股票市場已逐漸成為台灣目前投資的主要市場之一,股票市場變動快速而且影響層面錯綜複雜,容易使得投資者因聽取小道消息而盲目投資,資料探勘技術於股市方面已經有許多學者進行研究,其中類神經網路和支援向量機都有不錯分類能力,但其產生出來之預測結果不容易解讀,使得投資者有理解上問題。因此,本研究選擇可提供預測規則之決策樹,從數個簡單財務及非財務比率即可了解模型之建構過程。 在2004年以來陸續爆發博達、太電等掏空弊案,2007年次級房貸危機爆發,故本研究認為影響股價報酬因素必須加入公司治理變數與系統風險變數來進行研究。而為了提高預測能力,本研究也分別以流形學習及VIKOR排序法結合決策樹來建構預測模型;最後研究結果發現公司治理逐漸成為股價報酬的重要因素,而在使用流行學習對變數降維時,其因財務數據多群集性質而不適用,在運用VIKOR排序法時所取得的樣本時,不僅得以更少的樣本來建構模型,並且得到更好的預測準確率。
The stock market has gradually become one of the major investment markets in Taiwan. The stock market is very complex, and changes too fast that too invest. There are many studies using data mining technology in the stock market. Using neural network and support vector machine could provide a precise classification, but the results generated by black-box interpretation is not easy to understand. Therefore, this study provides prediction rules by choose the decision tree, from several financial and non-financial ratios to understand the process of model construction. Since 2004, The burst of financial scandal of PEWC and ProComp, and the second half of 2007, the outbreak of the subprime mortgage crisis in the U.S.A. This study is to investigate whether corporate governance variables and system risk has affected in the stock return. To the improve forecasting power, we take the methodology manifold learning and VIKOR together and combine with decision tree respectively to construct prediction models. The results found that corporate governance has gradually become an important factor in stock return, and VIKOR can serves better forecasting model.