A data-driven bayesian network model for risk influencing factors quantification based on global maritime accident database

Abstract

The Maritime transportation system is exposed to various risks, that can lead to accidents and sometimes resulting in severe economic losses and fatalities. The accident database of maritime accidents contains valuable knowledge about the causes of accidents. An in-deepth understanding of the impact of risk influencing factors (RIFs) on maritime accidents based on historical data helps to prevent accidents from happening in the future. Using a large dataset of 55469 maritime accidents from 2002 to 2022, a Bayesian network (BN) model is formulated to investigate how RIFs affect maritime accidents. The interdependencies between the RIFs are modelled using a Tree Augmented Network (TAN) with sensitivity analysis. The Most Probable Explanations (MPEs) for each type of accident are also identified. The results indicate that older, smaller, non-convenient flagships in the North Atlantic zone have a higher probability of accidents. The ranking of the most important RIFs for accident types is location, ship type, ship age, gross tonnage (GT), and deadweight tonnage (DWT). The effect of different RIFs on different types of maritime accidents is also examined. Ship type is the most important RIF for hull damage, fire or explosion, and contact accidents. Among the different ship types, Cargo ships are at the most significant risk of grounding while fishing ships exhibit the highest risk of hull damage, fire or explosion, and foundering. Age is the most significant RIF for foundering, while ship location is the most significant RIF for machinery damage, grounding, and collision accidents. Based on the above findings, recommendations for reducing maritime risk and promoting sustainable development and conservation of ocean and coastal areas are discussed in detail.