Risk analysis of damaged ships – a data-driven Bayesian approach

Abstract

An accident occurring at sea, though a rare event, has a huge impact both on the economy and the environment. A better and safer shipping practice always demands new ways to improve marine traffic and this essentially requires learning from past experience/faults. In this regard, probabilistic analysis of accidents and associated consequences can play a very important role in making a better and safer maritime transport system. Bayesian networks represent a class of probabilistic models based on statistics, decision theory and graph theory. This paper introduces the use of data-driven Bayesian modelling in risk analysis and makes a comparison with the different data-driven Bayesian methods available. The data for this study are based on the Lloyds database of accidents from 1997 to 2009. Important influential variables from this database are grouped and a Bayesian network that shows the relationship between the corresponding variables is constructed which in turn provides an insight into probabilistic dependencies existing among the variables in the database and the underlying reasons for these accidents.

Keywords:

• Bayesian network
• risk analysis
• data-driven Bayesian model
• damage database

Acknowledgement

The authors would like to acknowledge Lloyd's Register, London, for supporting this work and providing the necessary data.