This research proposed a novel algorithm to solve small sample size problem. The small sample size problem is difficult to solve due to it can’t use statistical methods to estimate the distribution of training samples. Therefore, the conventional method which applies to the large sample problem does not apply to the small sample size problem. This research generates virtual samples to increase the number of samples, and then calculating the probability of noise in order to filter data. After filtering, using linearly independence to select support vector. The experimental results indicate that the proposed method is effective.