以非線性獨立成份分析、支援向量迴歸及類神經網路建構財務時間 序列預測模型

Financial Time Series Forecasting using Nonlinear Independent Component Analysis, Support Vector Regression and Neural Network

楊榮力(Jung-Li Yang)

指導教授 ： 邱志洲

國立臺北科技大學/管理學院/商業自動化與管理研究所/碩士(2009年)

摘要

影響時間序列的因素有很多，由於這些因素間具有非線性關係且存在高雜訊，直接觀察常無法得到有用的資訊，所以若能利用特徵萃取工具(feature extraction methods)，將股價指數資料轉換到特徵空間(feature space)藉以消弭上述特性的影響，將有助於後續的分析與研究。本研究應用非線性獨立成份分析(Nonlinear independent component analysis, NLICA)技術於時間序列之特徵萃取，其假設混合訊號為原始訊號以非線性的方式組合，在沒有任何有關訊號混合機制的事前資訊下，將觀測到時間序列資料轉換到由獨立成份(independent components, ICs)所組成的特徵空間。由於這些獨立成份來自高階、非線性的轉換且彼此獨立，因而可以獲得較多的隱含資訊。本研究以財務時間序列資料為實驗樣本，將NLICA分別與類神經網路及支援向量迴歸結合，建構財務時間序列預測模型，由於NLICA可以有效的從觀測的資料中萃取出代表股價主要趨勢特徵的獨立成份，因此所提之整合NLICA的預測模型之預測績效優於使用LICA、PCA做為前處理之預測模型，也優於沒有使用NLICA當前處理的預測模型。

關鍵字

非線性獨立成份分析 ； 特徵萃取 ； 類神經網路 ； 支援向量迴歸

並列摘要

There are a lot of factors in time series, because have non-linear relation among these factors, it is often unable to receive useful information to observe directly. It is actually a difficult task because of the many correlated factors that become involved. Nonlinear Independent component analysis (NLICA) is a novel feature extraction technique. It aims at recovering independent sources from their mixtures, without knowing the mixing procedure or any specific knowledge of the sources. In this research a time series prediction model integrating NLICA and BPN/SVR is proposed for stock price. The proposed approach first uses NLICA on the input space composed of original forecasting variables into the feature space consisting of independent components representing hidden information of the original data. The hidden information of the original data could be discovered in these ICs. The ICs are then used as the input variables of the BPN and SVR for building the prediction model. In order to evaluate the performance of the proposed approach, the time series used as the illustrative example.

並列關鍵字

Nonlinear independent component analysis ； Feature extraction ； Neural network ； Support vector regression

參考文獻

[30] Hsu C. W., Chang C. C. and Lin C. J., “A practical guide to support vector classification,” 2003.

[40] Kao Y. H., Guo W. Y., Wu Y. T., Liu K. C., Chai W. Y., Lin C. Y., Hwang Y. S., Liou A. K., Cheng H. C., Yeh T. C., Hsieh J.C. and Teng M. H., “Hemodynamic segmentation of MR brain perfusion images using independent component,bayesian estimation and thresholding,” Magnetic Resonance in Medicine, vol. 49, no.5, 2003,pp. 885-894.

[1] Almeida L. B., “Linear and nonlinear ICA based on mutual information,” Adaptive Systems for Signal Processing, Communications, and Control Symposium, pp. 117-122

[2] Almeida L. B., “MISEP-linear and nonlinear ICA based on mutual information,” Journal of Machine Learning Reaching, vol. 4, 2003, pp. 1297-1318.

[4] Almeida L. B. , http://neural.inesc.pt/~lba/, 2001.