Feature selection with multi-view data: A survey.

Highlights • Representative fusion methods are investigated for multi-view feature selection. • Linear-weighted fusion techniques are applied to addressing the multi-view feature selection methods with adaptive weights. • Open issues and future trends on feature selection and fusion strategy are discussed. Abstract This survey aims at providing a state-of-the-art overview of feature selection and fusion strategies, which select and combine multi-view features effectively to accomplish associated tasks. The existing literatures on feature selection approaches are classified into three categories including filter method, wrapper method, and embedded method. Based on the feature selection methods mentioned above, feature-level fusion or known as low-level fusion methodology is further investigated from the perspective of the basic concept, procedure, and applications in analysis tasks as presented in the literatures. Moreover, several distinctive issues that influence the information fusion process such as the use of correlation, confidence level, synchronization, and the optimal features are also emphasized. Finally, we present the adaptive multi-view issues for further research in the area of feature selection and fusion by learning view-specific weights to each view data automatically. [ABSTRACT FROM AUTHOR]