A hybrid approach integrating case mining (CM) and the Copula Bayesian Network (CBN) for accident causation probabilistic reasoning of building construction collapses

Abstract

Construction collapses represent a prevalent type of structural failure in the building industry, involving multiple causal factors. Most current studies on these causal factors and their relationships are derived from expert knowledge and rely primarily on singular probabilistic estimates derived from expert input to predict the likelihood of accidents using Bayesian networks. However, fluctuations in expert knowledge can compromise the stability of Bayesian network structures, thus failing to accurately capture the uncertainty associated with construction collapses. To fill these gaps, a hybrid approach, integrating case mining (CM) and a Copula Bayesian Network (CBN), is proposed to infer the causation probability of construction collapse. To construct a stable Bayesian network structure, the CM method is used to extract the causal factors and their relationships from accident investigation reports. The Gaussian copula function and different marginal distribution functions are introduced to describe the probability distribution of the occurrence of causal factors. Correlation analysis, forward analysis and backward analysis are used to quantify the risk propagation relationships among the causal factors. The findings indicate that (1) construction collapses predominantly occur between 9:00-10:00 and 14:00-15:00 and (2) illegal construction, lack of administrative supervision and non-implementation of construction schemes are critical factors contributing to these incidents.