A MACHINE LEARNING AND SPATIAL CLUSTERING FRAMEWORK FOR URBAN AIR QUALITY PREDICTION

ABSTRACT:

Urban air pollution presents a significant public health risk, com- pounded by the complexity of multipollutant interactions and modelling uncertainty. This study integrates spatial analysis with ElasticNet regression to predict the Air Quality Index (AQI) using geolocated pollutant data. ElasticNet effectively handles multicollinearity while maintaining model interpretability. We also apply spatial clustering to categorize cities by pollution profiles and use bootstrapped confidence intervals to assess the stability of the prediction. The results show that PM2.5 is the dominant AQI driver and the model outperforms the baseline regressors. The approach provides a scalable framework for assessing urban air quality and prioritizing policy interventions.

Keywords: air pollution, ElasticNet, AQI prediction, PM2.5, spatial clustering, uncertainty quantification