Unleashing data power: Driving maritime risk analysis with Bayesian networks

Abstract

With the rapid growth of global shipping, increasing maritime traffic has heightened accident risks, posing threats to the economy, ecology, and public safety. This study introduces a data-driven Bayesian network (BN) framework to identify key risk factors for incident severity, considering data deficiencies. Firstly, boxplot techniques and the Adaptive Synthetic Sampling algorithm are introduced to handle outliers and imbalanced data, thereby supporting a valid dataset for model construction. Then, this study introduces the AcciMap theory, which provides a more comprehensive representation of accident causation from complex sociotechnical systems perspectives. Meanwhile, the K-means clustering method is employed to effectively overcome the high subjectivity inherent in traditional indicator state classification. Finally, we propose techniques to assess the framework performance and validate our framework. Our findings reveal: (1) “Standardized Operations” are identified as the key influential factor on maritime accidents, with a mutual information value of 0.134; (2) Human behavioral norms gain importance as incident severity increases; (3) Scenario analysis highlights that favorable weather conditions can paradoxically lead to more severe accidents. This study offers valuable insights for policymakers and industry practitioners, providing a robust framework for maritime risk management and accident prevention.