Modelling of ship navigation in extreme weather events using machine learning

Extreme weather events such as hurricanes are a significant hazard to shipping. We show that traditional methods to model weather related risks using naval architecture or historical incidents fail to accurately predict the potential risk of an accident by failing to account for risk mitigation actions taken by the bridge team. We therefore propose the use of unsupervised machine learning to identify clusters in risk response by ships to perceived high risk scenarios. This risk classification method is based on the analysis of large heterogenous datasets including vessel traffic, metocean and hurricane path data from the US Atlantic Hurricane Season. Clusters in vessel behaviour to these storms are identified and the risk perception by storm severity compared. The results of this analysis can be used to better understand the impact of extreme weather events on navigation safety and develop an early warning system for coast guard search and rescue response.