WAVELET RADIAL BASIS FUNCTION (WRBF) UNTUK PERAMALAN TIME SERIES NON STATIONER; WAVELET RADIAL BASIS FUNCTION (WRBF) FOR PREDICTION OF NON STATIONARY TIME SERIES

SAMINGUN HANDOYO, Subanar

2010 | Disertasi | PROGRAM STUDI ILMU KOMPUTER

This reaserch was conducted to implement and apply the hybrid method known as the Wavelet Radial Basis Function (WRBF) for forcasting non-stationary time series data. There are two format input data. The input data format 9 consist of 9 elemen data and is used to forecast data number 10. The input data format 17 consist of 17 elemen data and is used to forecast data number 18. There are four types data that were examined in this study ie: the caotic time series McGlass data is data that has stationary pattern, the monthly average electricity usage data that has non stationary in varian, the traffic fatalities data is non stationary data with trend and non constan varian, and the Canadian lynx data is non stationary and non linier data. After the system was completed in development, to determine the input parameter system is done by trial and error for system WRBF9 that is by trying various different values of spread and SSE. The combination of values that has smallest SSE is chosen as input parameter. Then MSE and the number of WRBF node formed to be used as input for the system WFFNN. The result of the application of on the McGlass data indicate that the method WRBF be superior, however, the WFFNN methods also have good performance for this data. This is indicated by MSE form WFFNN which is also very small. In general WRBF9 dominating performance for application in data that were discussed in this research, but with increasing level of complexity of the data pattren ie: non stationary with a mix of trend and varian not constan or non stationary and non linear data , the WFFNN methods has performance better than WRBF methods.

Kata Kunci : N