- 學位論文
資料前處理對於支援向量機之影響-以期貨預測為例
The Influence of Data Preprocess on Support Vector Machines - Applications to Futures Forecasting
若您是本文的作者，可授權文章由華藝線上圖書館中協助推廣。
摘要
在變化快速的金融市場上，已有多種投資商品供投資者選擇。臺灣股價指數期貨是一種低成本高報酬的商品，若能有效預測期貨的變化趨勢，可幫助投資者以有限的資金成本獲取最大的投資報酬。支援向量機(Support Vector Machines)具有處理線性與非線性問題的能力，它使用結構風險最小誤差的概念，解決類神經網路過度學習的問題。 本研究結合資料前處理方法與支援向量機預測期貨的漲跌幅度，將決策屬性分為漲跌與5類區間來探討。首先將屬性資料正規化至[0；1]的範圍內，再利用主成份分析、判別分析與粗略集合論等篩選重要的屬性，支援向量機預測期貨的漲跌幅度，並將結果與倒傳遞網路、判別分析比較。結果顯示，利用資料前處理與支援向量機有最佳的表現，可提供使用者在投資期貨時參考。
並列摘要
In the variable financial markets, many commodities have been provided for investors. Taiwan stock index futures is a commodity of using finite bankroll to earn profits. If we can forecast the movements of futures prices, it helps investors earn enormous profits. Support Vector Machines (SVM) can handle linear and nonlinear problems. It is based on structural risk minimization principle to explore the minimization of an upper bound with forecasting error. It can avoid the problems of over-fitting and improve performance. In this paper, we integrated data preprocess and SVM to forecast the price fluctuation of futures. We discussed price fluctuation and 5-classes interval of decision attribute. First, we normalized data to the range of [0；1]. Then, we used Principal Component Analysis (PCA), Discriminant Analysis (DA) and Rough Set (RS) to select important attributes and SVM to forecast volatility of Taiwan stock index futures. To evaluate the forecasting ability of SVM, we compared the performance with Backpropagation Neural Network (BP) and DA. The experiment results showed that the best performance integrated data preprocess and SVM to forecast futures can provide to investors.
參考文獻
陳正昌、陳新豐、程炳林、劉子鍵(民94)。多變量析方法-統計軟體應用(四版)，臺北市：五南。
葉怡成(民92)。類神經網路模式應用與實作(八版)，臺北市：儒林。
臺灣期貨交易所，http://www.taifex.com.tw/chinese/home.htm。
Ahn, B. S., Cho, S.S. & Kim, C. Y. (2000). The integrated methodology of rough set theory and artificial neural network for business failure prediction. Expert Systems with Applications, 18, 65–74.
Bergerson, K. & Wunsch, D. C. (1991). A Commodity Trading Model Based on a Neural Network-Expert System Hybrid. International Joint Conference on Neural Networks, I, 289-293, Seattle, 1991.