研究顯示類神經網路(Artificial Neural Network)技術能有較高之預測準確度,前提是使用適合的參數於類神經網路模型上,合適之參數設計會有比統計方法更準確的預測結果。當採納越多的資料數據做為分析的參考時,其範圍越廣亦越可能找到較相關參數,但相對地在分析時也會造成不必要的干擾,降低結果的準確性。 本研究選定臺股指數期貨(FITX)與國民生產毛額(GNP)為研究目標,利用迴歸決策樹與線性迴歸作為過濾器(Filter)篩選出哪些關鍵變數對預測結果影響較大,再將這些關鍵變數導入倒傳遞(Back-Propagation)類神經網路技術裡,臺股指數期貨實驗結果顯示平均均方誤差(MSE)只有0.136,比單純使用類神經網路的0.186結果為佳。國民生產毛額實驗結果顯示平均均方誤差只有0.17,比單純使用類神經網路的0.30為佳。
The related researches showed Artificial Neural Network technology have higher forecast accuracy of Prediction, the premise is use suitable parameters to network model, Because of the appropriate parameters design can have better prediction accuracy compared to the statistical method thought adopting more than data indication for analysis is more easily to find the relevant parameter. On contrariety, it will involve the unnecessarily interference and conduction in the prediction accuracy reduction. This research method combines Regression Trees with Artificial Neural Network technology to Taiwan Stock Index Future prediction, using the Liner Regression and Regression Trees as filters to find the indexes that affect the FIFX. Then adapting to Neural Network Model as input. The experiential results shows the MSE of our method is only 0.13 which is better than the previous research 0.15.
為了持續優化網站功能與使用者體驗,本網站將Cookies分析技術用於網站營運、分析和個人化服務之目的。
若您繼續瀏覽本網站,即表示您同意本網站使用Cookies。