A Study on Ultrafine Dust Prediction Model Estimation Using ARIMA Model and Multiplicative SARIMA Model

Abstract

Ultrafine dust data has seasonality. Therefore, in this study, the ARIMA model and the multiplicative SARIMA model, which are time series modeling methods, were estimated in consideration of seasonality, and the accuracy of the estimated prediction model was proposed using the MAPE measure to propose an ultrafine dust prediction model. For the ultrafine dust data used for the estimation of the ARIMA model, a seasonally adjusted estimate using the decomposition method was used assuming a multiplicative model, and the original ultrafine dust data was used for the multiplicative SARIMA model. The estimated prediction models were the ARIMA(0,1,4) model and the multiplicative   model. Residual analysis to validate the estimated ARIMA(0,1,4) model showed that the histogram of the Portmanteau statistic p-value was found to be significant. The predicted result increased in January and March, and no increase was observed from April to December. For the multiplicative  model, the significance probability of the chi-square statistic was significant at all lags. The prediction result showed that it increased in January and February, decreased continuously from March, and increased again in November. The prediction accuracy of the ARIMA (0,1,4) model was about 82.1%, and the multiplicative  model about 89.5%. The multiplicative   model was found to be about 7.4% better than ARIMA (0,1,4) in terms of the accuracy of the prediction model.