研究期間：10108~10207;The investment activities have become an integral part of our lives. Because stock investment has higher profit and is easy to sell them for cash, stock price analysis has long been regarded as an important research topic. Technical Analysis, one of the major stock prediction methodology, includes quantitative analysis studies (QAS) and graphical analysis studies (GAS), in which QAS uses technical indicators to predict market trends and stock price and to decide the best time to buy/sell stocks. On the other hand, GAS can be subdivided into candlestick charts and pattern analysis. Much related work focuses on either improving the precision rate of stock price trends, or increasing the returns of investment. However, QAS contain various technical indicators and there is no an exact answer about how many and what technical indicators are more representative for stock prediction. In addition, it is difficult to allow the investors and analysts to grasp the underlying idea of technical indicators. Moreover, GAS has no formal definition and fair evaluation methods of candlestick chart patterns, thus the efficiency and reliability is difficult to trace. In order to improve above limitations, the aim of this research project is to propose a novel approach to bring the idea of image retrieval into candlestick charts analysis, and to verify whether candlestick charts are suitable for stock price analysis. This study is expected to achieve the following objectives: (1) To show that the image retrieval techniques are applicable for candlestick charts analysis. (2) To research and develop new images features to represent the visual content of candlestick charts. (3) To construct and compare general machine learning based classifiers using candlestick charts for stock prediction. (4) To draw up some useful investment strategies.
