A data-driven risk assessment of Arctic maritime incidents: Using machine learning to predict incident types and identify risk factors

Abstract

The rise in Arctic temperatures has caused sea ice to melt, making the region more navigable through a new, shorter shipping route. Navigating the Arctic poses high risks due to extreme environmental conditions. This study proposes scalable data-driven predictive models to assess the risk of Arctic navigation under uncertain weather and sea ice conditions. Machine learning is applied to predict the type of Arctic incidents and identify their risk factors. Several models are investigated, and the most accurate model for the two Arctic routes, the North West Passage (NWP) and the Northern Sea Route (NSR), is determined based on several validation metrics. Random forest and Naïve Bayes models provide the best accuracy and F1-score for NWP and NSR, respectively. The wind speed, vessel type, length, and age are important risk factors for NWP while temperature at 2 m above the surface, vessel length, and age are important for NSR. Partial dependence plots are used to investigate the effect of each feature on predicting each incident type. Equipment failures are more common among newer and longer vessels. Collision related incidents are more likely to be predicted for longer vessels, while grounding related incidents are more frequent at higher air temperature.