股票市場的表現是國家經濟發展中最值得參考的領先指標。它不但會對一般的產業、企業有所影響,其盛衰更是牽涉著國家的整個經濟發展體系,因此要如何準確預測出股價走勢之變動一直是投資人或學者研究的目標。本研究之研究方法主要分為兩個階段:第一階段為資料的前處理,為了獲得更高的投資報酬率,首先必須選擇投資標的,根據選股原則篩選出值得投資的個股做為預測對象。;第二階段為案例式推理系統輔助類神經網路預測模型之建立,此預測模型包含兩個部份:第一部份為倒傳遞神經網路模型,將篩選出的因子輸入類神經網路中訓練,輸出值為個股的買賣點。第二部份為案例式推理系統模型,此模型是以動態時間視窗搜尋(Dynamic Time Window Search)的方式,從歷史資料當中,擷取過去幾期與近幾期趨勢波動最相近的時間區間,藉以預測出個股漲跌情況。以案例式推理系統輔助類神經網路的預測,更精確的判斷個股股價的轉折點。經由實驗結果證實,經由選股策略所篩選出的個股其投資報酬率皆大於未經選股策略篩選的個股,顯示選股的重要性及本研究制定出的選股策略確實可以篩選出能夠提昇投資報酬率的投資標的物;此外經由案例式推理的輔助判斷,確實可以減少類神經網路預測的買賣點次數,也可以排除買高賣低的預測誤差,也可以找出更佳的買點及賣點,提升更高的投資報酬。
The stock market is very important to the economic development of a country, because it will influence the general industry and economic development system of a country. Therefore, in this search we are trying to combine the neural network technique and case base reasoning technique to construct a trading system. Besides we will provide strategy of the stock selection decision. There are two steps in the prediction model: First step is neural network model, in which we use back-propagation network to train input data, in which the output data are buy-sell point; Second step is case-base reasoning model, in which we use dynamic time window search to retrieve history patten and find the most similar neighbouring solution in order to predict stock trading. The result of our experiment shows that our stock selection strategy can find investment ambition for increasing profits. Therefor case-base reasoning can help neural network model to determine the best tradind point.