China’s urban rail transit network has expanded rapidly and become highly complex. By the end of 2024 there were 361 operating lines totaling 12,160.77 kilometers and serving 6,651 stations. As system scale and complexity increase, operational uncertainty and vulnerability rise as well, and accidents can cause service disruption, passenger harm, equipment damage, and large socioeconomic losses. The authors note that many existing safety measures focus on single-factor control, hazard identification, or post-accident investigation, and may insufficiently address the complex interrelationships among multiple causative factors. To address this gap, the study integrates a causation model with data-mining and complex network methods to identify and prioritize risk factors in subway operation accidents from a system perspective.
The analysis used textual data drawn from 76 subway operation accident reports produced under China’s accident reporting framework. The authors emphasize that Chinese accident reports are relatively consistent due to unified management and classification systems, which provides a coherent dataset for systematic mining. All relevant data and supporting information are reported to be included within the manuscript and its supporting files.
To extract risk factors from narrative accident reports, the study employed the 24Model, a structured causation framework that enables systematic coding of human, organizational, technical, and environmental contributors to accidents. Using the 24Model, each accident report was analyzed to identify discrete causative elements that serve as nodes in subsequent association and network analyses. The causation-model approach is presented as a way to capture multidimensional factors beyond direct engineering causes and to improve detection of human and organizational contributors.
After factor extraction with the 24Model, the Apriori algorithm was applied to the coded factors to derive association rules that describe frequently co-occurring risk-factor combinations. These association rules formed the basis for constructing a directed/undirected risk-factor network in which nodes represent risk factors and edges represent statistically supported associations derived from the Apriori output. The resulting network is intended to reveal coupling relationships among factors and to support system-level analysis of how causes interact.
To validate the constructed risk-factor network and to identify the most influential factors, the authors used network robustness analysis and mutual information theory. Robustness analysis examines how network structure responds to perturbations or node removal, helping to identify nodes whose removal most degrades network connectivity. Mutual information was used to quantify the information coupling among factors and support ranking of factor importance based on interaction strength rather than solely on frequency counts.
The integrated analytical process identified multiple interrelated risk factors and highlighted those with greatest structural importance in the network. The four most critical risk factors reported were:
These factors emerged as highly connected and influential within the association network, indicating they play substantial roles in coupling chains that can precipitate or amplify accidents. The authors report that network-based ranking and robustness evaluation supported these factors as priorities for preventive management.
The study contrasts conventional single-factor or post hoc approaches with a system-oriented method that combines causation models, association-rule mining, and complex network analysis. The authors argue this integrated approach better captures interactions among human, organizational, and technical contributors and can reveal leverage points for resource allocation. Specifically, strengthening clear allocation of safety responsibilities for both front-line employees and managers, improving oversight of subcontractors, and tailoring training content are presented as priority interventions based on the network findings. The paper situates these conclusions in the context of China’s ongoing efforts to refine operational safety regulations, safety training, and intelligent monitoring practices.
The study demonstrates a methodological pipeline—24Model coding, Apriori association-rule mining, network construction, and validation via robustness and mutual information—to identify and prioritize subway operation risk factors from accident reports. The authors conclude that the identified top factors should guide safety management and resource allocation to support safer, more sustainable subway operations. The manuscript indicates that all data and supporting information are provided within the article and its supporting files. Specific methodological parameters, detailed rule thresholds, and full network metrics are reported in the paper; where particular details are not summarized here, those specifics are available in the original manuscript and supporting information.
The authors state that all relevant data are contained within the manuscript and its supporting information files. The study was conducted using 76 accident reports as described, and supporting tables and figures illustrating network structure, rules, and robustness analyses are provided within the published article.