Collaborative forecasting of tourism demand for multiple tourist attractions with spatial dependence: A combined deep learning model

Abstract

To forecast the tourism demand across a set of tourist attractions with spatial dependence, a new model is proposed, which has three stages: tourist attraction selection, base predictor generation, and base predictor combination. In stage 1, a method for selecting associated attractions based on multi-dimensional scaling is used to determine the strength of the spatial dependence between each pair of attractions. In stage 2, a hybrid base predictor based on LSTM networks and Autoregressive model is developed, where the LSTM networks are used to capture the spatial dependence among attractions, and the Autoregressive model is used capture the scale of tourist volume at each attraction. In stage 3, a strategy for combining these base predictors is proposed; it can alleviate the overfitting problem of LSTM and improve the stability of forecasts. Finally, the superiority of the model is verified through the data on tourist volumes at 77 attractions in Beijing.