Does Restraining End Effect Matter in EMD-Based Modeling Framework for Time Series Prediction? Some Experimental Evidences

摘要

Following the "decomposition-and-ensemble" principle, the empirical modedecomposition (EMD)-based modeling framework has been widely used as apromising alternative for nonlinear and nonstationary time series modeling andprediction. The end effect, which occurs during the sifting process of EMD andis apt to distort the decomposed sub-series and hurt the modeling processfollowed, however, has been ignored in previous studies. Addressing the endeffect issue, this study proposes to incorporate end condition methods intoEMD-based decomposition and ensemble modeling framework for one- and multi-stepahead time series prediction. Four well-established end condition methods,Mirror method, Coughlin's method, Slope-based method, and Rato's method, areselected, and support vector regression (SVR) is employed as the modelingtechnique. For the purpose of justification and comparison, well-known NN3competition data sets are used and four well-established prediction models areselected as benchmarks. The experimental results demonstrated that significantimprovement can be achieved by the proposed EMD-based SVR models with endcondition methods. The EMD-SBM-SVR model and EMD-Rato-SVR model, in particular,achieved the best prediction performances in terms of goodness of forecastmeasures and equality of accuracy of competing forecasts test.