Data-driven analysis of key factors in building fire risk

Purpose Building fire accidents arise from complex nonlinear risk interactions. Traditional static assessments often fail to capture these interactions accurately. This study elucidates the evolution mechanisms across diverse building categories to establish a scientific foundation for developing differentiated control strategies. Design/methodology/approach A framework integrating Text Mining, Association Rule Mining, and Complex Networks was applied to 853 verified Chinese accident reports. Fifty-one factors were extracted using the TF-IDF algorithm. A novel “Coupled Risk Index”, incorporating mutual information and posterior probability, was constructed to quantify risk transmission intensity. The topology and robustness were evaluated using targeted attack simulations. Findings The results reveal productive scenarios exhibit a “convergence-type” mechanism centering on “Lack of Fire Safety Training.” Conversely, non-productive scenarios show a “divergent” mechanism spreading from systemic and organizational flaws. The analysis confirms that synchronizing the targeted removal of high-importance nodes and high-vulnerability edges optimizes governance effectiveness under resource constraints. Originality/value By transforming text-based incident reports into structured networks, we uncovered unique risk patterns for different building types. Identifying these pathways helps shift fire safety from reactive responses to proactive interventions, thereby improving the urban resilience. This study systematically highlighted the primary risk association patterns.