Q-Learning Approach in Ship Safety Inspection Data

In the study, the authors focus the inspections of deficiencies basing 2000 – 2014 database of Tokyo MOU, and a Q-learning method to check the risk decision-making model. The input data are inspections of deficiencies ratio, current occurrences of deficiencies, and occurrences of deficiencies in the next 5 years, depending on the weight to feedback R value, and execute Q-learning algorithm and Markov chain to correct Q-Table, and the risk of changes of the vessels basing different flag state and classification societies are obtained. The results of study, Port State Control would pay effort on the priority of the vessels in future year, as the ship’s safety status will turn into high-risk in forecasting years. Port State Control would pay effort on the priority of the vessels of the vessels basing different flag state and classification societies to reduce the risk.