A Study And Application Of Fuzzy Support Vector Multi-Classifiers Based On GA Parameters Optimization

Fuzzy Support Vector Classifiers as a new means, by combining the fuzzy theories and Support Vector Classifiers together, is proposed to solve the problem that SVC is sensitive to noises or outliers in the training set. It was first put forward in 2001 by Takuga lnouet and Shigeo Abe, followed by some scholars continued to study and bring forward varieties of improved Fuzzy Support Vector Classifiers. The initial Fuzzy Support Vector Classifiers is designed for two-class classification problems. The ability of classification and anti-interference of Support Vector Two-Classifiers for the noise points or isolate points based on the introduction of the fuzzy membership function are enhanced, in order to enhance the veracity of classification and extendness. But in the practice, the problems of classification are usually the ones of muti-classification, which need to discuss the Fuzzy Support Multi-Classifiers based on the Fuzzy Support Two-Classifiers in order to explore more and more application fields.Aimed at the multi-class problem, the algorithm of improved Fuzzy Support Vector Machines combining with adaptive genetic algorithm is put forward.(1)To the question of the optimization of kernel parameters and error penalization parameter C , an adaptive genetic algorithm for which is bring forward. But Pc and Pm are fixed and have nothing to do with the evaluation of cluster for which the algorithm is fell into"early-maturing"phenomenon easily and the efficiency of algorithm is put down in traditional Genetic Algorithm. Therefore, a new means of optimizing parameters based on improved adaptive genetic algorithm(IAGA) is raised, which improves the diversity of the cluster and the ability of search, and holds back the algorithm.(2) To the question of the parameters optimization of the Fuzzy Support Vector Multi-Classification Machine, put one mixed kernel function,fuzzy membership function based on compact character and the 1-v-1 multi-classification method together to advance an improved Fuzzy Support Vector Multi-Classification Machine which is propitious to the extendness , exact classifying and training speed.(3)The Fuzzy Support Vector Multi-Classfiers based on GA parameters optimization is advanced by putting the genetic algorithm and improved Fuzzy Support Vector Muti-Classifiers together, which improves the veracity and the run speed of the Fuzzy Support Vector Multi-Classifiers.At last, both the improved adaptive genetic algorithm and the Fuzzy Support Vector Multi-Classfiers based on GA parameters optimization are exercised through the imitation experiment in order to show the the validity of them.