Marine pollution, especially oil spill-based, affects both marine and coastal environment is one of the most important issues for the maritime industry. The accurate prediction of the severity of oil spill is of great importance in order to determine the accurate response methods. In this perspective, this study aims to predict the severity of oil spill in possible vessel accidents by examining data based on vessel accidents that cause marine pollution. The United States Coast Guard (USCG) database covering 2002--2015 was utilized and a total of 1468 instances of vessel involved accidents leading oil spill were analysed using Decision Tree (DT) and data-driven Bayesian Networks (BN) called Tree Augmented Naive Bayes (TAN). As a result, the most important contributing factors affecting the severity of oil spill were revealed as "type of accident" and "type of vessel". This study would be a guide that will assist authorities and policy makers in predicting the severity of oil spill, and contribute to the development of important strategies and countermeasures for vessel accidents leading oil spills.