Machine learning-based classification of maritime accidents

Different working conditions in maritime transportation could cause misjudgement of any possible risk or near-miss situations that may lead to injuries or deaths. In this study, 99 accident reports gathered from the EMSA portal are analysed to use a machine learning approach to classify whether crew workload or fatigue under pilotage could lead to incidents or accidents. Operation types, vessel type/age, accident time, weather conditions, accident, flag state, cause of the event, injured person/death numbers and the pilotage status variables are used to classify minor or major accident cases, whether the vessel is under pilotage or not. The results revealed that the Logistic Regression method has the best accuracy of 87% along with the Ridge Classifier. The rest of the models SVM, Na & iuml;ve Bayes, Decision Tree, Gradient Boosting and Dummy have an accuracy of 85% for accident classification in the scope of pilotage operations.