Time series forecasting: An approach to modeling seasonality using neural networks

Artificial neural networks have been widely advocated as an alternative to statistical methods to forecast time series. Whether or not neural networks are able to model the seasonal component of a time series has been a matter of controversy. While some earlier studies suggest that neural networks can learn the seasonal component well, a couple of recent studies suggest that it is better to deseasonalise the data first. In this paper we present a neural network approach that reduces the forecasting errors of a seasonal series even without prior deseasonalisation of the data.