Tehran, the capital city of Iran, is among the world’s most polluted cities. Tehran is exposed to different types of pollutants one of which is the suspended particles of PM2.5. One of the steps that should be taken to reduce hazardous effects of this pollution on the health of society is timely prediction and announcement of its increased levels. Different methods can be used for predicting PM2.5 concentration. This study used a variety of models for predicting PM2.5 concentrations, including linear, nonlinear, and hybrid models. More specifically, the models which were used consisted of multiple linear regression, multi-layer perceptron (non-linear model), and a combination of ensemble empirical mode decomposition and general regression neural network (EEMD-GRNN) and Adaptive Neuro-Fuzzy Inference System (ANFIS) (hybrid of nonlinear models). The independent variables in the current study were air quality parameters, which were measured in reference to PM2.5, PM10, SO2, NO2, CO, and O3 and meteorological data which included average atmospheric pressure (AP), average maximum temperature (Max T), average minimum temperature (Min T), daily relative humidity level of the air (RH), daily total precipitation (TP), and daily wind speed (WS) in 2016 in Tehran. The results indicated that the ANFIS model exhibited the most accurate prediction in the training phase (R2=0.99, RMSE (root mean square error)=0.4794 and MAE (mean absolute error)=0.1305) and in the testing phase (R2=0.82, RMSE=3.2979 and MAE=2.1668). As it can be concluded, in comparison with a linear model, hybrid models are of higher precision in predicting PM2.5 concentration.
