Stock Prediction by Combining Multiple Feature Selection Methods

• Main Authors: Yu-Chieh Hsiao, 蕭鈺潔
• Other Authors: Chih-Fong Tsai
• Format: Others
• Language: en_US
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/00304036090036697532

碩士 === 國立中正大學 === 會計與資訊科技研究所 === 97 === Stock investment has become a popular investment activity in Taiwan. To effectively predict stock price for investors, it is a very important research problem and challenging. In literature, data mining techniques have been applied to stock price prediction. As feature selection is an important pre-processing step to select more representative variables for effective prediction, previous studies do not take all relevant variables into consideration seriously. In addition, they do not attempt to further combine multiple feature selection methods to filtering out more redundant variables. Therefore, the thesis takes fundamental indexes and macroeconomic indexes into consideration based on the TEJ dataset. Artificial Neural Network (ANN) is applied as the base-line prediction model. In particular, three well-known feature selection methods, which are principal component analysis (PCA), genetic algorithms (GA) and Decision Tree (DT), are used to filter out the redundant variables individually. Regarding the experimental result in this thesis, the combination of PCA and GA and the multi-intersection combination approach provide the better prediction performances. No matter the number of variables, the accuracy and the error rate for predicting stocks rise, these two models perform the best.