divisions. Thus, the dissertation proposes a new look at municipal organization in Brazil, combining different data sources and methodologies to deepen the understanding of the transformations that have occurred in the national territory.