Using data-driven Bayesian networks to analyze maritime accidents in congested waters

Abstract

This study adopted a data-driven approach using Tree Augmented Bayesian Networks (TAN) to analyze risk influencing factors for maritime accidents in congested waters in Japan. Specifically, data on 2,192 maritime accidents that occurred within the Tokyo Bay, Nagoya Port, Ise Bay, Osaka Bay, Bisan Seto, Kurushima Strait and Kanmon Strait during the period 2009-2022 were analyzed. In the analysis, 19 risk-influencing factors were extracted and the interactions between these factors were visualized and quantified using a Bayesian Networks (BN) structure. Particular focus was placed on seven factors with high mutual information values (human error, ship type, length, gross tonnage, number of crew, AIS on board, and total experience on board), and the impact of these factors on five accident types - collision, stranding, capsize/flooding, fire/explosion and inoperability - was analyzed in detail. In addition, a scenario analysis using the BN structure was used to assess the risk of maritime accidents under specific conditions. The analysis suggested the need for countermeasures against 'overconfidence' and 'pride' of experienced ship operators, maintaining vigilance under good navigational conditions, and risk management according to vessel characteristics.