Page 306 - Contributed Paper Session (CPS) - Volume 7
P. 306
CPS2105 Hermansah et al.
c. Modeling with Wavelet
Decomposition with MODWT is intended to stationary the detail series so
that the results are stationary. In this case, the wavelet family, Daubechies 4, is
̃
̃
̃
̃
used with 4 levels. Detailed and smooth values obtained are D , D , D , D ,
2
3
4
1
̃
and S . Furthermore, the forecasting result of close ISSI data is obtained from
4
the sum of forecasting values of each decomposition, visually seen in Figure
4. The results of the training data forecasting obtained MSE values of 0,5743
and MAPE of 0,0032 or 0,32% and testing data were obtained MSE value is
38,6204 and MAPE is 0,0328 or 3,28%. Because the MAPE value is below 10%,
so it can be concluded that the model has a good performance.
Figure 4: The plot of the actual data and forecasting results of Wavelet
models
4. Conclusion
The forecasting accuracy of the ARIMA, Neural Network and Wavelet
models for ISSI close data can be compared with the size of MSE and MAPE.
Based on the case study of ISSI close data, the best ARIMA model forecasting
results were obtained from the results of training data with MSE values of
1,5395 and MAPE of 0,0051 or 0,51%. Whereas the best ARIMA from the
results of data testing obtained the MSE value of 4,8462 and MAPE of 0,0112
or 1,12%. The best Neural Network model were obtained from the results of
training data with MSE values of 1,2639 and MAPE of 0,0046 or 0,46%.
Whereas the best Neural Network from the results of data testing obtained
the MSE value of 2,33735 and MAPE of 0,0075 or 0,75%. The Wavelet model
were obtained from the results of training data with MSE values of 0,5743 and
MAPE of 0,0032 or 0,32%. While the Wavelet model from the results of data
testing obtained MSE values of 38,6204 and MAPE of 0,0328 or 3,28%. Of the
three models, forecasting close ISSI data can be said to be the best model is
293 | I S I W S C 2 0 1 9