Research on Holiday Passenger Flow Prediction for Urban Rail Transit in Tourist Cities under Limited Sample Conditions: A Case Study of China

Abstract

Holiday passenger flow in urban rail transit systems is characterized by considerable fluctuations and unpredictability, posing significant challenges for operational management. In tourist cities, these challenges are compounded by an influx of visitors during holidays, leading to pronounced passenger flow peaks and making accurate prediction especially difficult. Traditional forecasting methods often fail to effectively capture the spatiotemporal features of holiday passenger flow, particularly when only limited holiday data samples are available, resulting in reduced prediction accuracy. To tackle these issues, this paper proposes an XGBoost-based transfer learning model for time-series prediction (XGB-TL-TSP), which utilizes regular Saturday data to train for holiday scenarios in China. Experiment results show that the XGB-TL-TSP model notably improves holiday passenger flow prediction accuracy while demonstrating robust performance and strong generalization in noisy data environments and across diverse holiday contexts. The proposed model offers effective technical support for holiday passenger flow prediction in tourist cities.