Table 1.
Overview of typical research works related to tourism demand forecasting.
| References | Destination | Data frequency | Models | Performance measure | Independent variables | Forecasting context |
|---|---|---|---|---|---|---|
| Sun et al. (2019) | Beijing | Monthly |
KELM, ARIMAX, ARIMA, ANN, SVR, LSSVR |
NRMSE, MAPE, DM | Tourist arrivals, Baidu index, and Google trends data | Tourist arrivals in Beijing from Mainland and overseas |
| Bi et al. (2021) | Jiuzhaigou and Mount Siguniang | Daily |
GAF/MTF/RP-CNN-LSTM, SVM,BPNN,CNN,LSTM,CNN-LSTM |
MAE, MAPE, RMSE | Tourist arrivals | Tourist arrivals at two well-known attractions in China, Jiuzhaigou, and Mount Siguniang |
| Kulshrestha et al. (2020) | Singapore | Quarterly | BBiLSTM, LSTM, SVR, RBFNN, ADLM | RMSE, MAE, MAPE, RRMSE | Tourist arrivals and economic variables | Tourist arrivals to Singapore from five major source countries, namely Australia, France, Germany, Netherlands, and New Zealand |
| Law, Li, Bayesian and Han (2019) | Macau | Monthly | LSTM-AM, Naïve, SVR, ANN, ARIMA, ARIMAX | RMSE, MAE, MAPE | Tourist arrivals, Baidu index, and Google trends data | Tourism arrivals in Macau from the global market and mainland China |
| Sun et al. (2021) | Beijing | Monthly | SN, SARIMA, SES, ARDL, SARIMAX, MLP, B-MLP, KELM, B-KELM, SAKE, B-SAKE | MAPE, NRMSE, DS, DM, PT | Tourist arrivals, economic variables and Baidu index | Tourist arrivals in Beijing from origin countries of the United States, the United Kingdom, Germany, and France |
| Silva, Hassani, Heravi, and Huang (2019) | European | Monthly | NNAR, DNNAR, ARIMA, ETS | RMSE, RRMSE, MAPE, DM, HS | Tourist arrivals | International tourism demand for tourist arrivals of ten European countries，namely, Germany, Greece, Spain, Italy, Cyprus, Netherlands, Austria, Portugal, Sweden, United Kingdom |
| Park et al. (2021) | Hong Kong | Monthly | SARIMAX, SARIMA, SNAIVE, ETS | MAE, MAPE, RMSE, RMSPE, RI | Tourist arrivals and online news data | Tourism arrivals in Macau from the US and mainland China |
| Hu et al. (2021) | Jiuzhaigou, Kulangsu and Siguniang Mountain | Daily | SN, SARIMA, ETS, SARIMAX, TBATS, k-NN and HPR | MAPE and MASE | Tourist arrivals and dummy variables for holidays | Tourist visits to three attractions in China |
| Höpken et al. (2020) | Sweden | Monthly | ARIMAX, ARIMA, ANN, ANNX | RMSE and Shapiro-Wilk test | Tourist arrivals and Google trends data | Inbound tourist arrivals to Sweden from major sending countries (Denmark, Finland, Norway, the Russian Federation, and the United Kingdom) |
| Li et al., 2019a, Li et al., 2019b | Hong Kong | Quarterly | Naïve, ES, SARIMA, STS, ADL, VAR, EC, TVP, Interval combination model | MAPE and Winkler scores | Tourist arrivals, economic variables, seasonal dummies and one-off event dummies | Hong Kong's inbound tourism demand from its eight key source markets: mainland China, Taiwan, South Korea, Japan, Macao, the Philippines, Singapore and the US |
| Bi, Liu, and Li (2020) | Jiuzhaigou and Huangshan Mountain |
Daily | Naïve, ARIMA, SVR, ANN and LSTM | MAE, RMSE and MAPE | Tourism volume data, Baidu index and weather data | Forecasting the daily tourism volume of Jiuzhaigou and Huangshan Mountain Area, two famous tourist attractions in China |
Notes: Models: Kernel extreme learning machine (KELM); Support vector regression (SVR); Autoregressive integrated moving average (ARIMA); Autoregressive integrated moving average with exogenous variables (ARIMAX); Artificial neural network (ANN); Least squares support vector regression (LSSVR); Gramian angular field (GAF); Markov transition field (MTF); Recurrence plot (RP); Convolutional neural network (CNN); Long short-term memory (LSTM); Support vector machine (SVM); Back propagation neural network (BPNN); Bidirectional long short-term memory optimized by Bayesian (BBiLSTM); Radial basis function neural network (RBFNN); Autoregressive distributed lag model (ADLM); LSTM augmented with the attention mechanism (LSTM-AM); Seasonal naïve (SN); Seasonal exponential smoothing (SES); Autoregressive distributed lag (ARDL); Multilayer perceptron (MLP); Bagging-based MLP (B-MLP); Bagging-based KELM (B-KELM); Stacked auto-encoder with KELM (SAKE); Bagging-based SAKE (B-SAKE); Neural networks auto-regression (NNAR); Denoised neural networks auto-regression (DNNAR); Hierarchical pattern recognition (HPR); Exponential smoothing state space model with Box-Cox transformation, ARMA errors, trend, and seasonal components (TBATS); k-Nearest neighbour (k-NN); Artificial neural network with exogenous variables (ANNX); Seasonal autoregressive integrated moving average (SARIMA); Seasonal autoregressive integrated moving average with exogenous variables (SARIMAX); Exponential smoothing (ETS); Structural time series (STS); Autoregressive distributed lag (ADL); Vector autoregressive (VAR); Error correction (EC); Time-varying parameter (TVP). Performance measures: Normalized root mean square error (NRMSE); Mean absolute percentage error (MAPE); Mean absolute error (MAE); Root mean square error (RMSE); The ratio of RMSE (RRMSE); Root mean square percentage error (RMSPE); Mean absolute scaled error (MASE); Diebold and Mariano statistic (DM); Pesaran and Timmermann statistic (PT); Hassani and Silva statistic (HS); Directional symmetry (DS); Relative improvement (RI). The bold font indicates the highest forecasting accuracy among the models.