Spatio-temporal impacts of unplanned service disruptions on public transit demand
Year: 2025
Type: article
Abstract: Public Transit (PT) operators often face unplanned subway service disruptions, which are challenging to handle because they need rapid and effective management strategies. Under such conditions, their goal is to limit user dropout by proposing attractive solutions. With the development of Automatic Fare Collection (AFC) systems and the ability to deal with increasing data volumes, research has recently focused on demand-oriented analysis of disruptions and demand-based strategies. However, AFC data can be incomplete and delayed when uploaded in real time and, therefore, are poorly monitored by PT operators. For this reason, this work seeks to show the relevance of using AFC data to enhance the implementation of disruption management strategies. It aims to understand when, where, and why service disruptions occur. Findings show that 39% of the service disruptions observed between 2021 and 2023 have a negative impact on demand. Thanks to the Gaussian Mixture Model (GMM) algorithm, we distinguish 3 levels of negative impacts: high, medium and low intensity. Results highlight that the PT system is more vulnerable to long-lasting disruptions, which occur during peak hours. The connection with the national railway system also increases the vulnerability of PT systems to disruptions. Using a Multinomial Logistic Regression model, this work highlights the causes that mostly harm the demand and calls for line-specific management strategies. In addition, the contextual analysis introduced in this study reveals the clues left by disruptions in demand levels and argues for the development of online disruption detection coupled with flow redistribution models to enhance decision-making. • This work shows the relevance of using Automatic Fare Collection data (AFC) as a proxy for demand to enhance the implementation of service disruption management strategies. Demand levels under disruptions are extracted and clustered using Gaussian Mixture Model (GMM) algorithm. Results highlight that 39% of service disruptions have a negative effect on demand, with three different levels of intensity: high, medium and low. • Using multinomial logistic regression, an explanatory model is built to understand which disruptions’ features play a role in the level of intensity observed. Findings indicate where, when, and why service disruptions occur. • Focusing on the high-intensity cluster, results demonstrate that the public transit system of Lyon (France) is more vulnerable to long-lasting service disruptions, during peak hours. It is more sensitive to the blockage of a subset of stations than the whole line. Also, they are mostly provoked by unattended baggage, power or informatics failure, and incidents related to infrastructure projects. • The study provides a contextual analysis that aims to highlight what happens in the spatio-temporal surroundings of the disruption. Noteworthy, this study reveals that some of the most intense disruptions provoke drops in demand that can be detected 5 min before the registered beginning of disruptions. This argues for the development of online disruption detection coupled with flow redistribution models to enhance operators’ decision-making. It also indicates high variability in the line’s response to disruptions, supporting the need for line-specific management strategies.
Language: English
Subfield: Transportation
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $2,400
OpenAlex ID: W4407414472