A bayesian network approach to ship safety assessment: integrating machine learning and expert opinions

• Published in: Reliability engineering & system safety Vol. 266; p. 111659
• Main Authors: Wang, Nanxi, Yuen, Kum Fai, Li, Duowei, Wong, Yiik Diew, Tan, Kim Hock
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2026
• Subjects: Bayesian network; Machine learning; Risk assessment; Safety index; Scenario deduction; Ship safety; Ship safety; Scenario deduction; Bayesian network; Risk assessment; Safety index; Machine learning
• ISSN: 0951-8320
• DOI: 10.1016/j.ress.2025.111659

•Provide the nSmE concept for analysing human-environment interactions in accidents.•Proposes a Bayesian Network-based hybrid model for modelling ship safety.•This model integrates ship, human, internal and external environmental factors.•Uses scenario-based deduction methods to establish factor structure relationships.•Uses machine learning and expert opinions to model conditional probability tables. Maritime accidents pose a significant threat to maritime operations. With technological advancements, a substantial amount of objective data can now be collected and utilized to improve ship safety assessment. This study proposes an assessment model that leverages comprehensive objective data collection to enhance maritime safety management. Theoretically, this study introduces a novel multi-factorial accident model, termed nSmE (n Subjects – m Environments), which emphasises the interaction between human behaviours and environmental conditions in accident causation. Methodologically, a Bayesian Network (BN)-based hybrid approach is employed to model the 2S2E (2 Subjects – 2 Environments) ship accident framework, systematically integrating factors across ship-related factors, human factors, internal environmental factors, and external environmental factors. A scenario-based deduction approach is employed to establish factor structure relationships. To model the conditional probability tables, a two-part modelling strategy is adopted. Machine learning techniques are applied to model context-dependent factors, while a combination of Failure Mode and Effects Analysis, IF-THEN rule-based reasoning, and Dempster-Shafer Evidence Theory is employed for directly measurable factors through incorporating expert opinions. Finally, case studies validate the model’s effectiveness and provide key recommendations for improving ship risk management. The results show that human reliability is a critical factor in accident prevention which requires targeted efforts to enhance seafarers' working conditions. Improving crew safety and performance through specialised equipment, such as protection against extreme temperatures and excessive noise exposure, is essential. Additionally, safety improvement strategies should be tailored to the specific risk factors present in each scenario to ensure effective risk mitigation and accident prevention.