Abstract:After analysis the disadvantage of unsupervised training of support vector data description(SVDD) and combine the advantage of optimal separation hyper-plane and SVDD, the supervision with the information of negative class and the hyper-sphere classification model with optimal Separation are proposed with one minimum hyper-sphere including positive class and one maximum hyper-sphere excluding negative class, and then the decision hyper-sphere can Separate the two hyper-spheres with max distance which improves the model’s description accuracy and generalization performance. To removal the interference of bad point, a method of double proportion control parameter is proposed which can realize soft separation. Experimental results on Banana and UCI data sets show that the proposed model has better classification performance than SVDD.