Data-driven causal factor analysis of metro construction incidents using complex network theory

Abstract

Near misses and accidents remain two of the most prevalent undesired incidents in metro construction, posing significant threats to safety. While these incidents often share common causal factors, the large number of factors involved in these incidents and the intricate relationships among them make it difficult to identify potential hazards and formulate targeted prevention measures. To address this challenge, this study applied complex network theory to systematically examine the interrelationships among causal factors of metro construction incidents. A case study approach was adopted, drawing on more than 4,000 near-miss and accident reports collected from multiple sources, such as government websites and construction sites. Accident chains and near-miss causation-attribute chains were extracted based on a comprehensive list of causal factors and work attributes (i.e., construction phase, construction area, and worker type). They were then used to construct a two-layer Metro Construction Incident Network (MCIN), capturing the multifaceted interactions between factors. Robustness assessment indicated that strength-based attack was one of the most effective strategies for incident prevention. Also, network topology analysis identified critical causal factors of accidents and near misses and revealed their occurrence patterns across different work attributes. Integrating work attributes into analysis provides greater flexibility for developing targeted prevention strategies for safety risks that are prone to incidents. The findings offer both theoretical insight for advancing accident causation analysis and practical guidance for improving safety risk management in metro construction.