Tourism forecasting: A dynamic spatiotemporal model

Abstract

In recent years, spatiotemporal modeling has become an effective method for predicting tourism demand. Nonetheless, existing forecasting models have neglected dynamic nature of spatial dependence. Furthermore, frequently used long short-term memory models often ignore spatial heterogeneity and are prone to overfitting in tourism contexts. To address these shortcomings, dynamic spatial-temporal convolutional network is proposed in this study. In this model, the spatial-temporal attention mechanism and convolution modules are employed to extract dynamic spatiotemporal dependencies and spatial heterogeneity. Based on two datasets with different time granularities, this empirical study shows that the proposed model outperforms baseline models. The results confirm that incorporating dynamic spatial dependencies and spatial heterogeneity can significantly improve predictive performance.