Neural Nets: An Approach to the Forecasting of Time Series

Abstract

We propose to compare the performance of a parallel distributed processing system with an operational time series model to predict mean length of service (LOS) in a military personnel system After testing a simple model to gauge the suitability of neural networks for pattern recognition, various standard network types were tested, including Adaline, counterpropagation, and backpropagation. The networks are being tested to see if they can provide forecasts that are more sensitive to turning point fluctuations than standard time series models Keywords: neural nets, forecasting, time series analysis