Satellite-derived NO₂ data to predict hospital admissions : a machine learning-based approach

Abstract

Rising Levels of air pollution and changing environmental conditions are the global concern. There is substantial evidence linking short-term and long-term exposure to NO2 (nitrogen dioxide) with multiple disorders such as asthma, Bronchitis, Chronic Obstructive Pulmonary Disease (COPD), and Respiratory Infections in humans. However, multiple researchers have mainly depended on time-based exposure data, possibly overlooking fluctuations in NO2levels across different locations. We aim to analyze the relationship between NO2fluctuations and emergency department hospital admissions. We adopted machine learning (ML) based regression approaches to predict hospital admissions, we obtained emergency department hospital admissions from 2018–2019, while NO2tropospheric concentration was collected from the Sentinel 5P satellite. The results are evaluated using various evaluation matrices such as MSE (Mean Square Error), RMSE (Root Mean Square Error), R2 score, etc. Our study demonstrated that Random Forest and ExtraTreesRegressor models gained substantial predictive accuracy compared to other applied regression models. ExtraTreesRegressor proves its significance with R2 score of 67.11. These results deliver significant insights for hospitals and healthcare authorities by offering resource management through anticipating fluctuating NO2concentrations. Moreover, the findings of our study highlight the potential of extending this research to include several air pollutants and meteorological factors.