Daily Tourism Demand Forecasting with the iTransformer Model

Accurate forecasting of tourist volume is crucial for the sustainable development of the tourism industry. Deep-learning methods based on multivariate data can enhance the accuracy of tourism demand forecasting, enabling tourism management departments and enterprises to make evidence-based decisions. This study adopts an inverted transformer approach with a self-attention mechanism, which can improve the extraction of correlation features from the time series of multiple variables. Taking Zhejiang Province, a major tourist destination in China, and Hangzhou, a famous tourist city in China, as case studies, this research considers historical tourist volume, search engine data, weather data, date pattern data, and seasonal data in daily tourism volume forecasting. By comparing the forecasting results with three benchmark models, including CNN, RNN, and LSTM, the inverted transformer model’s effectiveness in forecasting the daily total visitors and overnight visitors is validated. This study’s findings can be applied to forecast the regional daily tourist arrivals, enabling decision-makers in the tourism sector to make more precise forecasts and devise more dependable plans.