Bayesian network model construction of ship accidents under small sample conditions

Ship accident risk research is often affected by insufficient data and cannot be analyzed accurately. In this paper, to make up for the defect of not being able to accurately identify the interrelationships among the influencing factors in the case of small samples, an information diffusion- based method is proposed to diffuse the information that can be collected at present, so as to obtain a sufficient sample size to realize the accurate mining of the influencing factors. Secondly, considering the defect that traditional Bayesian network assumes that the factors are indep intent of each other, a tree Bayesian network is introduced to further identify the interrelationships among the influencing factors. The dataset adopts the grounding, reefing, explosion, collision, and touchdown accidents of China Maritime Safety Administration in 2014- 2022, and 21 risk fact ors are identified from four perspectives, namely, ship, environment, management, and human factors, and the results of the study indicate that the failure of the shipping company to implement the safety training program, the insufficient professional knowledge of crew members, and the insufficient lookout of crew members are the main factors leading to the ship accidents. Finally, the validity of the data is verified by the mutual information value and belief variance among the indicators, and the five important indicators with the highest correlation are analyzed separately. In order to verify the effectiveness of the model, this paper randomly selects 30 test samples to bring in by means of scenario simulation, and the accuracy of the test results are all above 65%, which indicates that the model has a high degree of confidence, and it can provide reference suggestions for the shipping enterprises and the related maritime safety management departments.