In this chapter four combinations of input features and the feedforward, cascade forward and recurrent
architectures are compared for the task of forecast tourism time series. The input features of the ANNs
consist in the combination of the previous 12 months, the index time modeled by two nodes used to the
year and month and one input with the daily hours of sunshine (insolation duration). The index time
features associated to the previous twelve values of the time series proved its relevance in this forecast
task. The insolation variable can improved results with some architectures, namely the cascade forward
architecture. Finally, the experimented ANN models/architectures produced a mean absolute percentage
error between 4 and 6%, proving the ability of the ANN models based to forecast this time series.
Besides, the feedforward architecture behaved better considering validation and test sets, with 4.2%
percentage error in test set.
The modulation of tourism time series was used in this work for forecast purposes.
The Tourism Revenue and Total Overnights registered in the hotels of the North region of Por-
tugal were used for the experimented models. Several feed-forward Artificial Neural Networks
(ANN) models using different input features and number of hidden nodes were experimented
to forecast the Tourism time series. Empirical results indicate that the Dedicated ANN models
perform better than models with several outputs. Generally the usage of previous 12 values of
the same time series is very important to a good quality forecast. For the prediction of Tourism
Revenue the Foreign Overnights and GDP of contributing countries are relevant. This time series
was predicted with an error of 4.7% and a Pearson correlation of 0.98. The forecast of Total
Overnights had an error of 6.0% and Pearson correlation of 0.98. Domestic Overnights are more
predictable than Foreign Overnights.
