應用資料探勘技術於股票投資-以健鼎為例

資料探勘是一種利用大量的資料，透過機器學習將資料分類、預測的方法，不同的輸入資料、不同的資料探勘演算法，都會影響機器分類、預測的結果。本研究提出使用資料探勘對股票投資的買點信號做分類，主要希望能達成二個目標，分別是(1)輸入變數的選擇以及建模，(2)買點信號的分類。本研究透過輸入變數的選擇以及建模，得到資料探勘模型對買點信號的分類績效。同時提出交易策略，整合模型分類績效與交易策略提出適合模型的交易方法，計算出投資報酬率。最後利用健鼎測試本研究提出的新交易方法是否適用，以達成本研究提出利用資料探勘建立新的股票交易方法是否有效的目標。

關鍵字

資料探勘；機器學習；股票投資；交易策略；投資報酬率；交易方法

並列摘要

With the trend of big data, more and more data mining methods are adopted to analyze the data sets and to support machine learning technologies. The ultimate goal is to classify objects/situations and/or predict future. However, the accuracy of classification/prediction is highly affected by the input data set and the data mining algorithms. In this research, we focus on Tripod Technology Corp. and build “buy-day” classification models to determine whether a trading day is a “buy-day”. A trading day is called a “buy-day” if the stock closing price of Tripod Technology Corp. can rise over 10% in the coming 70-day trading period. There are two features in our research. The first is the selection of proper input variables. The second is the construction of classification models. We first select proper input variables for modeling and then tune the parameters of classification models for higher accuracy. Then, these “buy-day” classification models are applied to predict whether a specific trading day is a “buy-day”. When a “buy-day” is identified, several accomplished trading policies are proposed to raise the confidence. Finally, historical trading data of Tripod Technology Corp. is collected and applied to verify the combined performance of our classification models and trading policies.