A Self-learned Advanced Booking Model to Forecast Railway Arrivals

Tsung-Hsien Tsai National Quemoy University Kinmen, Taiwan Abstract This paper utilizes data collected during the booking period and presents a novel advanced booking model to predict railway arrivals. As the date is approaching the departure day, more and more booking information is cumulated in the railway database. We extract useful patterns based on temporal features and formulate a two-stage model. The first stage evaluates the similarity among booking curves which describe the variation of passenger behavior. Then a group of samples which are similar to the booking patterns of the forecast target are chosen to project the prediction value based on a weighted scheme. Since the proposed model has several parameters for capturing the distinctive influence of temporal features, a direct search method is incorporated to determine the values of the parameters in the second stage. In addition, two common advanced booking models termed pick-up model and regression are also constructed in this study for comparative purposes. The results show that the proposed model may achieve at least 18% improvement in terms of predictive accuracy comparing with the benchmarks. However, more evidences are needed to validate the capability of the proposed concept in the future.