Prediction of tropospheric ozone using artificial neural network (ANN) and feature selection techniques

Tropospheric ozone (O3), as an air pollutant is increasing at an alarming rate in urban areas. The concentration of ozone is affected by precursor pollutants, such as particulate matter (PM10, PM2.5), carbon monoxide (CO), nitrogen dioxide (NO2), sulfur dioxide (SO2), carbon dioxide (CO2), and nitric oxide (NO), and meteorological parameters, such as air temperature (AT), relative humidity (RH), global solar radiation (SR), wind direction (WD), and wind speed (WS) of the area. Ozone is a secondary pollutant and strong oxidizing agent injurious to human health. The present study aimed to identify the most crucial factors that influence ozone formation and to develop an ozone prediction model using artificial neural network with optimal inputs. The data obtained from Limbayat, real-time air pollutants monitoring station of Surat city, have been used to evolve the model, followed by feature selection techniques, namely, sensitivity analysis, Boruta algorithm, and recursive feature elimination algorithm (RFE). Finally, 6/14 influencing parameters have been identified using an attribute selection approach. Interestingly, “hour of the day” was found the most prominent and governing parameter among the 14 parameters after applying various feature selection techniques in the experiments. The result showed that the efficiency of the prediction model was 79.4% when six parameters were used in the machine learning algorithms.