Forecasting Algorithm Adaptive Automatically to Time Series Length

Abstract

The developed forecasting algorithm creates trend models based on varying length time series by eliminating its oldest member. The constructed criterion evaluates the attained models through estimating the ratio between the average of the stochastic errors for the forecasted period and the average of real values. The best model and forecasting are automatically achieved in contrast to statistical software systems SPSS, STATISTICA, etc. where this process is accomplished progressively by the user. Therefore, this forecasting algorithm is adaptive to the length of time series. Component oriented approach has been used for software implementation. Simulation experiments have been carried out to test the forecasting algorithm using the multidimensional time series database on fishery in Greece.