Maritime occupational accidents analysis: A data-driven Bayesian network approach

Abstract

The challenging working conditions aboard ships continuously expose seafarers to significant risks, yet research into maritime occupational accidents remains sparse. To address this gap, this study employs both statistical techniques and the Tree-Augmented Naive Bayesian Network (TAN-BN) model for a comprehensive analysis of Risk Influential Factors (RIFs) in maritime occupational accidents. This study analyses 505 maritime occupational accident cases from 2013 to 2021, identifying 17 RIFs related to consequences on crew, ship factors, human factors, and external environment factors. The approach involves using the database to: 1) conduct statistical analyses that delineate the principal characteristics and trends of maritime occupational accidents, and 2) develop and refine the TAN-BN model to pinpoint the five primary factors impacting accident severity, the number of injured crew members, the specific body part affected, the nature of the injury, the rank of the injured personnel, and their age. Further validation through sensitivity analysis and real-world accident cases confirms the model's robust predictive accuracy, aiding in the identification of underlying causes. This study provides innovative practical implications for maritime stakeholders to develop regulations and measures in the prevention of maritime occupational accidents, protection of occupational safety, and post-accident management.