A SVM multi-classification method based on fuzzy weighted SVDD

Aiming at sensitivity of noise on WSVDD and circumstance of can not be separated in multi-classification problem, the paper presented a multi-classification method based on fuzzy weighted support vector description algorithm. Inspired by weighed SVDD, the method assigned weight to each training sample to build super-ball, while its weight does not take into account the effect of characteristics of sample data itself and noise. Noise fuzzy kernel clustering method was used to determine membership degree of samples and fuzzy weighted SVDD model was built. The multi-classification algorithm and simple classification rules were also provided. The example proves that the method can effectively reduce the effect of noise on classifier, and it can also achieve relatively better training accuracy.