A data-driven Bayesian Network model for oil spill occurrence prediction using tankship accidents

Oil spills are one of the most important issues facing the maritime industry, with a wide range of catastrophic environmental, social, and economic effects. While all marine accidents can cause pollution, tankships are most likely to cause oil spills due to their cargo content. Accordingly, this study develops a model based on a data -driven Bayesian Network (BN) algorithm to predict whether oil spills may occur following tankship accidents using a total of 2080 accident reports of non-US flagged vessels from the database of the United States Coast Guard (USCG). The analysis shows that the developed model has a very high predictive power with an accuracy value of 75.96%. The most important variables affecting oil spill probability are accident type, vessel age, vessel size and waterway type. The findings are also supported by various scenario tests. These findings will be especially useful for decision-making authorities to predict as quickly as possible whether an oil spill will occur following an accident in order to reduce the time to intervene.

Automated summary

This study develops a model based on a data-driven Bayesian Network algorithm to predict the occurrence of oil spills following tankship accidents. The findings suggest that accident type, vessel age, vessel size, and waterway type are the most important variables affecting oil spill probability. These findings provide valuable insights for decision-making authorities.