The Best Choice Of Parameters Of SVM And Application

With the development of modern biology, especially after the accomplish-ment of the Human Gene Project, it is generally accepted that the research on the molecule to study the human disease; Spurred on the advance of high-throughput data collection techniques, Gene Expression Profile data on thou-sands of molecules in humans and most model species have become available. The flood of information presents exciting new opportunities for understanding cellular biology and disease.However, the Gene Expression Profile data is characterized by high di-mensionality of predictors, which is far beyond the number of samples. It determines that most traditional statistical methods can't be put directly into application. Therefore the great challenge we face is to develop new statistical model to analyze and interpret the Gene Expression Profile data effectively and efficiently.Statistical Learning Theory is a small-sample statistics by Vapnik et al. which concerns mainly the statistic principle and the properties of learning procedure when samples are limited. SLT provides us a new frame work for the general learning problem. Based on this theory, support vector machine is a new learning algorithm, which adopts Structural Risk Minimization and can solve small-sample learning problems better.In the application of SVM in solving specific problems, we should con-sider the choice of kernel function and parameters. Recently, there is a few achievements on the study and application of kernel function, yet the theory which guides the choice of kernel function is to be formed. This paper study the best parameters of SVM with RBF, it improves the expansion strategy of grid search and intends to seek best parameters in the application of rand search.At last, the data test shows:the SVM in application of the procedure mentioned in this paper can predict better and enjoys priority in time and space complexity.