A New Semi-Fuzzy Algorithm for Cluster Detection and Characterization

摘要

Researchers propose a new algorithm for detecting homogeneous clusters within sets of unlabeled objects represented by numerical data of the form X = {x_1, x_2,…, x_n}⊂R~p. By quickly exploring the available data using an inter-objects similarity measure plus an ambiguity measure of individual objects, this algorithm provides the number of clusters present in X, plus a set of optimized prototypes V = {v_1, v_2,…, v_n} ⊂Rp where each prototype characterizes one of the c detected clusters. The performance of the algorithm is illustrated by typical examples of simulation results obtained for different real test data.