Different models and different selections of architectural parameters for each model give different prediction performance. A combination of them may provide a better performance then that provided by each indivdual. in this paper, we study this issue based on two models. One is a slightly modified version of Backk-Propagation model. and the other one is a new model we propose called RPCL-CLP. Diffeent instances of the two models with different selections of architectural parameters are employed as experts, whose outputs are comined by the modified Mixture of Experts model (Xu, Jordn and Hinton, 1994) to predict the real Foreign Exchange Rate time series.
展开▼