A novel approach for the prediction and analysis of daily concentrations of particulate matter using machine learning

Abstract

Traditional air quality analysis and prediction methods depend on the statistical and numerical analyses of historical air quality data with more information related to a specific region; therefore, the results are unsatisfactory. In particular, fine particulate matter (PM2.5, PM10) in the atmosphere is a major concern for human health. The modelling (analysis and prediction) of particulate matter concentrations remains unsatisfactory owing to the rapid increase in urbanization and industrialization. In the present study, we reconstructed a prediction model for both PM2.5 and PM10 with varying meteorological conditions (windspeed, temperature, precipitation, specific humidity, and air pressure) in a specific region. In this study, a prediction model was developed for the two observation stations in the study region. The analysis of particulate matter shows that seasonal variation is a primary factor that highly influences air pollutant concentrations in urban regions. Based on historical data, the maximum number of days (92 days in 2019) during the winter season exceeded the maximum permissible level of particulate matter (PM2.5 = 15 μg/m3) concentration in air. The prediction results showed better performance of the Gaussian process regression model, with comparatively larger R2 values and smaller errors than the other models. Based on the analysis and prediction, these novel methods may enhance the accuracy of particulate matter prediction and influence policy- and decision-makers among pollution control authorities to protect air quality.