Authors: Manoj Kumar Yadav, Deepak Kumar Singh
Abstract: Air pollution has emerged as one of the most serious environmental and public health challenges in South Asia, with fine particulate matter (PM2.5) identified as the most pernicious pollutant due to its ability to penetrate deep into the human respiratory system. Uttar Pradesh, the most populous state in India, frequently records PM2.5 concentrations that exceed national and international standards. This study presents an integrated framework that combines geostatistical interpolation and machine learning regression to predict PM2.5 levels across ten non-attainment cities in Uttar Pradesh. Daily PM2.5 data for the period 2021–2024 were obtained from continuous monitoring stations and subjected to rigorous preprocessing. Spatial interpolation using Ordinary Kriging was implemented to generate high-resolution exposure surfaces, while machine learning algorithms including Random Forest, Gradient Boosting Regressor, Extreme Gradient Boosting, Support Vector Regression, and K-Nearest Neighbour were trained to capture temporal and spatial variability. Results demonstrate that PM2.5 concentrations consistently exceeded permissible limits, with pronounced seasonal peaks in winter and relative minima during monsoon months. Kriging revealed spatial clustering of pollution hotspots in Ghaziabad, Kanpur, and Lucknow, while peripheral cities exhibited lower but still concerning levels. Among machine learning models, XGBoost achieved the highest predictive performance with R² values above 0.74, followed by Gradient Boosting. Integration of Kriging-derived features into machine learning workflows improved prediction accuracy by 8–12%. The study demonstrates that hybrid geostatistical–machine learning approaches provide reliable and high-resolution PM2.5 predictions, enabling early-warning systems, spatially targeted interventions, and evidence-based policy planning.
