Machine learning approach to PM2.5 forecasting and health risk assessment during stubble burning period in Delhi

Delhi experiences higher air pollutant levels in October and November owing to burning agricultural residue in neighboring states, adverse meteorological conditions, and anthropogenic emissions. This study explores the health concerns related to exposure to high concentrations of PM2.5 in Delhi due to stubble burning between 2018 and 2023 along with forecasting PM2.5 using different machine learning approaches. We have employed the AirQ + model to assess the health risks due to short-term and long-term exposure to elevated PM2.5 concentration during the stubble-burning months. The respiratory health of residents of Delhi was found to be significantly vulnerable to short-term exposure to PM2.5 (6-year average: estimated attributable proportion (EAP) = 18.06%, estimated number of cases (ENC) = 46,278, and estimated number of cases per 100,000 population at risk (ENCPR) = 228). Long-term exposure to PM2.5 during stubble-burning months significantly impacted mortality rates due to chronic obstructive pulmonary disease (COPD) and stroke in adults and mortality due to acute lower respiratory infection (ALRI) in children. Additionally, three machine learning models: Artificial Neural Network (ANN), Support Vector Machine using Radial Basis Function (SVM RBF), and Random Forest (RF) were utilized for PM2.5 forecasting in stubble-burning months. The random forest was found to be consistent in the training phase (R2 = 0.895, RMSE = 0.058, and MAE = 0.037) as well as in the testing phase (R2 = 0.842, RMSE = 0.06, and MAE = 0.045) for PM2.5 forecasting. The novelty of this study lies in its integrative approach that combines health risk assessment with comparative machine learning based PM2.5 forecasting in stubble-burning months in Delhi. The outcomes will be helpful in informed policy formulation along with potential mitigation strategies for improving the air quality of Delhi. Copyright © 2025 American Association for Aerosol Research