Analysis of the Risk Evolution of Fatal Accidents in the Construction Industry Based on Bayesian Network

Abstract

Fatal construction accidents, which cause significant loss of life, often arise from complex risks that evolve over time rather than from isolated or static hazards. However, existing research largely neglects this dynamic nature and offers limited insight into the temporal evolution of construction safety risks. To address this gap, this study systematically investigates the interperiod changes of construction safety risks to improve understanding of the dynamic mechanisms underlying fatal construction accidents. Specifically, tree-augmented naive Bayes models were developed for 357 fatal accidents in Hong Kong across two intervals (2000–2010 and 2011–2021), enabling explicit visualization of the state probability distributions for each descriptive and causal factor. Sensitivity analyses were subsequently performed to identify the most impactful factors, followed by scenario-based simulations to evaluate the risk evolution trends of various accident typologies during the two periods. Findings indicate the time-varying dynamics of construction safety risks: (1) falls from height remain the most prevalent accident type, increasing from 50.0% in the earlier period to 59.2% in the later period. (2) Among descriptive factors, specific risk agents and work types are strongly associated with distinct accident types; among causal factors, lack of skills/knowledge and unsafe workplace arrangements have recently become more critical in determining accident type. (3) The probabilities of inadequate risk assessment, insufficient safety training, and deficient safety rules or procedures have increased significantly across all accident types. This study deepens understanding of the temporal evolution of construction safety risks and highlights the need to allocate adequate safety resources and establish effective safety management systems, particularly for preventing falls from heights.