Short-term lagged interactions between freight and passenger volumes in urban traffic - inter- and intra-modal effects with explainable machine learning

Aug 6, 2026

Upload by Aria

Urban transport systems face increasing complexity as freight and passenger flows compete for limited road capacity. While multimodal forecasting methods have progressed, short-term interactions between vehicle classes remain underexplored, especially in operational settings. This study tests whether recent freight or passenger volumes are associated with current traffic conditions across modes using 6,003 hourly records from Liverpool. An interpretable framework combining K-means clustering, XGBoost, and DALEX shows that one-hour lagged freight volume improves classification of passenger traffic states. Feature importance and partial dependence identify nonlinear effects beyond about 500 vehicles per hour.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages26

Reading Time1 h 6 min

AccessFree to download

Introduction

Motivation and problem context

Research gap and study aim

Data-driven framework and methodology

Transportation Research Part A 206 (2026) 104927 Contents lists available at ScienceDirect Transportation Research Part A journal [homepage:](homepage: www.elsevier.com/locate/tra)[ www.elsevier.com/locate/tra](homepage: www.elsevier.com/locate/tra) | Short-term lagged interactions between freight and passenger volumes in urban traffic: inter- and intra-modal effects with explainable machine learning
E. Amirnazmiafshara , D.P. Song a,* , B. Kenny b, J.M. Wu c, B. Kulcs´ar c, Y.Z. Liu d, C. Olaverri-Monreald
a University of Liverpool, UK b ESG Consultants Ltd, UK
c Chalmers University of Technology, Sweden d Johannes Kepler University, Austria | | | | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | Keywords:
Freight-passenger interaction Urban traffic management Multimodal transport planning Short-term demand forecasting Lagged traffic volumes Explainable machine learning | | Urban transport systems face increasing complexity as freight and passenger flows compete for limited road capacity. While multimodal forecasting methods have progressed, short-term interactions between vehicle classes remain underexplored, particularly in real-world operational settings. This study addresses that gap by examining whether recent freight or passenger volumes are significantly associated with current traffic conditions across modes. Using 6,003 hourly records from Liverpool, UK, we develop an interpretable machine learning framework combining Kmeans clustering, XGBoost classification, and the DALEX explainability toolkit. Results show that one-hour lagged freight volume significantly improves the classification of current passenger traffic states, while the reverse effect is limited. Global feature importance and local interpretability analyses consistently identify freight volume as the most influential predictor. Partial dependence plots (PDPs) reveal a nonlinear inflexion point, where freight volumes exceeding roughly 500 vehicles per hour in this Liverpool case study are associated with reduced passenger flow. McNemar’s test confirms a statistically significant improvement, and robustness checks, including alternative lag structures, interaction terms, and reciprocal models, reinforce the stability of this finding. These insights offer practical value for short-term forecasting, corridor-level coordination, and longer-term multimodal planning. The observed directional asymmetry, wherein freight volumes more reliably predict passenger conditions than the reverse, highlights the potential benefits of incorporating freight data into real-time traffic management systems. More broadly, the study demonstrates how interpretable machine learning can uncover crossmodal dependencies and support the development of more integrated, responsive, and equitable urban mobility systems. | 1. Introduction Urban transport networks today face increasingly complex challenges due to the simultaneous rise in passenger and freight * Corresponding author. [E-mail address:](E-mail address: Dongping.Song@liverpool.ac.uk)[ Dongping.Song@liverpool.ac.uk](E-mail address: Dongping.Song@liverpool.ac.uk) (D.P. Song). [https://doi.org/10.1016/j.tra.2026.104927](https://doi.org/10.1016/j.tra.2026.104927) Received 6 August 2025; Received in revised form 20 January 2026; Accepted 10 February 2026 0965-8564/© 2026 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license ([http://creativecommons.org/licenses/by/4.0/](http://creativecommons.org/licenses/by/4.0/)). mobility demands. As e-commerce and just-in-time logistics continue to expand, freight movement (e.g., via light goods vehicles and heavy goods vehicles) has become more temporally dispersed and spatially embedded within the urban landscape (Safikhani et al., 2020). At the same time, cities must accommodate growing levels of passenger travel (e.g. via cars, taxis, buses, and coaches), often using the same road infrastruct

FAQs

What research question does the study address?

The study examines whether short-term increases in freight or passenger volumes are significantly associated with changes in the other vehicle class’s current traffic conditions across modes.

How is the proposed model made interpretable?

It combines K-means clustering and XGBoost with the DALEX explainability toolkit, using global importance and local interpretability plus partial dependence plots.

What do the results show about directional interaction effects?

One-hour lagged freight volume significantly improves classification of current passenger traffic states, while the reverse effect is limited, indicating directional asymmetry.

Aug 6, 2026

Upload by Aria