Research On Factors Affecting The Marine Accident Based On Tree Augmented Naive Bayesian Network

Maritime transportation is a crucial foundation of international trade.The volume of maritime transportation continues to increase with the rapid development of the global economy.Despite the industry’s significant efforts to enhance maritime safety,the frequency of marine accidents has not declined to the expected level.Therefore,maritime safety remains a hot topic in both the shipping industry and academia.The causality and severity of marine accidents are two vital topics in maritime safety research.Marine accident causality involves analyzing and predicting the evolution mechanism of accidents due to numerous factors such as the cause and accident environment.Research focusing on the severity of marine accidents mainly concentrates on evaluating the degree of crew injuries and ship damage in specific types of accidents or hazardous environments.There are still certain limitations in existing studies on marine accidents,particularly the lack of research on the correlation between accident influential factors and accident severity.This paper addresses the correlation between marine accident influential factors and accident severity through the following steps:(1)Firstly,extensive collection and filtering of marine accident data was performed,and1294 ship accident investigation reports from 2000 to 2019 were selected based on the completeness of ship accident investigation reports.Text analysis methods were used to classify and mark the main influencing factors of ship accidents,and a database of 35 ship accident severity influencing factors was established.(2)Secondly,a tree-enhanced naive Bayesian network model(TAN-BN)was established based on a data-driven method,and the established accident influential factor database was used as the training data of the machine learning model.The parameter learning and structure optimization of the TAN-BN model were conducted.Correlation degree indicators,such as mutual information,joint probability,and real risk impact,were combined to analyze the correlation between accident influential factors and the severity of ship traffic accidents.(3)Finally,the effectiveness of the model was validated through sensitivity analysis,receiver operating characteristic(ROC)curve,and practical cases.The most likely explanation(MPE)model analysis method was used to explain the possible emergency situations that may occur in different accident scenarios and the possible chain phenomena that may occur under fixed information configurations to help identify potential related risks.This paper analyzes the most important influential factors concerning several levels of accident severity,including the types of accidents most likely to cause "particularly serious accidents." The study finds that "capsizing/sinking," "hull/equipment failure," and "collision" are the most likely causes of particularly severe accidents.Fishing boats and other small vessels are also more likely to experience "particularly serious accidents" than any other type of vessel.The results of this paper can help relevant maritime departments analyze and forecast ship accidents,propose safety measures,and maintain safe ship navigation.