Métodos de inteligência computacional aplicados em análises espaço temporais de poluição atmosférica, condições climáticas e socioeconômicas e saúde populacional

Abstract

The use of computational intelligence has expanded across engineering and environmental sciences, enabling the integration and analysis of large datasets to understand complex systems and support evidence-based decision-making. In the field of air pollution and public health, there has been a growing application of machine learning algorithms and multivariate analysis; however, studies that integrate socioeconomic, environmental, and health dimensions in spatiotemporal analyses remain limited. The objective of this study was to develop and apply computational methods based on clustering algorithms and multivariate analysis to identify spatiotemporal patterns of air pollution, climatic conditions, and respiratory hospitalizations in municipalities of Paraná, Brazil, by integrating environmental, meteorological, socioeconomic, and health data at different levels of granularity (daily, monthly, and seasonal), collected between May 2024 and May 2025. The methodology involved the collection and integration of daily air quality data, meteorological variables, and hospital records, associated with income, sanitation, and human development indicators. Hierarchical and K-Means clustering techniques were applied and validated using metrics such as Silhouette (0.43-0.44), Davies-Bouldin (0.68-0.89), and Calinski-Harabasz (11.09-393.5), along with Principal Component Analysis (PCA) and bootstrap-based stability testing. The quantitative results revealed the formation of three distinct seasonal profiles (low, intermediate, and high pollutant variability), with higher PM₂.₅ concentrations and respiratory hospitalizations during winter. The qualitative analysis indicated that municipalities with lower income and inadequate sanitation exhibit greater environmental and health vulnerability, underscoring the role of socioeconomic determinants in exposure and disease burden. It is concluded that the integration of multivariate statistics, unsupervised learning, and socioenvironmental indicators represents a methodological advancement in the scientific field, enhancing the capacity to identify regional vulnerability patterns and supporting evidence-based environmental and public health policies.