An improved method of support vector machine and its applications to financial time series forecasting

Abstract

A novel method for kernel function of support vector machine is presented based on the information geometry theory. The kernel function is modified using a conformal mapping to make the kernel data-dependent so as to increase the ability of predicting high noise data of the method. Numerical simulations demonstrate the effectiveness of the method. Simulated results on the prediction of the stock price show that the improved approach possesses better forecasting precision and ability of generalization than the conventional models.

Keywords :

• support vector machine (SVM)
• data-dependent
• information geometry
• conformal mapping
• financial time series
• stock price prediction

Supported by the Key Project of National Education Ministry of China (Grant No. 02090)

Supported by the Key Project of National Education Ministry of China (Grant No. 02090)

Notes

Supported by the Key Project of National Education Ministry of China (Grant No. 02090)