A Fuzzy Support Vector Machine For Three Classes Classification Problem

Support Vector Machines (SVM) proposed by Vapnik and his group at Belllaboratory, which is a novel machine learning system based on statistical learning theorywhich have attracted increasing attention from Learning Machine Community becauseof their optimal applications in classification data. SVM have demonstrated their abilityin solving classification problems in an optimal way with a solid mathematicalbackground. An SVM employs a two part approach to binary classification, the first partis transformation method known as the kernel function. The kernel function transformsa given set of vectors to a higher dimensional space. The second part often called theinduction engine, attempts to find a hyperplane to separate the transformed vectors intoto regions. Traditional SVM is developed for binary classification problems. How toextend it for multiclass classification is a significant purpose.In this paper, we introduced a combined type kernel function, it is combined withthree kernel function, which are Gauss RBF kernel function, polynomial kernel functionand Sigmoid kernel function. We use this new function to transform plane problem intoa higher dimensional problem. And then a new fuzzy membership function isproposed in this paper, it is introduced into the decision function based on binaryclassification and extended to three-classes data set. The experimental results show thatthe proposed method is a better way to solve the three-classed classification problem,and five us a measure of the quality of the ultimately chosen model.