Multi-view evidential c-means clustering with view-weight and feature-weight learning

Multi-view clustering remains a challenging task due to the potential overlap of clusters and variability across different views, which causes uncertainty and imprecision in cluster assignment. This paper explores multi-view evidential c-means clustering with view-weight and feature-weight learning (MVECM-VFL), grounded in the theory of belief functions (TBF). The aim is to effectively capture the uncertainty and imprecision in cluster assignment while concurrently assessing the contributions of view and feature weights in the clustering framework. We integrate view and feature weights and credal partition into a joint learning framework and design a new objective function to find the best results. In MVECM-VFL, each object can belong to different clusters with various masses of belief, thereby characterizing the uncertainty. Furthermore, when an object resides in the overlapping area of several singleton clusters, it can be attributed to a meta-cluster (defined as the union of these singleton clusters) to represent the local imprecision in cluster assignment. Additionally, view-weight and feature-weight learning can help to obtain the credal partition better. Compared to the related methods, the effectiveness of MVECM-VFL is demonstrated based on synthetic and real-world datasets.