A marine accident analysis based on data-driven Bayesian network considering weather conditions and its application to Taiwanese waters

Abstract

Few marine accident studies contemplate the weather effects on the places where and dates when shipping casualties occur. This study conducts a marine accident analysis based on Bayesian Network (BN) considering the weather characteristics of the accident spots and occurrence dates. The weather element is adopted as a node in the BN that comprises a group of five clusters illustrating the states of current speed, wave height and wind speed. A total of 431 accidents occurring in the western coast of Taiwan during the period from 2014 to 2022 are analyzed. Wave with height greater than slight sea has a certain effect on the accident frequency and severity. Mishaps are likely to suffer severe consequences if encountering the weather conditions featured by strong current, moderate/fresh breeze and slight/moderate sea. In addition, Accident Type is the most influential node of Accident Severity among the parameters, followed by Ship Type, Gross Tonnage (GT), Weather Cluster, Location and Ship Age. Finally, the three most crucial parameters leading to Very Serious Marine Casualty are Accident Type, Ship Type and GT.