The Research And The Application Of The Fuzzy Support Vector Machines

Based on a solid theoretical foundation of statistical learning theory, it developed a new universal learning method - Support Vector Machine(SVM). Using VC theory and structure risk minimization principle (SRM) as the theoretical foundation, SVM have strong generalization ability in solving some practical problems such as limited samples, and effectively overcome the traditional machine learning in relation to the local minimum and the curse of dimensionality and so on. At the same time, support vector machine can be based on limited sample information in the model complexity (ie,Learning accuracy) and learning ability (ie, error-free capacity to identify any samples) in order to obtain the best generalization ability.Support vector machine as a new technology is not yet mature, but also has shortcomings and limitations. Through improved and perfected, it can enhance the applicability, of which there are several problems to be solved:First, support vector machine was originally made for the two-class classification, how extended to multi-class classification become a research hotspot. Currently, the construction of multi-class classifier includes both direct construction and indirect method of construction, indirect construction method requires more than two classifier construction, and direct construction method to construct only one classifier, but the classification accuracy is relatively lower. How to improve the classification accuracy of direct construction method and effective multi-class classification is one of hot issues to be studied.Second, SVM is very sensitive isolated pointsand noise dada, the number of samples of each class may not be balanced. How to overcome the impact of noise on the training process and the classification results of the imbalance samples, further research is needed to improve and enhance the problem.Third, the choice of kernel function and parameters of the optimization has been also a hot issue. In the limited samples study, because of the lack of prior information, the search intervals are often large, in the case of multiple parameters to be optimized, the training process takes a lot of time.In this paper, support vector machine based on the fuzzy theory. Support vector machines can not reflected the importance of each sample ,the fuzzy support vector machine theory, on the basis of the existing research,proposed to removal the edge of the data, to obtain a better classification. The algorithm is also applied to human data and UCI data, and used in the comparative experiments with standard support vector machines.