A data-driven approach to identify key hazards and coupling pathways in metro construction accidents

Purpose Metro construction has become increasingly hazardous due to its complexity, long durations, and dynamic environments. The purpose of this study is to develop a data-driven approach to identify key hazards and reveal their coupling pathways in metro construction accidents, ultimately enhancing risk management and accident prevention strategies. Design/methodology/approach This study integrates three analytical techniques. First, the Latent Dirichlet Allocation (LDA) topic model is applied to extract hazards from 210 metro construction accident reports. Next, the N-K model is used to quantify the coupling degrees among hazards across subsystems. Finally, social network analysis (SNA) is employed to visualize and analyze the coupling network and determine key hazards and their interaction pathways. Findings The analysis reveals four major subsystems comprising 35 distinct hazards. Key hazards include M8 (non-compliance with rules), M7 (inadequate safety control), M5 (lack of technical safety briefings), M12 (unreasonable construction planning), EI6 (object insecurity), H4 (staff irregularities), and H5 (failure to observe hazardous areas). The results highlight that poor safety management frequently triggers multi-hazard coupling, especially along the H–EI–E–M pathway, suggesting critical intervention points for reducing accident risks. Originality/value This study offers a novel framework that combines text mining, coupling theory, and network analysis to uncover the complex interactions between hazards in metro construction. The approach provides actionable insights for safety managers to prioritize high-risk interactions and design targeted interventions.