Tourist Scale Prediction Based On Baidu Index And Spatial Dependency Of Scenic Spots

The prediction of tourist scale helps the government to optimize the allocation of human resources and ensure the personal safety of tourists.However,the tourist scale prediction is complicated.This research uses Baidu search index to establish a spatiotemporal tourist scale prediction framework based on LightGBM.The main research contents and conclusions are as follows:(1)Analyze the time lag between Baidu Index and the tourist scale of Shanghai based on the datasets of 2018 and 2021.The result shows that there is a lag time of 1-14 days between Baidu index and daily tourists scale(in the range of 1-14 days,Baidu Index passes the Granger test as the reason for the change of tourists number,that is,the search comes first,and the actual tourist behavior lags 1-14 days);the best p value is obtained in the lag period of 3 days;there is a 2-week lag between the Baidu index and weekly tourist scale.(2)In this paper,the 2018 and 2021 daily tourist scale prediction model and the2021 weekly tourist scale prediction model are established based on LightGBM.The R~2of the three models are 92.75%,77.33%,and 82.92%,respectively.On this basis,by constructing the spatial characteristics of scenic spots,the influence of the spatial location of scenic spots is introduced into the prediction model to improve the above models.After introducing the spatial features of scenic spots,the R~2 of the three models are 93.40%,81.28%and 84.52%,respectively.The result shows that the introduction of spatial features can improve the R~2in different models.This paper also found that the spatial feature ranked among the top places of the feature importance.(3)By designing different experiments,this study analyzes the prediction accuracy,applicability,stability and generalization of the model,and draws the following conclusions:compared with ARIMA algorithm and simple machine learning model,the spatial LightGBM tourist scale prediction model can obtain better accuracy;for 4A with more complex tourist composition or scenic spots with moderate search volume,the prediction error of the model is higher than that of famous scenic spots with more foreign tourists,and also higher than that of less well-known scenic spots mainly with local tourists.The impact of the epidemic on the model accuracy is mainly reflected in the following aspects:the prediction accuracy of the model decreased when there was a large-scale epidemic or a local epidemic in Shanghai.Nevertheless,the R~2 of the model proposed in this study can still maintain above 70%.The accuracy of the daily tourist scale prediction model trained on the July-September 2018 dataset has decreased when it was applied to the 2021 dataset,but the R~2 could still reach 78.89%in July-September 2021.