AN IMPROVED FUZZY SVM WITH DUAL MEMBERSHIP VALUES
Support vector machine (SVM) is a popular machine learning technique in various domains,because of its solid mathematical background, high generalization capability, especially when solving the classification issues with small datasets.However, the SVM algorithm is sensitive to outliers and noise points present in the datasets.On the other hand, the SVM algorithm divides the training dataset into mutually two exclusive classes in the binary classification issues absolutely,ignoring the possibility of the overlap region between the two training classes.Although the existing methods like fuzzy SVM can handle the problem of outliers and noise, they can still suffer from the problem of sample overlap.In this paper, we present a new FSVM algorithm which can be used to handle the problem of sample overlap.In this method, we cluster the training samples into two classes by spectral clustering method, and let each sample belong to the two classes according to membership values.The most important innovation lies in the proposed membership model, by which we can calculate the two membership values at the same time.Simulation results show that the proposed method can solve the problems of misclassification and over-fitting, and its classification performance is much better than the other existing methods and this advantage is more outstanding for small data sets.
Dual Membership Values Fuzzy Support Vector Machine Spectral Clustering Over-fitting
Xiaodong Song Liyan Han Xiyue Deng
School of Economics and Management, Beihang University, Beijing 100191, China
国际会议
The 11th International Conference on Industrial Management(第十一届工业管理国际会议)
日本
英文
369-374
2012-08-29（万方平台首次上网日期，不代表论文的发表时间）