Une nouvelle méthode de classification pour des données intervalles

Mathématiques et sciences humaines. Mathematics and social sciences

Kasoro, Nathanael¹  Hardy, André¹ 

关键词: clustering;    maximum likelihood;    hypervolumes criterion;    Poisson process;    decision tree;   

DOI  :  10.4000/msh.11138

学科分类：数学（综合）

来源: College de France * Ecole des Hautes Etudes en Sciences Sociales (E H E S S)

【 摘 要 】

This paper presents a new clustering method for interval data. It is an extension of a classical clustering method to interval data. The classical procedure is based on the theory of point processes, and more particularly on the homogeneous Poisson process. The first part of the new method is a monothetic divisive procedure. The cut rule is an extension to interval data of the Hypervolumes clustering criterion. The pruning step uses two statistical likelihood ratio tests based on the homogeneous Poisson process: the Hypervolumes test and the Gap test. The output is a decision tree. The second part of the method is a merging process, that allows in particular cases to improve the classification obtained at the end of the first part of the algorithm. The method is applied to a generated data set and to a real data set. It is compared with other clustering methods available for interval data.

【 授权许可】

Unknown