期刊名称：International Journal of Advances in Soft Computing and Its Applications
印刷版ISSN：2074-8523
出版年度：2016
卷号：8
期号：3
出版社：International Center for Scientific Research and Studies
摘要：Forecasting exchange rate requires a model that can capture thenon-stationary and non-linearity of the exchange rate data. In thispaper, empirical mode decomposition (EMD) is combines with leastsquares support vector machine (LSSVM) model in order to forecastdaily USD/TWD exchange rate. EMD is used to decomposeexchange rate data behaviors which are non-linear and nonstationary.LSSVM has been successfully used in non-linearregression estimation problems and pattern recognition. However, itsinput number selection is not based on any theories or techniques.In this proposed model, the exchange rate is decompose first byusing EMD into several simple intrinsic mode oscillations calledintrinsic mode function (IMF) and a residual. Permutationdistribution clustering (PDC) is used to cluster the IMF and theresidual into few groups according to their similarities in order toimprove the LSSVM input. After that, LSSVM is used to forecasteach of the groups and all the forecasted value is sum up in order toobtain the final exchange rate forecasting value where the bestnumber of input for the LSSVM is determine by using partialautocorrelation function (PACF). The result shows that the modifiedEMD-LSSVM (MEMD-LSSVM) outperforms single LSSVM andhybrid model of EMD-LSSVM.
