A NEW METHOD OF DETECTING CLUSTERING IN THE DATA

The present study describes a new method of detecting clustering in a data set. For each object in the data, the distances to all other obje cts are calculated, sorted in ascending order, normalized and plotted as so called distance curves. n curves are obtained for data containin g n objects. The shape of these curves, together with their distributi on, give information on clustering of the data and possible distributi on. It is also possible to evaluate the populations of the clusters. T he method, however, fails for very close clusters and for elongated cl usters whose separation distance is much smaller than the range of the ir greatest variability. The method is explained using simulated and r eal data. (C) 1998 Elsevier Science B.V. All rights reserved.