A stratified hybrid model for scenario-weighted causal risk assessment of construction sites

Purpose Interdependent hazards and changing operating conditions complicate construction safety prioritization. This study develops a stratified probabilistic framework to assess these influences jointly and support context-specific decisions. Design/methodology/approach The framework was applied to 36 risk factors across nine dimensions at a large mixed-use construction project in Türkiye under six external states. Fuzzy decision-making trial and evaluation laboratory (DEMATEL) informed network structure, evidential reasoning generated prior distributions, and best–worst method (BWM) importance weights parameterized Bayesian conditional probability tables. State probabilities were elicited from five experts through BWM likelihood comparisons of the six states. Comparative state probabilities were derived from a Markov transition matrix proposed by a large language model, and the final ranking was compared with a BWM-based benchmark. Findings Final risk scores ranged from 0.5125 to 0.7146. The highest priorities were identified as work schedule pressure (0.7146), stock control and planning (0.7001), documentation practices (0.6908), storage and organization (0.6907) and working hours/shifts (0.6889). The final ranking showed high agreement with the comparative Markov-based ranking (Spearman's ρ = 0.994) and moderate agreement with the BWM-based benchmark (ρ = 0.547). Practical implications The findings identify scheduling, material management, documentation and working arrangements as priorities alongside physical hazard controls. Results remain specific to the assessed project and modeling assumptions; independent empirical validation remains necessary. Originality/value The framework links expert-informed risk structuring, probabilistic propagation and scenario-weighted prioritization, with distinct roles for importance weights and probability distributions.