Integration of ANFIS model and forward selection method for air quality forecasting

In the last decade, air pollution in the city of Kermanshah has become a major concern. In this study, adaptive neuro-fuzzy inference system (ANFIS) was developed to predict five daily air pollutants in the atmosphere of Kermanshah city on the same day and 1day in advance from 2014 to 2016. The selected pollutants were the particulate matter PM10, sulfur dioxide (SO2), nitrogen dioxide (NO2), carbon monoxide (CO), and ozone (O-3). The temperature, relative humidity, dew point, wind speed, precipitation, pressure, visibility, and the pollutant concentration on the previous day were considered as predictors in the ANFIS model. In order to reduce the computational cost and time, the collinearity tests and forward selection (FS) technique were utilized to remove the redundant input variables and select the different combinations of input variables, respectively. Results showed that input combination for MODEL 2 (six input conditions) and MODEL 3 (five input conditions) performed well between observed and predicted values of CO in the same day forecasting (SDF) and 1day in advance forecasting (1DAF). For other pollutants such as NO2, SO2, and PM10, the results obtained from MODEL 3 were better compared to the other input subset of the MODELs in the SDF and 1DAF. Developing the ANFIS model for O-3 pollutant showed that MODEL 4 with the lowest normalized mean square error (NMSE) can be used to forecast the O-3 concentration in both cases. It can be concluded that the integration of the FS method and ANFIS model led to an improvement in air quality forecasting.