K線圖探勘於股票預測之研究;Mining candlestick charts for stock prediction

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/54939

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/54939

Title:	K線圖探勘於股票預測之研究;Mining candlestick charts for stock prediction
Authors:	蔡伯煜;Tsai,Po-Yu
Contributors:	資訊管理研究所
Keywords:	影像資料探勘;K 線圖影像;股價分析;stock prediction;candlestick chart;image mining
Date:	2012-07-06
Issue Date:	2012-09-11 19:13:05 (UTC+8)
Publisher:	國立中央大學
Abstract:	股價預測 (stock prediction) 對於投資者而言是一個有趣的議題之外，在學術上也是一個重要的議題。然而由於股價的變動因素過多，使得投資者難以預測股價，於是發展出「基本分析」及「技術分析」來輔助投資者進行決策。K 線圖是一種技術分析的方法，其中包含歷史交易的價格和交易量資料，因此投資者可以根據 K 線圖所顯示的波動趨勢和型態來分析股價趨勢。然而以往的 K 線圖分析法皆是根據分析師個人的經驗，因而缺乏一套客觀且自動化的方法來解讀 K 線圖，為了改善這個限制，本研究提出一個 K 線圖探勘 (Candlestick Chart Mining，CCM)，其作法是透過影像處理技術來萃取出 K 線圖影像的特徵向量，結合傳統技術分析的技術指標來預測股價漲跌趨勢，並且透過固定模式和滑動視窗模式來訓練分類器。根據實驗結果，本研究所提出的新方法，結合影像特徵及技術指標特徵的混合特徵對於提升股價預測正確率是有用的。且在本研究所做的短、中、長期股票預測，中期股票預測中的影像特徵預測效能比技術指標特徵好，而滑動視窗模式相較於固定模式而言，更適合用於中期股票預測。Stock prediction is an interesting and important issue for many investors. However, the factors that affect stock price are very complicated and difficult to analyze. Therefore, it is very hard to effectively predict stock price. In general, both fundamental analysis and technical analysis have been used for stock prediction. The analysis of candlestick chart (also called K chart) is one of the technical analysis methods since such figures usually contain lots of trading information which allow the investors to analyze the stock trend. However, previous studies of K chart analysis have all been based on the analysts' personal experiences. In the other words, an objective and automated method to interpret those figures is lacking. To solve this limitation, we propose a novel method called Candlestick Chart Mining (CCM). In CCM, the image processing technique is used to extract the image features from K charts. Particularly, the texture features are extracted as the image descriptors to combine with some technical indicators. The results demonstrate that the new method that we proposed to combine the image features with the technical indicators is useful for improving stock prediction accuracy, and in the mid-term stock prediction. Moreover, using the image feature alone can make the neural network classifier to perform better than using the technical indicators. Furthermore, the sliding windows mode for training the prediction model is more suitable than the stable training mode for the mid-term stock prediction.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	863	View/Open

社群 sharing

Loading...