PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Kernel-based outlier preserving clustering with representativity coefficients
Marie-Jeanne Lesot
In: Modern Information Processing: From Theory to Applications (2006) Elsevier , pp. 183-194.


Kernel learning methods provide a framework to implicitly enrich data representation and to handle non-vectorial data without requiring to adapt to each data representation or nature. In this paper, we consider the kernel extension of the Outlier Preserving Clustering Algorithm in order to identify both major trends and atypical behaviours in datasets and to define exceptionality coefficients to measure the subgroups' representativeness, independently of the data nature. We illustrate its principles on an artificial two-dimensional dataset and on xml data representing student results to several exams.

EPrint Type:Book Section
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Theory & Algorithms
ID Code:2923
Deposited By:Marie-Jeanne Lesot
Deposited On:23 November 2006