Post-pandemic tourism forecasting with ensemble RNN

Tourism forecasting plays a critical role in the tourism industry, enabling strategic planning for diverse stakeholders. However, it is a challenging task influenced by numerous factors. This study investigates the development of automated, data-driven approaches, by introducing an ensemble model that combines two forecasting methods of recurrent neural networks. It integrates COVID-19-related explanatory variables and automatically learns the spatial relationship across destinations. The model outperformed benchmark methods in forecasting China's outbound tourism to twenty destinations before and during COVID-19, using data from 1989 to 2022. Subsequently, our approach achieved 1.4723 mean absolute scaled error and third runner-up for the Point Forecasting Track in Tourism Forecasting Competition amid COVID-19 Round II, for the forecast period between August 2023 and July 2024.