Development and performance evaluation of DE based time series prediction model

Abstract

The paper proposes a new model for efficient prediction of small and long range exchange rate forecasting. The model employs an adaptive linear combiner with its weights trained using Differential Evolution (DE). A new training scheme of model parameters is proposed using DE based optimization rules. The prediction results are obtained using LMS, GA as well as DE based method. In all cases simulated it is concluded that the DE based training model shows improved prediction performance for all exchange rates as well as for various months' ahead prediction.