The appraisal of machine learning techniques for tourism demand forecasting

Abstract

Machine learning (ML) methods are being increasingly used with forecasting purposes. This study assesses the predictive performance of several ML models in a multiple-input multiple-output (MIMO) setting that allows incorporating the cross-correlations between the inputs. We compare the forecast accuracy of a Gaussian process regression (GPR) model to that of different neural network architectures in a multi-step-ahead time series prediction experiment. We find that the radial basis function (RBF) network outperforms the GPR model, especially for long-term forecast horizons. As the memory of the models increases, the forecasting performance of the GPR improves, suggesting the convenience of designing a model selection criteria in order to estimate the optimal number of lags used for concatenation.