Interpreting the influential factors in ship detention using a novel random forest algorithm considering dataset imbalance and uncertainty

Abstract

Port State Control inspects foreign ships in national ports to verify that ships' conditions and equipment obey international regulations and that the ships are crewed and operated in accordance with these regulations. Port State Control has proven useful in ensuring a “safer ship and cleaner ocean.” To support the effectiveness and efficiency of inspections, targeted ships should only be considered if they are at a high risk for accidents. The key factors for ship selection have been included in inspection regimes, but their combined effect on causing ship detention is unclear. Meanwhile, certain factors are characterized by data uncertainty that may influence inspection results and even the time window of an inspection. Furthermore, although tens of thousands of inspection data items are produced yearly, the probability of ship detention is around 3%. Therefore, a new uncertain random forest algorithm has been developed to address factor uncertainty and data imbalances. This algorithm generates rules for the relationships between the multi-factors and ship detention with high accuracy and robustness performance. Based on uncertain random forest models, the following three results are presented. First, the optimal data balancing strategy is a detention ratio of 30% rather than 50%, which could better balance inspection accuracy and efficiency. Second, data uncertainty influences the prediction probability of ship detention; as the uncertainty interval range increases, the prediction probability decreases. Third, the uncertain random forest algorithm generates Port State Control's association rules. Thus, this algorithm can help port authorities identify substandard vessels more efficiently. © 2024 Elsevier Ltd