This dissertation introduced a new clustering method, which deals with symbolic data. The method is hierarchical divisive monothetic, and is based on the Homogeneous Poisson Process. It builds a clustering tree by selecting, at each step, a variable to cut a group in two subsets. Its criterion is deduced from the criterion of maximum likelihood. It cuts in the middle of the biggest gap between the data. Then a phase based on the Gap Test named pruning has to simplify the tree. Examples (artificial with two variables, then real) show how the new method works, and compares it with other existing clustering methods.
|Date of Award||2006|
|Supervisor||Andre Hardy (Supervisor), MARCEL REMON (Jury) & Jean Paul Rasson (Jury)|