本文研究將獨立成份分析與倒傳遞類神經網路應用於預測台灣股價與漲跌。 用獨立成份分析演算法，將解釋變數分離成數個彼此獨立的獨立成份，透過扣除象徵雜訊的獨立成份後將其還原，再以倒傳遞類神經網路建模。 比較原變數與扣除雜訊還原後的變數在預測及分類上的準確性，本論文分為兩個部分，第一部分的預測變數為當日收盤價；第二部分的預測變數為股價當日的漲跌。 透過實際資料的研究結果可知在扣除特定獨立成份還原後變數的預測模型可降低預測上的誤差，及提高預測的準確性。 In this paper, we apply independent component analysis and backpropagation neural network to prediction the price of Taiwan Stock and up and down. We separated explanatory variables into the several independent components by independent component analysis and delete one independent component (noise) to recover the variables, that is new explanatory variables. We use the new explanatory variables to building the backpropagation neural network model and compare it with originalmodel.We can see the new variables can reduce error and improve the accuracy in the prediction by the real data.
