Abstract: To identify the critical causes of various metro construction accidents and to support the decisionmaking of the accidentrelated parties in risk analysis, prevention, and control, the reported data from 202 metro accidents from 2011 to 2021 in China is collected and statistically processed. The risk indicators are extracted, integrated, and screened. A graphical model structure is constructed and model parameters are determined by tree augmented naive Bayesian and expectationmaximization algorithms, respectively, while considering the perspectives of accident occurrence, direct causes, and indirect causes. The final analysis model is established by training with GENIE software. The analytical results of the Bayesian model reveal the following. First, the essential riskcausing factors for various types of accidents are clarified through forward reasoning, and the overall degree of impact for each risk factor triggering accidents is ranked in importance. Second, the reverse diagnosis illustrates the role of the proposed model in supporting decisionmaking for risk prediction under various risk factor combinations. Third, the 10fold crossvalidation confirms the validity of the model.

Key words: metro section, station construction, safety accident; tree augmented naive Bayesian, risk factors