Spatio-temporal forecasting of tourism demand using graph Wavenet: evidence from the Hawaiian Islands

Tourism demand forecasting is essential for destination management and policy planning. This study develops a spatio-temporal deep learning framework based on Graph WaveNet to forecast daily tourist arrivals across the Hawaiian Islands. The proposed model jointly captures temporal dynamics and dynamic spatial dependencies among destinations, enabling more accurate multi-step demand prediction. Using island-level arrival data, we compare the proposed approach with classical statistical models and neural network–based benchmarks. The results show that the Graph WaveNet model consistently outperforms baseline methods across multiple forecasting horizons. These findings demonstrate the value of incorporating spatio-temporal structures into tourism demand forecasting and provide practical insights for regional tourism analytics and decision-making.