A semi-supervised collaboration-training based on genetic algorithm for unlabeled data selection

When unlabeled data is selected for updating classifier, it is easy to introduce noise or unreliable data. In this paper, a semi-supervised collaboration-training based on genetic algorithm (SCGA) is proposed. This algorithm uses optimization function of genetic algorithm to help collaboration-training algorithm to select valuable unlabeled data. Experiments on UCI datasets prove that the algorithm is useful for updating classifiers effectively and can prevent the introduction of noise.