Parameter optimization of ε-support vector machine by genetic algorithm
ICNC'09 Proceedings of the 5th international conference on Natural computation
Hi-index | 0.00 |
In this paper, an introduction of traditional time series prediction model using SVM has been first given, and then followed by description of a new network training algorithm and a nonlinear regression algorithm of support vector machine which are based on classification. Compared with traditional SVM regression algorithm, CSVR algorithm is less sensitive and more robust. It is another advantage that the value of the parameters can be set according to individual situation. More importantly, this method can also escape from over-fitting. Finally, an analysis of this new method has been given to demonstrate the validity of this method.