Two Kinds Of Improved Fuzzy Support Vector Machines

Fuzzy support vector machine (FSVM), whose membership function is based on class centers, can effectively overcome the problem that the traditional support vector machine (SVM) is sensitive to the noises and outliers. However, FSVM assigns smaller memberships to support vectors, which may decrease the effects of these support vectors upon the construction of classification hyperplane. To tackle the foresaid problem, a novel method to determine membership function is proposed. At the same time, the training time of FSVM is generally long which is caused by the high computational complexity for constructing its kernel function. To reduce the training time of FSVM, the training samples are handled by an effective method called the dismissing margin method. The proposed method may remove some training samples that are not support vectors and improve training speed by optimizing the number of reduced training samples. According to the novel method for determining membership function and the method of dismissing margin, a new FSVM is constructed. Experimental results show that the new FSVM can effectively enhance the training speed and classification accuracy rate. However, this new FSVM has some disadvantages in dealing with the non-equilibrium data classification. Therefore, a novel method to determine membership function is proposed, and a new FSVM is constructed. Experiments show that the new FSVM can effectively reduce the misclassification rate produced by the class with fewer samples in dealing with non-equilibrium data classification problem. Therefore, the proposed FSVM may make the misclassification rates upon two classes approximately equal.