Forecasting Air Pollution Concentrations in Iran, Using a Hybrid Model

Abstract

The present study aims at developing a forecasting model to predict the next year's air pollution concentrations in the atmosphere of Iran. In this regard, it proposes the use of ARIMA, SVR, and TSVR, as well as hybrid ARIMA-SVR and ARIMA-TSVR models, which combined the autoregressive part of the autoregressive integrated moving average (ARIMA) model with the support vector regression technique (ARIMA-SVR). The main concept of generating a hybrid model is to combine different forecasting techniques so as to reduce the time-series forecasting errors. The data used in this study are annual CO2, CO, NOx, SO2, SO3, and SPM concentrations in Iran. According to the results, the ARIMA-TSVR Model is preferable over the other models, having the lowest error value among them which account for 0.0000076, 0.0000065, and 0.0001 for CO2; 0.0000043, 0.0000012, and 0.000022 for NOx; 0.00032, 0.00028., and 0.0012 for SO2; 0.000021, 0.000014, and 0.00038 for CO; 0.0000088, 0.0000005, and 0.00019 for SPM; and 0.000021, 0.000019, and 0.0044 for SO3. Furthermore, the accuracy of all models are checked in case of all pollutants, through RMSE, MAE, and MAPE value, with the results showing that the hybrid ARIMA-TSVR model has also been the best. Generally, results confirm that ARIMA-TSVR can be used satisfactorily to forecast air pollution concentration. Hence, the ARIMA-TSVR model could be employed as a new reliable and accurate data intelligent approach for the next 35 years' forecasting.