Predictive power of AIS on marine insurance : a demonstration of how activity level and operational patterns of the merchant fleet can be used to predict P&I insurance claims using machine learning

Abstract

Digitalisation is making a growing appearance across all sectors, and traditional P&I insurance is no exception. Marine insurance is said to be several years behind traditional land-based insurance when it comes to digitalisation. This thesis is attempting to narrow the gap, by investigating the potential of applying machine learning on AIS-information against the extensive database on P&I insurance claims from the P&I Club Skuld. The thesis aims at investigating the potential to predict P&I insurance claims based on variables retrieved from AIS. AIS-information from 2013-2017 and Skuld's claims data for the same period was combined, and a total of five machine learning methods were tested to assess the predictive power of AIS-information. An extensive pre-processing was executed to make the data available for machine learning, and this section provides detailed information to anyone that aims at utilising AIS in their research. The research finds that AIS-information has predictive power for claims, as it links claims to activity level and operational patterns of the merchant fleet. The findings have implication for two fields in marine insurance; risk assessment/ pricing and loss prevention. In relation to loss prevention; average distance sailed, number of unique ports visited, and total distance sailed were found to have the most predictive power. Regarding risk assessment, the strongest model was able to predict 79 % of all cargo claims for Bulk & Cargo small. The research has revealed that machine learning has potential to create significant value in P&I insurance and that an extensive amount of data is ready to be applied in the pursuit of more accurate risk assessments and more precise loss prevention measures. Estimates vary between a potential yearly reduction in claims of 7-14%, in addition to increased revenue as a result of correct pricing.