In this paper is presented a new cluster-segmentation algorithm. Its distance measure, derived by using Fisher's likelihood theory, depends on the probability density function (frequency function) of the observations. The resulting measure of similarity or dissimilarity is consistent with the likelihood theory. It shows attractive features: (a) curtailment of cluster-segmentation techniques; each probability density function has its own optimal measure of similarity or dissimilarity; (b) detection of dependencies between variables; and (c) all the advantages of hierarchical divisive techniques, which makes it suitable for analysis of large transportation surveys. The use of the new algorithm is illustrated by using a large data base, the Netherlands National Travel Survey. The goal of this research is to analyze mobility (expressed in daily mileage) by constructing homogeneous population groups. This example clearly demonstrates that the methodology can satisfactorily deal with numerous observations.
Abstract