An Optimized Method for Semi-supervised Support Vector Machines

In view of problem of non-convex optimization problem that semi-supervised support vector machines use margin maximization principle to classify labeled and unlabeled samples, a method EDA_S3VM was proposed which using estimation of distribution algorithm to optimize semi-supervised support vector machines. Labels of unlabeled samples are taken as optimized parameters to obtain a combinatorial optimization problem on standard support vector machines, which can be solved by estimation of distribution algorithm through learning and sampling of probability model. The experiment results of artificial and UCI datasets showed that EDA_S3VM has better classification accuracy than other methods of semi-supervised support vector machines.