Artificial Neural Network Models for Forecasting Tourist Arrivals to Barcelona

In order to reach accurate tourism demand forecasts, various forecasting methods have been proposed in the literature [1]. These approaches can be divided into two subclasses. One of them is conventional methods such as autoregressive moving average (ARIMA) or exponential smoothing. And, the other one is advanced forecasting techniques such as fuzzy time series, artificial neural networks (ANN) or hybrid approaches. The main purpose of this study is to develop some efficient forecasting models based on ANN for tourism demand of Barcelona in order to reach high accuracy level. Different ANN models are constructed by changing architectures and activation functions used in neurons of hidden layer. Three activation functions such as stepwise, logistic and hyperbolic tangent functions are utilized for neurons of hidden layer. The number of neurons in the input layer is changed from 1 to 6 and the number of neurons in the hidden layer is changed from 1 to 15. Thus, 90 architectures are examined for each activation function since one neuron is used in the output layer. In the implementation, 270 different ANN model constructed and applied to tourism demand of Barcelona. The tourist arrival to Barcelona between 2000 and 2014 has 15 yearly observations. When this time series was analyzed by ANN models, the first 12 observations were used as training data and the rest of them as test data. First of all, all models are trained and model parameters are determined by using the training set. Then, the forecasting performances of the models are evaluated by using a performance measure. Then, the best architecture that has the minimum performance measure value calculated over the test set is selected. All obtained forecasting results are presented and discussed.