Abstract:
In recent years, artificial neural networks have received increasing attention as a decision making tool when prediction of financial time series is concerned. Modeling issues associated with artificial neural network model like the size of sample data and the general architecture of the model affect the performance of the model. For this reason, artificial neural networks outputs are prone to over-fitting or under-learning resulting to large mean squared errors which affect the accuracy of the prediction. In this paper, we investigate if the application Kalman filter algorithm to artificial neural networks model output can improve the model accuracy through the reduction of the mean squared error. Performance measures for prediction accuracy were used to compare the two models over the datasets for dollar, Euro and Pound exchange rates in Kenya Shilling for a period of five years. In the entire cases artificial neural networks model performed better than artificial neural networks with Kalman filter model.
