Strengthening safety in the first line: An advanced data-driven approach to optimize flag state implementations

Abstract

Ship inspections are one of the most important implementations for ships to maintain standards in the fields of safety, security, and environmental management. The main objective of this research is to develop an objective ship targeting model based on machine learning using port state control reports for flag state implementations considered as the first line of safety. In this context, the Turkish flag state was selected as the target fleet, and 6008 inspection reports from four memorandums in which this fleet sailed most frequently were analyzed using three different Naive Bayes-based algorithms. Moreover, a model was developed not only for detecting substandard ships, but also for identifying the specific areas in which these ships may be deficient. It was determined that the accuracy value of the model predicting the detection of a deficiency on the ship reached 73.4 %, and for the deficiency areas, these values were between 64.6 and 99.4 %. Models with satisfactory levels of performance metrics were also supported by scenario analyses. The most important variables affecting the detection of deficiency on the ship were found to be "ship age", "classification society" and "ship deficiency index", respectively. The research novelty is that it has feasible approach for flag state implementations by integrating machine learning approaches into ship inspections. The developed models can minimize the risks of the ships in terms of safety, security, and environment by detecting the substandard ships at the first stage for the flag state implementations and may be facilitators for other inspection implementations, especially port state controls.