Using Iterative Methods to improve time series forecasts from Radial Basis Function Nueral Networks

Marietta J Tretter

Department of Business Analysis Texas A&M University College Station, TX 77843-4217


Abstract

The reports on the accuracy and practicality of time series forecasting using neural networks in general have been mixed. However, radial basis function neural networks have demonstrated forecast accuracy and speed of computation comparable or better than established forecasting methods. The purpose of this paper is to demonstrate how this performance can be enhanced using iterative methods. A previous paper proposed the use of multigrid methods to locate centers of data for improved forecasting. This paper further explores the use of iterative methods in more efficiently solving the linear systems generated by the radial basis function networks. Marietta Tretter Department Business Analysis Texas A&M University College Station, TX 77843-4217