Forecasting tourism demand for Lisbon’s region through a data mining approach

Tourism stakeholders such as the government, passenger transport companies, accommodation establishments, restaurants, recreational businesses, among others, rely on tourism demand indicators’ forecasts to make decisions. Most of tourism demand forecasting models are time-series and econometric based. Machine learning methods are emerging and have been proved to be quite suitable for non-linear modelling. These methods are part of an interdisciplinary field named “Data Mining” which is known by the process of knowledge discovery in databases (KDD). The core drive of this work is to enhance the available public sources of tourism forecast information and to contribute to the tourism stakeholders’ strategy in Portugal. More specifically, a multivariate model to forecast international tourism demand was developed through a Data Mining approach, which assessed models derived by Regression Trees (Random Forests), Artificial Neural Networks and, Support Vector Machines (SVM). The model development was constrained to machine learning methods, publicly available data, and minimum data assumptions. The forecasted demand variable was the nights spent at tourist accommodation establishments in Lisbon’s region, one of the country’s main foreign tourist destinations. The objectives were achieved, as the selected model (SMOReg, support vector regression) was successful in generalization capability. The accuracy of the produced forecasts provides some evidence of the reliability of the proposed model. If institutions and decision makers have information regarding the evolution of the explanatory variables used in this model, the impact on Lisbon’s tourism demand can be assessed, even in case of an emerging recession, as shown using three future plausible scenarios.