ePrints

Short-Term Passenger Flow Prediction of Urban Rail Transit Integrating Multiple Features: A Hybrid Model with Two-Stage Feature Selection

Lookup NU author(s): Ziye Lan

Downloads

Full text for this publication is not currently held within this repository. Alternative links are provided below where available.

Abstract

© 2026 American Society of Civil Engineers. Urban rail transit systems encounter challenges in short-term passenger flow prediction due to complex spatiotemporal data and dynamic travel patterns. Traditional models often struggle to capture these complexities. This study introduces a hybrid prediction framework that employs a two-stage feature selection strategy and multiscale decomposition to enhance forecasting accuracy. Key innovations include (1) a two-stage feature selection combining correlation analysis and recursive feature elimination, enabling the identification of the most influential temporal, spatial, and external features; (2) complete ensemble empirical mode decomposition with adaptive noise for isolating multiscale patterns; (3) parallel modeling with modern temporal convolutional networks and extended long short-term memory; and (4) a cross-attention mechanism for spatiotemporal feature fusion. Tested on 22 stations of Fuzhou Rail Transit Line 2 (8 million automatic fare collection records), the proposed model significantly outperforms classical and deep learning baselines. An ablation analysis further validates the contribution of each module to the overall performance. By bridging advanced AI methodologies with practical transit management needs, this work advances scalable, data-driven decision-making for sustainable urban mobility systems.

Publication metadata

Author(s): Lai Y, Zhang J, Zhou X, Xu L, Lan Z, Easa SM, Wang S

Publication type: Article

Publication status: Published

Journal: Journal of Transportation Engineering Part A: Systems

Year: 2026

Volume: 152

Issue: 6

Print publication date: 01/06/2026

Online publication date: 27/03/2026

Acceptance date: 05/12/2025

ISSN (print): 2473-2907

ISSN (electronic): 2473-2893

Publisher: American Society of Civil Engineers (ASCE)

URL: https://doi.org/10.1061/JTEPBS.TEENG-9429

DOI: 10.1061/JTEPBS.TEENG-9429

Altmetrics

Share