Seasonal time series prediction with artificial neural networks and local measures

Forecasting is one of the most challenging fields in the industrial research, due to its importance in practice and to the variability and number of elements that should be considered. In this context, the usage of Artificial Neural Networks has proved to be particularly advisable, thanks to their ability to approximate any kind of function within a desirable rangeThe paper proposes a novel approach to the time series forecasting activities through the identification and exploitation of information hidden – or latent – into the values and the structure of a seasonal time series. In order to do this, the time series is decomposed into parts, for each of which some local measures are evaluated: such measures are intended to improve the forecasting ability of the ANN. To exploit the “regularity” of a seasonal time series, the concept of Seasonal Periodic Index has been introduced.