Data-driven Bayesian network for risk analysis of global maritime accidents

Maritime risk research often suffers from insufficient data for accurate prediction and analysis. This paper aims to conduct a new risk analysis by incorporating the latest maritime accident data into a Bayesian network (BN) model to analyze the key risk influential factors (RIFs) in the maritime sector. It makes important contributions in terms of a novel maritime accident database, new RIFs, findings, and implications. More specifically, the latest maritime accident data from 2017 to 2021 is collected from both the Global Integrated Shipping Information System (GISIS) and Lloyd's Register Fairplay (LRF) databases. Based on the new dataset, 23 RIFs are identified, involving both dynamic and static risk factors. With these developments, new findings and implications are revealed beyond the state-of-the-art of maritime risk analysis. For instance, the research results show ship type, ship operation, voyage segment, deadweight, length, and power are among the most influencing factors. The new BN-based risk model offers reliable and accurate risk prediction results, evident by its prediction performance and scenario analysis. It provides valuable insights into the development of rational accident prevention mea-sures that could well fit the increasing demands of maritime safety in today's complex shipping environment.