Time series prediction using minimum description length

Abstract

Artificial neuron networks typically consist of a large number of nonlinear functions (neurons) each with several linear and nonlinear parameters that are fitted to data through a computationally intensive training process. Longer training results in a closer fit to the data, but excessive training will lead to over-fitting. We use the Optimal Fitting Routine together with the MDL (Minimum Description Length) Principle to construct the optimal model in order to prevent over-fitting or under-fitting. Then it will be compared with the modified version of it and the aid of back-propagation. We apply these algorithms to several time series and satisfactory results are obtained.