A temporal case retrieval model to predict railway passenger arrivals

This paper proposes a three-stage model to predict final sales when advanced booking, which is prevalent in the service industry, is available. The concept behind the proposal is that similar booking patterns during the reservation period indicate the trend of sales. Booking curves which record accumulated reservations were collected from a railway company. The first stage is to evaluate the similarity of booking patterns between the collected samples and the days to be predicted. Then samples with high similarity to the forecasting target are chosen from the collected observations. Integrating the final sales of these selected samples to project future volumes is the main job in the last stage. Regression and Pick Up models, common in practice, are also constructed for comparing purposes. The results show that the proposed model can significantly improve predictive accuracy in the testing cases.