| Summary: | 碩士 === 東吳大學 === 企業管理學系 === 98 === Sales prediction is an important issue for most enterprise in every industry. However, in the work of sales forecasting, the data of sales are usually affected by government’s policy, business cycle … etc. which make sales data include noises and instabilities. The noise of data will make time series model over fitting or under fitting and the instability will make it hard to construct a predicting model.
For overcome the problem as above. This study uses the approaches of Hilbert-Huang transform (HHT), back-propagation neural network (BPN) and support vector regression (SVR). First, we use “Empirical Mode Decomposition” method of HHT to transform non-stationary and non-linear times series information into several “Intrinsic Mode Functions (IMFs)”. Second, we import IMFS by using BPN and SVR method to construct predicting model in order to reduce noise and instability. Finally, we use the sales data of six industries in Taiwan to test and verify the effectiveness of our method by compare with the data which are not transformed by HHT.
Additionally, we also compare our method with Auto regressive Integrated Moving Average Models (ARIMA), Wavelet analysis and Independent Component Analysis. The results shows that our method is better than other model on predict errors.
|
|---|
