Abstract: This paper mainly studies the following problems: if there is a training set containing only A and B classes,and a test set containing more than these two categories,how should the samples in the test set be classified? For this problem, three new category classification methods based on SVM and minimum enclosing ball method are proposed. These three new methods not only can solves the weakness of SVM that can't correctly identifying new categories, but also can obtain good effect in the real data analysis. The data set used in this paper is breast cancer molecular subtype data set. The final sample classification accuracy rate can reach more than 90%,and the classification accuracy of the new category samples can be more than 99%.

Key words: machine learning, multi-classification problem, support vector machine, MEB, SVDD