Incorporation of statistical methods in multi-step neural network prediction models

This paper addresses the problems associated with multistep ahead prediction neural networks models. We will see how some concepts from the statistical theory field can be applied in various ways to improve the modelling. The generalization and error autocorrelation problems will he addressed using topological and methodological approach among which network committees, statistical bootstrap and principal component analysis will play a key role. These methods will be applied to the sunspot time series