Artificial neural networks in modelling seasonal tourism demand - case study of Croatia

Abstract

The purpose of this paper is to design an artificial neural network in the attempt to define the data generating process of the number of German tourist arrivals in Croatia considering the strong seasonal character of empirical data. The presence of seasonal unit roots in tourism demand determinants is analysed using the approach developed by Hylleberg, Engle, Granger and Yoo – Hegy test. The study is based on seasonality analysis and Artificial Neural Networks approach in building a model which intend to describe the behaviour of the German tourist flows to Croatia. Different neural network architectures were trained and tested, and after the modelling phase, the forecasting accuracy and model performances were analysed. Model performance and forecasting accuracy evaluation was tested using the mean absolute percentage error. Based on the augmented HEGY test procedure it can be concluded the German tourist arrivals to the Republic of Croatia have nonstationary behaviour associated with the zero frequency and seasonal frequency. Taking this into consideration, in the analysis of the phenomenon it is necessary to consider its seasonal character. Given the importance of the tourism for Croatian economic development, the research results could be useful, for both, researchers and practitioners, in the process of planning and routing the future Croatian hotel industry development and improvement of business performances.Svrha je ovog rada dizajnirati umjetnu neuronsku mrežu sa ciljem definiranja procesa generiranja podataka o broju dolazaka njemačkih turista u Hrvatsku, uzimajući u obzir sezonski karakter empirijskih podataka. Prisutnost sezonskoga jediničnog korijena u odrednicama turističke potražnje analizirana je pristupom koji su razvili Hylleberg, Engle, Granger and Yoo – Hegyijevim testom. Istraživanje je temeljeno na analizi sezonalnosti i umjetnim neuronskim mrežama u stvaranju modela koji za cilj ima opisati ponašanje njemačkih turističkih tokova u Hrvatskoj. Trenirane su i testirane različite arhitekture neuronskih mreža, te je nakon faze modeliranja ocijenjena pouzdanost i analizirane su performanse modela. Prognostička pouzdanost modela testirana je srednje apsolutnom postotnom pogreškom. Temeljem rezultata Hegyijevog testa može se zaključiti kako je serija broja dolazaka njemačkih turista nestacionarna, na razini nulte i sezonske frekvencije. Uzimajući to u obzir, u analizi navedene pojave potrebno je uvažiti njezin sezonski karakter. S obzirom na važnost turizma za ekonomski razvoj Hrvatske, ovo bi istraživanje trebalo koristiti, kako istraživačima tako i praktičarima, u procesu planiranja razvoja hrvatske turističke industrije i u poboljšanja poslovnih performansi.