A global data-driven Bayesian network model for ship collision accident analysis
利用1978-2024年全球碰撞事故报告数据，构建树增强朴素贝叶斯网络模型，同时分析事故、环境、船舶和驾驶员四类风险因素，识别高发区域，为海事管理提供决策支持。
Ship collision accidents remain one of the most frequent and severe types of maritime incidents worldwide, often resulting in significant property damage, loss of life, and pollution. Despite extensive regulatory and technological advancements, the underlying risk mechanisms and consequence formation processes of collision accidents are not yet fully understood, particularly from a global and data-driven perspective. To address this gap, this study develops a data-driven Bayesian Network framework to analyze and predict the consequences of ship collision accidents on a global scale. The proposed model is constructed using a Tree-Augmented Naive Bayes (TAN) structure and is trained on a long-term dataset of collision accident investigation reports covering the period 1978–2024, compiled from multiple international maritime bodies. Unlike most previous studies that primarily focus on causal factors or rely on expert judgment, this research simultaneously investigates accident related, environmental, ship related, and bridge operator related risk influencing factors (RIFs), while explicitly accounting for the characteristics of both vessels involved in a collision. In addition, spatial analysis based on accident coordinates reveals distinct geographical clustering patterns in high traffic coastal and port areas. The findings of the study provide practical value for maritime administrations, ship operators, and policy makers by supporting evidence-based decision making in collision prevention, risk informed regulation, targeted training strategies, and the prioritization of proactive safety measures at both operational and strategic levels.