Design and implementation of the SARIMA–SVM time series analysis algorithm for the improvement of atmospheric environment forecast accuracy

With the recent increased interest in atmospheric pollutants in South Korea, studies on the analysis and forecast of atmospheric pollution using Internet-of-Things technology have been actively conducted. To forecast atmospheric pollution, a multiple regression analysis technique based on statistical techniques, data mining, and an analysis technique combining time series models have typically been used. In terms of accuracy, however, multiple regression analysis is insufficient for analyzing atmospheric environment data in South Korea. In addition, although the time series analysis technique is appropriate for analyzing linear data, it is inappropriate for analyzing atmospheric environment data in South Korea, where linear and nonlinear data are mixed. Therefore, this study proposes a seasonal auto regressive integrated moving average–support vector machine (SARIMA–SVM) time series analysis algorithm, combining time series analysis and nonlinear analysis, for data analysis of atmospheric environment information and improvement of pollution forecast accuracy. The proposed algorithm analyzes the seasonality in environmental contamination by using the SARIMA model, and succeeds in improving accuracy in the contamination forecast through an analysis of linear and nonlinear characteristics by applying an SVM nonlinear regression model. A comparative assessment with the existing atmospheric contamination forecast algorithm was conducted as well. The assessment results show that the forecast accuracy of the proposed algorithm improved by 20.81% for fine dust, and by 43.77% for ozone, compared to the performance of the existing models.