Designing Forecasting Models using Composite AutoML for Air Quality Modeling in a Coastal Urban Area

Resumo

Air pollution remains one of the most critical environmental risks to human health, particularly in urban areas where industrial activities and vehicular emissions prevail. Accurate forecasting of air quality plays a pivotal role in public health protection and supports global initiatives such as the United Nations Sustainable Development Goals (SDGs), in special  SDG 3 (Good Health and Well-being) and SDG 11 (Sustainable Cities and Communities). In this study, we explore the use of machine learning (ML) and automated machine learning (AutoML) approaches for forecasting air quality time series data. As a case study, we employed a comprehensive dataset collected in Vitória, the capital of Espírito Santo State, Brazil—an urban and industrial coastal region frequently affected by atmospheric pollutants. Predictive modeling pipelines were automatically designed and optimized using the FEDOT AutoML framework, enabling the capture of both linear and non-linear temporal behaviors. The performance of the AutoML-generated models was compared against conventional forecasting methods typically employed in air quality modeling. The findings show that AutoML achieves competitive predictive accuracy while significantly reducing the human effort in model development. Furthermore, it was observed that the integration of domain-specific physical knowledge enhances model robustness. This work highlights the potential of AutoML to streamline the development of adaptive and scalable models for air quality monitoring, offering valuable insights for environmental policy-making, public health planning, and sustainable urban development.