Multi-view structural twin support vector machine with the consensus and complementarity principles and its safe screening rules

Nowadays, numerous multi-view algorithms are proposed to achieve better performance in classification tasks. In this article, we propose a multi-view structural twin support vector machine (MvSTSVM), which is guided by both complementarity and consensus principles. In order to keep the consensus information, structural information of inter-class and intra-class is explored in two hyperplanes by clustering methods, which significantly enhances the classification performance. Besides, MvSTSVM introduces the weights to multi-view datasets to dig the complementarity information among different views. In addition, our model achieves both intra-class compactness and inter-class separability. To deal with the optimization problems, an iterative strategy is adopted for the solution of MvSTSVM. Moreover, we derive the safe screening rules for MvSTSVM, which speed up the tuning process obviously without sacrificing accuracy. A series of numerical experiments confirm the effectiveness of MvSTSVM and the safety of proposed screening rules. •A novel multi-view classifier: MvSTSVM is proposed.•It accelerates the parameter selection obviously by safe sample screening.•It achieves both intra-class compactness and inter-class separability.•It keeps the consensus information in the structural regularization item.•It achieves complementarity principle through weighting ideas.