Paper Title
Prediction of Tourist Arrivals using Neural Networks

Abstract
Tourist arrivals, as an extremely important variable for Croatian economy, requires an in-depth analysis. Namely, the share of tourism’s total contribution to GDP in Croatia amounted to almost 25% in 2019. In years to follow the number of tourist arrivals declined significantly due to COVID circumstances with a decline in the share of tourism’s total contribution to GDP as well. Its accurate forecast, nevertheless, remains vital for the tourism industry as well as government economic policy and decision making.However, due to unfulfilled assumptions of monthly tourist arrivals, i.e. non-normality and nonlinearity, using linear parametric models is not suitable and leads to inability to forecast its values properly. Moreover, it exhibits extremely seasonal patterns. In order to capture all processes and to successfully forecast the tourist arrivals in Croatia, it is necessary to choose a suitable methodology, i.e. in this case seasonal neural networks. Therefore, the goal is to examine their ability to optimally forecasttourist arrivals in Croatia. Additionally, neural networks are compared to traditionally used linear models, i.e. seasonal ARIMA models, in their forecasting ability. The research confirms neural networks as superior to its linear counterparts in forecasting monthly tourist arrivals in Croatia. Moreover, lower number of hidden units yields better predictive performances of neural networks. JEL codes: C32, C45, C53 Keywords - Seasonal ARIMA, Seasonal Neural Networks, Tourist Arrivals, Croatia, Post-Covid Period