Clustering Belief Functions Using Agglomerative Algorithm

This paper proposes an approach for clustering belief functions. The approach is composed of agglomerative clustering and how to determine the cluster number. The former one is achieved by taking belief distance as dissimilarity measure between two belief functions and selecting complete-link algorithm to measure the dissimilarity between two clusters. The latter one is completed by utilizing metaconflict when there is priori information on cluster number, and by setting appropriate threshold value of dissimilarity when there is no any priori information. The advantage of the proposed approach is that there is no need to set the cluster number which is unknown in advance. Illustration results are presented to demonstrate the usability of the proposed approach.