Forecasting of internet usage by singular spectrum analysis with trend extraction method

The Singular Spectrum Analysis (SSA) technique is applied to forecast internet usage data. In this paper, we forecast internet usage by SSA with Trend Extraction Method. Trend extraction is one of automatic grouping methods in SSA. there are three methods in this step; series, eigen and vector-factor methods. We select suitable methods in autogrouping by measuring MAPE (Mean Absolute Percentage Error) value of forecasting result. The obtained results comfirm that autogrouping based on vector-factor method approach can provide more accurate forecasting results than series and eigen methods. The results indicate that vector-factor approach has potential in selecting the value of r for signal extracting. Best of all result, series method is 1.25% with L =5, outsample =3, for eigen method is 1.25% with L=40 and outsample=3. Finally, the best MAPE value of vector-factor method is 0.58% with L=25 and outsample =3.