A data-driven risk model for maritime casualty analysis: A global perspective

Abstract

Maritime casualty analysis needs to be addressed given the increasing safety demand in the field due to the accidents’ low-frequency and high-consequence features. This paper aims to delve deeper into the factors that affect maritime accident casualties by establishing a new database and conducting an accident casualty evolution analysis. Based on the refined dataset, a pure data-driven Bayesian Network (BN) model is developed to conduct the casualty analysis of maritime accidents that occurred under different ship operational conditions. Methodologically, it introduces new risk factors to improve maritime casualty analysis accuracy through the enriched updated maritime accident database. Furthermore, the new database is categorised into five new datasets based on temporal development trends to better analyse the evolution of the casualty. Five risk analysis models are individually constructed based on different timeframes to illustrate the dynamics of the casualties and compared by seven evaluation indexes to demonstrate the effectiveness of the proposed data-driven BN model. It, for the first time, investigates the changing roles of different risk factors on maritime casualties with time. The insights gained from this model are invaluable, contributing to improved risk prediction and maritime safety strategies by acknowledging the changing patterns of maritime accidents.