Hypergraph based semi-supervised support vector machine for binary and multi-category classifications

Currently, most graph regularization algorithms including LapSVM and LapPPSVM utilize unlabeled data for semi-supervised learning by introducing manifold regularization term. However, the graph construction in the manifold regularization term ignores the higher-order relation between data and cannot accurately express the multivariate and complex relation between data. In this paper, we propose a semi-supervised support vector machine algorithm based on hypergraph (HGSVM) for semi-supervised classification. Hypergraph is used to replace simple graph to fully explore the inherent manifold structure between labeled data and unlabeled data, and the hypergraph Laplacian matrix is calculated to form the manifold regularization term, which is embedded in the semi-supervised SVM model. Furthermore, a multi-category semi-supervised algorithm terms as KSRU-HGSVM is proposed, which introduces OVR strategy, oversampling and undersampling techniques into HGSVM. Experiments validate the effectiveness of the proposed semi-supervised classification algorithms in binary classification and multi-category classification.

Automated summary

A semi-supervised support vector machine algorithm based on hypergraph was proposed in this study to effectively utilize unlabeled data for semi-supervised learning, introducing manifold regularization term to capture the complex relations between data. Experiments demonstrated the effectiveness of the algorithm in binary and multi-category classification tasks.