Deciphering particulate matter dynamics in Asansol city: A novel approach using machine learning for enhanced prediction accuracy

Understanding and predicting Particulate Matter (PM) concentrations is critical for managing urban air quality. This study evaluates machine learning models, including Ridge Regression, Lasso Regression, Gaussian Process Regression, XGBoost, and Recurrent Neural Networks (RNN), to forecast PM 2.5 and PM 10 levels in Asansol, India. The XGBoost model emerged as the best performer with an R² of 0.97, low Mean Error (-0.06), and Root Mean Square Error (4.0). Seasonal trends and meteorological influences were analysed, revealing NO₂, Barometric Pressure, and SO₂ as key predictors of PM levels. Variable importance ranking and correlation analyses provided insights into the complex interactions between air pollutants and environmental factors. The findings highlight the potential of XGBoost for precise pollution forecasting, aiding policy interventions and sustainable urban development.