A data-driven Bayesian network (BN) model is developed to capture the individual and collective impacts of accident influential factors (AIFs) on different attributes of maritime accidents, including accident type, severity, and loss. To develop the data-driven BN model, a dataset comprising 980 accidents from 2000 to 2023 in the Liaoning Sea area is initially established. Subsequently, the data-driven BN model is constructed using the Tree Augmented Network (TAN) learning algorithm and expectation-maximization (EM) algorithm. Finally, comprehensive BN analysis encompassing strength of influence assessment, sensitivity analysis, scenario simulation, and model verification is conducted based on the developed TAN-BN model. The results demonstrate that distinct types of maritime accidents exhibit varying sensitivities to seasonal variations and time of day; contact and groundings are primarily encountered by general cargo ships while fire and explosions are more prevalent among bulk carriers. Fishing boats tend to be vulnerable to collisions whereas sinking accidents are most commonly associated with sand carriers.