International Journal of Computational
Intelligence Research (IJCIR)
Volume 2, Number 1 (2006)
Optimizing local modeling for times series prediction
Coadou Yves Le
University of Paris13, LIPN-CNRS, 99 Avenue Jean Baptiste Clément 93430
IBCP-CNRS, 7, Passage du Vercors 69367 Lyon Cedex 07, France
In this paper, we present a modular system of times series prediction. This system is based on three conventional methods, a self organizing map (SOM) for learning the past of the times series, an ascendant hierarchical clustering (AHC) for optimizing the number of classes formed by SOM and a set of multi layer perceptrons (MLPs) for predicting the evolution of data in the future. The number of MLPs depends on the number of classes formed by AHC. Our approach called (SAM) will be compared with both a global approach only based on MLP and modular one using SOM and MLP. We will also demonstrate that SAM method is rather more efficient than the other previous methods.
SOM, MLP, prediction, times series, local modeling.