Identification of key risk-influencing factors in merchant ship collision occurrence using a data-driven bayesian network with emphasis on ship-related factors

Identifying key risk-influencing factors (RIFs) for collision occurrence is essential for preventing accidents. While most studies focus on accident severity, few address occurrence. Accident data alone are insufficient for understanding the occurrence, which requires comparing vessels involved in accidents with those that were not—necessitating exposure data. Automatic Identification System (AIS) data are often used for this purpose but limit the analysis to short-term or regional scopes. To overcome these limitations, this study constructed a ship × year-based database using long-term, global data from world fleet and maritime casualty databases. This study used a tree-augmented naïve Bayes-based Bayesian network (BN) to identify RIFs contributing to collision occurrence, focusing on ship-related RIFs. The relative importance of each factor was evaluated by introducing evidence into the BN and calculating the posterior probability of collision. The results reveal that the period and built year have a greater influence on collision occurrence than age. Other significant RIFs included the gross tonnage, draught, speed, and deadweight tonnage. Incorporating these RIFs into conventional risk assessments can enhance objectivity, reducing reliance on subjective expert judgement and strengthening decision-making in safety management.