Clusterwise PLS regression on a stochastic process

Cristian Preda (1) , Gilbert Saporta (2)

1 Faculté de Médecine Henri Warembourg - Université de Lille
2 CEDRIC - MSDMA - CEDRIC. Méthodes statistiques de data-mining et apprentissage

In this paper we propose to use the PLS approach for clusterwise linear regression in the particular case where the set of predictor variables forms a L2-continuous stochastic process . We have adapted the k-means algorithm to this case and we give necessar conditions for its convergence.The results of an application of the clusterwise PLS regression to stock-exchange data are compared with those obtained by other methods.

Mots clés: Clusterwise regression; PLS regression; stochastic process

Domaines: Statistiques [math.ST]

Cristian Preda, Gilbert Saporta. Clusterwise PLS regression on a stochastic process. COMPSTAT 2002. 15th Conference on Computational Statistics, Aug 2002, Berlin, Germany. ⟨hal-01124745⟩