Subyek : Wavelet Neural network,MODWT,koefisien skala, koefisien wavelet, time series musiman
Pengarang : Subanar
Kontributor : Suhartono
Tahun terbit : 2009
ABSTRACT The aim of this research is to study further some latest progress of nonlineartime series analysis, particularly about Wavelet Neural Networks (WNN).There are three main issues that be considered further in this research. Thefirst is some properties of scale and wavelet coefficients from MaximalOverlap Discrete Wavelet Transform (MODWT) decomposition, particularly atseasonal time series. The second is about development of model buildingprocedures of WNN based on the properties of scale and waveletcoefficients. Then, the third is empirical study about the implementation ofprocedures that have been developed and comparison study about theforecast accuracy of WNN to other model. The results show that scalecoefficients of MODWT at seasonal time series has stationary pattern andwavelet coefficients has seasonal pattern. The result~ of model buildingprocedure development yield two procedures both for seasonal (withouttrend) time series and seasonal with trend time series. In general, theseprocedures accommodate input lags of scale and wavelet coefficientsintroduced by Renaud et al. (2003) and other additional lags, particularlyseasonal lags of scale and wavelet coefficients. The result of empirical studyshows that these procedures work well for finding the best WNN model forseasonal time series forecasting. The comparison study of forecast accuracyshows that the second (with stepwise) procedure of seasonal with trend timeseries yields the best forecast compared to MAR, and WNN models by usingother procedures. It’s showed by the smallest RMSE at testing data.
Keywords: Wavelet Neural Network, MODWT, scale coefficient, waveletcoefficient, seasonaltime series.
Artikel selengkapnya klik di sini