Review of Passenger Flow Modelling Approaches Based on a Bibliometric Analysis

Jonathan Hecht, Weilian Li, Ziyue Li, Youness DehbiarXiv 2025 · DOI · arXiv

Introduction

Accurate short-term passenger flow forecasting serves as the critical "brain" for smart city transit systems, enabling operators to optimize bus frequency, manage station overcrowding, and implement effective rerouting during delays. Despite its importance, the research landscape has historically been fragmented, often limited to specific transit modes or narrow timeframes. This review provides a quantitative, "bird's-eye view" of the field, analyzing 814 publications from 1984 to 2024 to map the evolution of modeling techniques and identify emerging frontiers.

Methodology

The study employed a multi-stage bibliometric and natural language processing (NLP) workflow to treat the existing literature as a structured dataset.

The three-stage research methodology: Data retrieval, Methodology, and Results.

The methodology consisted of three primary phases:

  1. Performance Analysis & Science Mapping: Utilizing the R package bibliometrix to track publication metrics, top-tier journals, and citation networks.
  2. Thematic Modeling: Implementing BERTopic, a transformer-based technique, to cluster document abstracts into thematic areas, supplemented by Biterm Topic Modeling (BTM) for co-word analysis.
  3. Network Analysis: Developing directed co-citation networks to visualize the intersection between transit-specific research and broader machine learning and time-series fields.

Data Sources

The dataset comprised 814 eligible documents, including articles, conference papers, and book chapters. Data was harvested from Web of Science (WoS) and Scopus, followed by a rigorous cleaning process to exclude irrelevant studies (e.g., aviation or maritime transit) and duplicates.

Key Insights

The Evolution of Modeling

Research activity remained sporadic until a significant turning point in 2008. Since 2017, the field has experienced an explosion in publication volume, mirroring the broader "Deep Learning revolution." The modeling paradigm has shifted from traditional statistical methods like ARIMA and basic machine learning (SVM) toward specialized architectures such as Long Short-Term Memory (LSTM) networks for temporal dependencies and Graph Convolutional Networks (GCNs) for spatial relationships.

Geographic and Modal Biases

The field is heavily dominated by Chinese institutions, with approximately 72% of analyzed papers featuring a Chinese corresponding author. Beijing Jiaotong University stands out as a global leader. Additionally, there is a pronounced bias toward rail-based transit (subways and metros) over bus systems, likely due to the more structured and "closed" nature of rail data.

Results and Thematic Clusters

The BERTopic analysis successfully identified 11 distinct thematic clusters, illustrating that the field extends beyond simple prediction to include train scheduling, flow simulation, and advanced data pre-processing techniques like wavelet analysis.

BERTopic visualization of the 11 key thematic clusters in passenger flow research.

Influential publication channels include Transportation Research Part C: Emerging Technologies and IEEE Transactions on Intelligent Transportation Systems, which serve as the primary "heavy hitters" in terms of citation impact.

Future Directions and Limitations

While accuracy continues to improve, several critical gaps remain:

  • Explainable AI (XAI): As models transition into "black boxes," there is an urgent need for interpretability so transit dispatchers can trust and understand model predictions.
  • Foundation Models: A nascent shift toward massive, pre-trained models (e.g., Chronos or Lag-Llama) suggests a future where models are fine-tuned rather than trained from scratch.
  • Data Fusion: Future research must integrate auxiliary data—such as weather, social media, and land-use patterns—to move beyond univariate fare-card analysis.
  • Practical Deployment: Researchers are encouraged to balance marginal accuracy gains with computational efficiency and real-world deployment constraints.

Note: This study primarily focused on English-language papers in WoS and Scopus, potentially omitting significant research published in other languages or on preprint servers.

Summary updated

BibTeX
@misc{hecht2025review,
  author = {Jonathan Hecht and Weilian Li and Ziyue Li and Youness Dehbi},
  title = {{Review of Passenger Flow Modelling Approaches Based on a Bibliometric Analysis}},
  year = {2025},
  eprint = {2511.13742},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2511.13742},
}
Figure