Development of Anfis Model with Optimised Inputs to Reduce the Computational Cost and Time for Ground Level Ozone Forecasting

Abstract

This study aims to develop adaptive neuro­fuzzy inference system (ANFIS) for forecasting of daily ozone (O3) concentrations in the atmosphere of a mega city. The ANFIS model predictor considers the value of seven meteorological factors (pressure, temperature, relative humidity, dew point, visibility, wind speed, and precipitation), NO2 concentration, and previous day’s ozone concentration in different combinations as the inputs to predict the 1­day advance and same day ozone concentration. Collinearity tests were conducted to eliminate the redundant input variables. A forward selection (FS) method is used for selecting the different subsets of input variables. The method reduces the computational cost and time. The performances of the models were evaluated on the basis of four statistical indices [(coefficient of determination (R2), normalized mean square error (NMSE), index of agreement (IOA), and fractional bias (FB)].