Ship detention prediction via feature selection scheme and support vector machine (SVM)

Ship detention decision plays a key role in port state control (PSC) inspection process, which is compactly related to navigation safety and maritime environmental protection. Many focuses were paid to exploit intrinsic relationship among ship attributes (ship age, type, etc.), detention events and typical ship deficiencies. It is noted that many ship detention prediction frameworks were implemented considering single type of factors regardless of internal relationship between ship crucial deficiencies and ship attributes. To address the issue, we proposed a support vector machine (SVM) based framework to exploit crucial ship deficiencies, and thus forecast the probability of ship detention event. Firstly, we design a feature selection scheme to determine ship fatal deficiency types by exploring historical PSC inspection data. Secondly, we predict the ship detention event via conventional support vector machine (SVM) with support of ship feature selection outputs. Thirdly, we verify the proposed framework performance by predicting ship detention event from historical PSC data, which is quantified with the indicators of accuracy (Acc) and area under ROC curve (AUC). The research findings help PSC officials easily identify fatal ship deficiencies, and thus make more reasonable ship detention decision in real-world PSC activity.

Automated summary

The research highlights the importance of ship detention decisions in the PSC inspection process and proposes an SVM-based framework to predict the probability of ship detention events. This framework utilizes a feature selection scheme and support vector machine to achieve the prediction, with performance validation done on historical data.