| dc.description.abstract | The Taiwan High-Speed Railway (THSR) has been in operation since 2007, the passenger of THSRs is increasing by the day and THSRs are already an important vehicle for Taiwan residents. The objective of this study is to forecast passengers′ departure and arrivals at the main stations including Taipei, Taichung, and Kaohsiung. Firstly, the passengers′ departure and arrivals data were collected and the date was from Jan. 2007 to Dec. 2022, and the data number by month is 72. Simple Smooth, ARIMA and Genetic Expression Programming (GEP) were adopted to establish those forecasting models of passengers′ departure and arrivals at the three main stations, mean absolute percent Error (MAPE) and Determination coefficient(R2) were used to evaluate the model performance. In addition, the passengers′ departures and arrivals of the three main stations Jan. 2023 to May 2023 were tested on the accuracy of the forecasting model. The SPSS was utilized to build up the Simple Smooth, and ARIMA model, and the best feasible model of SPSS was simple smooth. Based on the results of the Simple smooth using SPSS, the R2 and MAPE of the passengers′ arrivals model of Taipei were 0.92 and 5.1%, the R2 and MAPE of the passengers′ departure model of Taipei were 0.92 and 5%, the R2 and MAPE of the passengers′ arrivals model of Taichung were 0.91 and 5.3%, the R2 and MAPE of the passengers′ departure model of Taichung were 0.91 and 4.8%, the R2 and MAPE of the passengers′ arrivals model of Kaohsiung were 0.89 and 4.9%, and the R2 and MAPE of the passengers′ departure model of Kaohsiung were 0.88 and 5.2%. Based on the results of GEP, the training, and validation R2 of all models of the main stations were above 0.7 and there is no overfitting situation. The MAPE of the passengers′ departure and Arrivals model using GEP of Taipei were 4% and 7%, The MAPE of the passengers′ departure and Arrivals model using GEP of Taichung were 10% and 7%, and The MAPE of the passengers′ departure and Arrivals model using GEP of Kaohsiung were 11%. Comparing the prediction capacity between Simple smooth and GEP from Jan. 2023 to May 2023, the MAPE of the GEP Model is 10%, the MAPE of the Simple Smooth Model is 15%, and the forecasting performance of the GEP model is better than Simple Smooth model. Based on the above, the GEP is the better feasible method to build up a passenger forecasting model of THSR and it is worthy of further study. | en_US |