Artificial neural networks-based econometric models for tourism demand forecasting

Purpose - Tourism is a growing sector, playing an important role in many economies, always looking for methods to provide tourism demand forecasting and new creative ideas to develop local tourist offer. Early prediction on the tourist inflow represents a challenge helping local economy to optimize and develop tourist income. Forecasting models for international tourism demand have usually mainly been focused on factors affecting the tourist inflow, following an approach that is time-consuming and expensive in developing econometric models. Design - We modelled a backpropagation Artificial Neural Network (a Machine Learning Method for Decision Support and Pattern Discovery) to forecast tourists arrivals in Croatia and compared the results with those obtained with the linear regression methods. Methodology - The accuracy of the neural network has been measured by the Mean Squared Error (MSE) and compared to MSE and R-2 obtained with the linear regression. Approach - Our approach consists in combining ideas from Tourism Economics and Information Technology, in particular Machine Learning, with the aim of presenting creative applications of algorithms, such as the Artificial Neural Networks (ANN), to the tourism sector. Findings - The results showed that using the neural network model to predict tourists arrivals outperforms linear regression techniques. Originality of the research -The idea to use ANN as a Decision Making tool to improve tourist services in a proactive way or in case of unexpected events is innovative. Moreover, in our final consideration, we will also present other possible creative improvements of the method.