New Multi-class Classification Based on Support Vector Date Description

Most of the multi-class learning methods transfer the multi-class classification problems to two-class classification problems,which not only are time-expensive but also have some region undiscriminating.A direct multi-class learning algorithm named multi-SVDD was proposed.Based on the consideration that there is within-class imbalance in large data sets and multi-class data sets,every class of the training data was firstly clustered.Some minimum bounding hyperspheres were formed by Support Vector Date Description (SVDD) according to the clustering results.A test sample is assigned to the label of hyperspheres if its distance to the sphere center is smaller than or equal to the radius.Compared with minimum enclosing hypersphere algorithm,the multi-SVDD algorithm doesn’t become worse in time and space cost,and the experiment result is better.