Abstract: This paper proposed two constraint project based selective ensemble algorithms of support vector machines.Firstly,projective matrices were determined upon randomly selected must link and cannot-link constraint sets, with which original training samples were transformed into different representation spaces to learn a group of base classifiers.Then, two selective ensemble techniques of genetic optimization and minimizing deviation errors were utilized to combine base classifiers. Experiments on UCI datasets show that both proposed algorithms improve the generalization performance of support vector machines significantly, which are much better than classical ensemble algorithms, such as Bagging,Boosting,fcaturc Bagging and LoBag.

Key words: Constraint project, Selective ensemble, Support vector machines, Classifiers