An Artificial Intelligence Model Considering Data Imbalance for Ship Selection in Port State Control Based on Detention Probabilities

Port state control inspection is seen as a safety net to guard marine safety, protect the marine environment, and guarantee decent onboard working and living conditions for seafarers. A substandard ship can be detained in an inspection if serious deficiencies are found onboard. Ship detention is regarded as a severe result in port state control inspection. However, developing accurate prediction models for ship detention based on ship's generic factors (e.g. ship age, type, and flag), dynamic factors (e.g. times of ship flag changes), and inspection historical factors (e.g. total previous detentions in PSC inspection, last PSC inspection time, and last deficiency number in PSC inspection) before an inspection is conducted is not a trivial task as the low detention rate leads to a highly imbalanced inspection records dataset. To address this issue, this paper develops a classification model called balanced random forest (BRF) to predict ship detention by using 1,600 inspection records at the Hong Kong port for three years. Numerical experiments show that the proposed BRF model can identify 81.25% of all the ships with detention in the test set which contains another 400 inspection records. Compared with the current ship selection method at the Hong Kong port, the BRF model is much more efficient and can achieve an average improvement of 73.72% in detained ship identification.

Automated summary

Port state control inspection is crucial for marine safety, and developing accurate prediction models for ship detention is challenging due to imbalanced inspection records dataset. The proposed balanced random forest (BRF) model shows promising results in identifying detained ships at the Hong Kong port, achieving an average improvement of 73.72% in ship identification efficiency compared to the current method.