Accuracy assessment of multiple linear regression, (ARIMA), and (GRNN) models to prediction of particulate matter PM2.5 concentration

Existence of environmental crises and problems in the world has led the increasing of importance of the discussion about the environment and environmental issues in the past half century. In the recent decades, the air pollution as one of the environmental crises has been known as a most harmful natural disasters. Due to harmful effects of particulate matter on human health, prediction of particulate matter concentration in the coming days could be reduce these effects. Therefore, in this study the accuracy of linear models including multiple linear regression (MLR), autoregressive integrated moving average (ARIMA), and nonlinear model of General Regression Neural Network (GRNN) to prediction of particulate matter PM2.5 in Sanandaj city were tested to choose the most accurate prediction. To date, no study has been conducted to investigate the accuracy of the GRNN model to prediction of atmospheric pollution in Iran. Air quality data consisting of O3, CO, NO2, SO2, PM10, and meteorological data such as average minimum temperature (MinT), average maximum temperature (MaxT), average atmospheric pressure (AP), daily total precipitation (PR), daily relative humidity of the air (RH) and daily wind speed (WS) of 2015 as independent variable and the PM2.5 concentration as a dependent variable were considered. The results showed that the GRNN model with a R2 = 0.81, RMSE = 6.9468 and MAE = 5 in the training phase and the amount of R2 = 0.74, RMSE = 5.0725 and MAE = 3.4874 in the test phase had been best performance to predict of particulate matter PM2.5 compared to linear models in Sanandaj city.