Neural Network Modeling for Forecasting Tourism Demand in Stopića Cave: Balancing Visitor Management and Geoconservation

We predict the number of tourist visits to Stopića Cave using three different approaches: the Auto-Regressive Integrated Moving Average (ARIMA) model, Support Vector Regression (SVR), and the hybrid NeuralProphet method. ARIMA and SVR are classical statistical and machine learning techniques, respectively, while NeuralProphet combines elements of both - incorporating seasonality, trend decomposition, and neural network structures. Forecasting performance across all methods is evaluated on the available dataset, with NeuralProphet outperforming the other models. This hybrid approach enables more effective modeling of non-linear patterns and temporal variability, resulting in greater predictive accuracy. Additionally, Google Trends data is integrated as an external variable to improve model precision. In practical terms, the results can provide essential information for decision-making and sustainable tourism management, offering valuable insights into visitor dynamics at Stopića Cave and a data-driven foundation for addressing issues such as carrying capacity in Serbia’s most visited cave. Accurately forecasting tourism demand is essential for managing visitor pressure at natural attractions, particularly sites experiencing pronounced seasonal fluctuations such as Stopića Cave. Reliable predictions support planning, resource allocation, and sustainable management efforts.