Trend Prediction based on SVM with Nonlinear Feature Selection
| Autor: | YU, SONGTING, 余松庭 |
|---|---|
| Rok vydání: | 2017 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 105 Prediction of the stock market is one of the popular research issues in recent years. The stock market trend in Taiwan is highly related to the industry. There are many factors considered as the main effects of fluctuation of the stock's price such as political factors, economic factors and the monetary policies. Stockholders and investors used lots of technical indicators to estimate the flow of stock market. Moreover, many researchers have applied Support Vector Machine (SVM) to predict the fluctuation of stock trend. In this study, the nonlinear kernel-based feature selection method was applied to determine an appropriate subset (i.e., suitable indicators) with the largest nonlinear class separability. Furthermore, an automatic SVM based on the automatic parameter selection method was used to predict the stock trend (Raise or Down). The examination on the stock price of TSMC (Taiwan Semiconductor Manufacturing Company), Hon Hai Precision Industry Company Ltd and United Microelectronics Corporation (UMC) from 23 June 2008 to 4 October 2016 showed that the prediction accuracy is close to 0.885, 0.933 and 0.957 respectively based on selected technical analysis indicators. The Coh-Metrix education data set related to reading ability is also our target to analyze. When we utilize both indices such as Content word frequency and Age of acquisition, the accuracy appears to be 0.862 that is similar to the highest accuracy 0.865 based on about twenty three selected indices. Keywords: Stock market prediction, Support vector machine, Kernel method, Nonlinear feature selection method, Automatic parameter selection method |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|