Creating Prototypes for Fast Classification in Dempster-Shafer Clustering

Abstract

We develop a classification method for incoming pieces of evidence in Dempster-Shafer theory. This methodology is based on previous work with clustering and specification of originally nonspecific evidence. This methodology is here put in order for fast classification of future incoming pieces of evidence by comparing them with prototypes representing the clusters, instead of making a full clustering of all evidence. This method has a computational complexity of O(M * N) for each new piece of evidence, where M is the maximum number of subsets and N is the number of prototypes chosen for each subset. That is, a computational complexity independent of the total number of previously arrived pieces of evidence. The parameters M and N are typically fixed and domain dependent in any application.