[Paper Review] Potential Passenger Flow Prediction: A Novel Study for Urban Transportation Development

Yongshun Gong, Zhibin Li+3 more|arXiv (Cornell University)|Dec 7, 2019

Traffic Prediction and Management Techniques22 references4 citations

TL;DR

This paper proposes a multi-view localized correlation learning model (MLC-PPF) for predicting potential passenger flows (PPF) in urban transportation networks, leveraging spatial proximity, regional attributes, and multi-view data to forecast passenger movements to unconstructed metro stations. The method achieves state-of-the-art performance on real-world datasets and effectively addresses the cold-start problem in recommender systems.

ABSTRACT

Recently, practical applications for passenger flow prediction have brought many benefits to urban transportation development. With the development of urbanization, a real-world demand from transportation managers is to construct a new metro station in one city area that never planned before. Authorities are interested in the picture of the future volume of commuters before constructing a new station, and estimate how would it affect other areas. In this paper, this specific problem is termed as potential passenger flow (PPF) prediction, which is a novel and important study connected with urban computing and intelligent transportation systems. For example, an accurate PPF predictor can provide invaluable knowledge to designers, such as the advice of station scales and influences on other areas, etc. To address this problem, we propose a multi-view localized correlation learning method. The core idea of our strategy is to learn the passenger flow correlations between the target areas and their localized areas with adaptive-weight. To improve the prediction accuracy, other domain knowledge is involved via a multi-view learning process. We conduct intensive experiments to evaluate the effectiveness of our method with real-world official transportation datasets. The results demonstrate that our method can achieve excellent performance compared with other available baselines. Besides, our method can provide an effective solution to the cold-start problem in the recommender system as well, which proved by its outperformed experimental results.

Motivation & Objective

• Address the novel challenge of predicting potential passenger flows (PPF) for new metro station planning in urban areas before construction.
• Overcome the limitations of existing passenger flow models that focus only on known stations and ignore inter-area mobility patterns.
• Develop a method robust to the cold-start problem, analogous to predicting preferences for new users or new stations with no historical data.
• Integrate spatial, temporal, and domain-specific features (e.g., land use, population) to improve prediction accuracy in spatio-temporal passenger flow forecasting.
• Demonstrate transferability of the method to the cold-start problem in recommender systems, showing strong generalization capability.

Proposed method

• Construct a k-nearest neighbor-based similarity matrix using geographical proximity and regional properties (e.g., residential, commercial) to define localized spatial correlation.
• Apply a weighted correlation learning strategy that adaptively learns passenger flow relationships between target areas and their k-nearest neighbors.
• Incorporate multi-view side information from urban statistical data (e.g., population, employment, land use) to guide the learning process and enhance model generalization.
• Use a regularization term (λ) to balance the influence of auxiliary data (ABS data) and the learned correlation structure, improving robustness.
• Formulate the prediction task as a multi-view learning problem where features from different domains (spatial, temporal, demographic) are jointly optimized.
• Leverage collective matrix factorization principles in the cold-start transfer experiment to validate the model’s adaptability beyond transportation contexts.

Experimental results

Research questions

• RQ1 Can a model accurately predict passenger flows to new, unconstructed metro station areas using only historical flows from existing stations?
• RQ2 How can spatial and temporal correlations be effectively modeled to improve PPF prediction in urban transit networks?
• RQ3 To what extent can the proposed method mitigate the cold-start problem in transportation planning and recommender systems?
• RQ4 How do different hyperparameters (e.g., k, λ) affect the prediction performance and model stability?
• RQ5 Can the model generalize to other domains, such as recommender systems, by transferring knowledge from spatial correlation learning?

Key findings

• The MLC-PPF model outperforms all baseline methods in PPF prediction, achieving the lowest MAE on real-world metro transactional data.
• The model achieves a MAE of 0.396 on the Amazon book domain in cold-start recommendation tasks, significantly outperforming CMF (0.834), CDTF (0.755), and WITF (0.740).
• Setting k=2 yields the best performance, and the model is robust to variations in k between 1 and 5, indicating strong stability.
• The regularization factor λ = 10⁻¹ produces optimal results, with stable performance across the range [10⁻⁵, 10⁰], demonstrating parameter insensitivity.
• Case studies show that MLC-PPF accurately predicts flows from target areas like Homebush to major Sydney destinations such as Sydney-Haymarket and Parramatta-Rosehill.
• The method successfully transfers to the cold-start problem in recommender systems, proving its broader applicability beyond urban transportation planning.