Predicting Air Pollution Occurrence Using Climate-Responsive Machine Learning Models

Rapid urbanization and climate-induced shifts in meteorological conditions are increasingly influencing air quality in megacities worldwide. Dhaka, Bangladesh, provides a particularly important case study because of its exceptionally high population density, rapid urban expansion, and growing vulnerability to changing weather patterns. This study investigates the relationship between meteorological variability and particulate matter concentrations (PM₂.₅ and PM₁₀) in Dhaka, one of the world's largest and fastest-growing urban centers [1]. A one-year dataset comprising more than 7,000 hourly observations of rainfall, relative humidity, maximum and minimum temperature, wind speed, PM₂.₅, and PM₁₀ was obtained from a publicly available repository. Climate-responsive machine learning models were developed using meteorological variables as predictors and particulate matter concentrations as target variables. The modeling framework integrates ensemble-learning approaches and deep neural networks to capture the complex and nonlinear interactions between weather conditions and urban air pollution. Model performance was evaluated through training, validation, and testing using an 80/20 data split. The model was evaluated by regression, error, and feature-importance analyses. The results demonstrated that both ensemble-based and deep-learning models can accurately predict particulate matter dynamics, while also identifying the dominant meteorological drivers influencing pollution levels. By focusing on Dhaka's unique urban and environmental conditions, this study provides new insights into weather-sensitive air quality prediction and establishes a foundation for assessing how future climate-driven changes in meteorological patterns may affect particulate matter exposure in rapidly urbanizing megacities.

¹United Nations, Department of Economic and Social Affairs, World Urbanization Prospects 2025