|
Application of Exponential Smoothing Holt Winter andARIMA Models for Predicting Air Pollutant Concentrations |
|---|---|
| รหัสดีโอไอ | |
| Creator | 1. Arie Dipareza Syafei 2. Nurul Ramadhan 3. Joni Hermana 4. Agus Slamet 5. RachmatBoedisantoso 6. Abdu Fadli Assomadi |
| Title | Application of Exponential Smoothing Holt Winter andARIMA Models for Predicting Air Pollutant Concentrations |
| Publisher | The Thai Society of Higher Education Institutes on Environment |
| Publication Year | 2561 |
| Journal Title | EnvironmentAsia |
| Journal Vol. | 11 |
| Journal No. | 3 |
| Page no. | 251-262 |
| Keyword | ARIMA, Holt Winter, Prediction, Air Pollution |
| URL Website | http://www.tshe.org/ea/index.html |
| Website title | EnvironmentAsia |
| ISSN | 2586-8861 |
| Abstract | Two time series models, Holt Winter and Autoregressive Integrated Moving Average (ARIMA), were adapted to predict the concentrations of daily air pollutants in Surabaya, Indonesia. Two scenarios were developed to assess model performance in predicting PM10, sulfur dioxide, carbon monoxide, nitrogen dioxide, and ozone concentrations. In the first scenario, we used measured data, and, in the second scenario, we tested model performance when the data contained many missing values. We varied the percentage of missing values for three different sets of trained data and filled them withinterpolations. It was found that the Holt Winter model was best at predicting carbon monoxide, nitrogen dioxide, and ozone concentrations using measured data, whereas the ARIMA model was better at predicting PM10 and sulfur dioxide concentrations. An assessment of model performance when there were missing values shows that the Holt Winter model was not affected by the number of missing values and missing data patterns in the prediction of carbon monoxide and ozone concentrations, although it was affected in the prediction of nitrogen dioxide. On the other hand, the ARIMA model, which was used for the prediction of PM10 and sulfur dioxide concentrations, was not affected by the amount of missing data and missing data patterns. The Holt Winter model is recommended for the prediction of carbon monoxide concentrations based on the following model goodness of fit criteria for three different experimental runs with various amounts of missing data: the mean error, ME, (0.039; -0.878; -1106); root mean square error, RMSE, (0.315; 0.985; 1.175); coefficient of determination, R square, (0.516; 0.612; 0.785); and correlation (0.719; 0.782; 0.886). |
