Modelagem integrada de redes Multilayer, Análises Espaciais e Aprendizado de Máquina na investigação da disseminação de COVID-19 e de arboviroses (Dengue, Zika e Chikungunya) em Fortaleza

Abstract

The rise of arboviral diseases—and, in particular, COVID-19—calls for approaches that integrate multiple territorial dimensions. This thesis adopts an integrated approach combining multilayer networks, statistical and spatial analyses, and interpretable machine learning to understand diffusion patterns of COVID-19, dengue, zika, and chikungunya in Fortaleza (CE), 2020–2024. Layers comprise epidemiological, socioeconomic/demographic, infrastructure, and political/HDI variables; intralayer relations are defined by neighborhood similarity, and interlayer relations by associations between variables. The COVID-19 dataset was built from weekly epidemiological bulletins, aggregated by month and neighborhood, prioritizing territorial patterns comparable across diseases. Spatially, we use global and local autocorrelation (Moran’s I, LISA) to identify clusters and local patterns; in the network domain, we compute centralities and apply community detection. Predictive models (Random Forest, XGBoost) and SHAP-based explanations assess the variables’ predictive power and contributions. K-means clustering synthesizes territorial profiles by combining structural and epidemiological characteristics. Findings reveal spatial heterogeneity and cross-layer differences, with territorial convergences and divergences that inform vulnerability assessment and prioritization. As a practical contribution, the thesis includes an Appendix: Methodological Enhancement Proposal (“Integrated Framework”) to guide future standardization and support municipal decision-making.