Transductive graph-based semi-supervised learning methods usually build anundirected graph utilizing both labeled and unlabeled samples as vertices.Those methods propagate label information of labeled samples to neighborsthrough their edges in order to get the predicted labels of unlabeled samples.Most popular semi-supervised learning approaches are sensitive to initial labeldistribution happened in imbalanced labeled datasets. The class boundary willbe severely skewed by the majority classes in an imbalanced classification. Inthis paper, we proposed a simple and effective approach to alleviate theunfavorable influence of imbalance problem by iteratively selecting a fewunlabeled samples and adding them into the minority classes to form a balancedlabeled dataset for the learning methods afterwards. The experiments on UCIdatasets and MNIST handwritten digits dataset showed that the proposed approachoutperforms other existing state-of-art methods.
展开▼
