Etude comparative d'une méthode de détermination du nombre de classes pour des données quantitatives: la Gap Statistic

Abstract

The goal of this report is the study of a recent method to determine the number of clusters for a set of quantitative data: the Gap Statistic method. \\ This method will be theoretically justified, and then it will be extended to cope with data characterized by symbolic features, called "intervals". \\ The idea is to compare the changes in within cluster dispersion to these expected under an appropriate reference null distribution. \\ This report contains a lot of examples. On each of the applications, the output of the method is analysed; that method is compared with other well-known methods available in the scientific literature.

Date of Award	2007
Original language	French
Supervisor	Andre Hardy (Supervisor), Jean Paul Rasson (Jury) & Marcel Remon (Jury)