Multi-class Fuzzy Support Vector Machine Based On Dismissing Margin

In the 1990s Support Vector Machines which was proposed by Vapnik is a new technology of data mining. It uses quadratic optimization method to solve the traditional difficulties, such as "over learning" and "dimension disaster". Support vector machines have achieved good effect whether in theory or algorithm research, and have already become a hot research. Support vector machines have been well researched and applied in two-class classification questions. But in reality there are more and more multi-class classification problems, therefore extending two-class classification problem to multi-class is a trend of research.Each time the largest interval SVM separating multi-class problem into two-class tends to divide one category, which has low efficiency. Intra-class interval matrix method has been proposed to raise efficiency. But the time-consuming and calculation quantity are still great when calculating the intervals between each two classes every time. In view of this shortcoming, this paper presents a center-class distance matrix method instead of intra-class interval matrix. For large number samples, categories fuzzy and isolated point situation, this paper extends two-class dismissing arithmetic to multi-class and proposes a new method—multi classification fuzzy support vector machine based on dismissing margin. This thought changes the sensitivity of the traditional support vector machine towards isolated point. Removing some sample points which do not possibly be support vector before training, which can reduce the corresponding optimization question's dimension, and the requirement of memory and calculation quantity, raises the efficiency finally.