A mixture model approach for on-line clustering
Allou Samé, Christophe Ambrosie and Gérard Govaert
In: Compstat 2004, 23-27 August 2004, Prague, Czech Republic.
This article presents an original on-line algorithm dedicated to mixture model based clustering. The proposed algorithm is a stochastic gradient ascent which maximizes the expectation of the classification likelihood. This approach requires few calculations and exhibits a quick convergence. A strategy for choosing the optimal number of classes using the Integrated Classification Likelihood (ICL) is studied using simulated data. The results of the simulations show that the proposed method provides a fast and accurate estimation of the parameters (including the number of classes) when the mixture components are relatively well separated.