Prediction of the concentrations of PM2.5 and NOx using machine learning-based models

Abstract

Air pollutants such as fine particulate matter (PM2.5) and nitrogen oxide (NOx) are important to study due to their harmful health effects; therefore, it is essential to accurately predict them. However, due to a lack of effective monitoring and modeling methodologies, the accurate prediction of these air pollutants is challenging. The advancement in computational methodologies and the availability of larger data sets have expanded the opportunities of modeling for application in wide-ranging air pollution studies. Previous studies have utilized a lesser amount of data to predict air quality; therefore, a machine learning-based approach can be useful by utilizing a larger dataset for accurate pollutant predictions. Hence, a pilot study was carried out by utilizing two-month datasets for May and June for the year 2018 to predict PM2.5 and NOx concentration at three locations in Jaipur city, India. A multiple linear regression (MLR) based machine learning model has been developed in this study. The input parameters were selected using the Pearson correlation coefficient method. Specifically, PM10, NOx, and benzene were selected for PM2.5 prediction, and NH3, CO, O3, and benzene were used for NOx prediction. The model results were further refined by the outlier’s removal method. The final values of R2 for PM2.5 varied from 0.59 to 0.68, and for NOx, it ranged from 0.56 to 0.81 at the three different sites of Jaipur city, representing the effectiveness of the model developed. Overall, the results of this study will be useful in measuring and managing air quality in urban cities in India.