Foreign Exchange Rates Forecasting with Multilayer Perceptrons Neural Network by Bayesian Learning

In order to avoid the over-fitting in the training of neural networks, we apply Bayesian learning to neural networks. We illustrate the advantages of Bayesian learning by concentrating on multilayer perceptrons (MLP) neural networks and Markov Chain Monte Carlo (MCMC) method for computing the integrations. We conduct the experiments on the foreign exchange rate forecasting by using the approach. The experiment results show that Bayesian learning is better at avoiding over-fitting than the traditional parameter optimization method during the training phase of neural networks.