The projection pursuit learning network for nonlinear time series modeling and forecasting

Abstract

Nonlinear time series modeling and forecasting is one of important problems of nonlinear time series analysis. In this paper, we prove that projection pursuit learning network can approximate to nonlinear autoregression at any given precision in Lk space, where k is some positive integer. The mathematical formulation, training strategy and calculative procedures are also presented. The results of application to real data show that the projection pursuit learning network outperforms the traditional methods. © Springer-Verlag Berlin Heidelberg 2005.