Introducing an Appropriate Model using Support Vector Machine for Predicting Carbon Monoxide Daily Concentration in Tehran Atmosphere

Backgrounds and Precise air pollutants prediction، as the first step in facing air pollution problem، could provide helpful information for authorities in order to have appropriate actions toward this challenge. Regarding the importance of carbon monoxide (CO) in Tehran atmosphere، this study aims to introduce a suitable model for predicting this pollutant. We used the air pollutants and meteorological data of Gholhak station located in the north of Tehran; these data provided 12 variables as inputs for predicting the average CO concentration of the next day. First، support vector machine (SVM) model was used for forecasting CO daily average concentration. Then، we reduced the SVM inputs to seven variables using forward selection (FS) method. Finally، the hybrid model، FS-SVM، was developed for CO daily average concentration forecasting. In the research، we used correlation coefficient to evaluate the accuracy of both SVM and FS-SVM models. Findings indicated that correlation coefficient for both models in testing step was equal (R~0. 88). It means that both models have proper accuracy for predicting CO concentration. However، it is noteworthy that FS-SVM model charged fewer amounts of computational and economical costs due to fewer inputs than SVM model. Results showed that although both models have relatively equal accuracy in predicting CO concentration، FS-SVM model is the superior model due to its less number of inputs and therefore، less computational burden.