Forecasting Visitor Arrivals at Tourist Attractions: A Time Series Framework with the N-BEATS for Sustainable Tourism

Contemporary techniques built on deep learning technologies enable precise forecasting of tourism demand, particularly for the relaunch of sustainable tourism following COVID-19. We developed a novel framework to forecast visitor arrivals at tourist attractions in the post-COVID-19 period. To this end, a time-based data partitioning module was first pioneered. The N-BEATS algorithm with multi-step strategies was then imported to build a forecasting system on historical data. We imported visualization of curve fitting, metrics of error measures, wide-range forecasting horizons, different strategies for data segmentations, and the Diebold-Mariano test to verify the robustness of the proposed model. The system was empirically validated using 1604 daily visitor volumes of Jiuzhaigou from 1 January 2020 to 13 May 2024 and 1459 observations of Mount Siguniang from 1 October 2020 to 18 May 2024. The proposed model achieved an average MAPE of 39.60% and MAAPE of 0.32, lower than the five baseline models of SVR, LSTM, ARIMA, SARIMA, and TFT. The results show that the proposed model can accurately capture sudden variations or irregular changes in the observations. The findings highlight the importance of improving destination management and anticipatory planning using the latest time series approaches to achieve sustainable tourist visitation forecasts.