Accident data-driven human fatigue analysis in maritime transport using machine learning

In maritime transport, fatigue conditions can impair seafarer performance, pose a high risk of maritime incidents, and affect safety at sea. However, investigating human fatigue and its impact on maritime safety is challenging due to limited objective measures and little interaction with other risk influential factors (RIFs). This study aims to develop a novel model enabling accident data-driven fatigue investigation and RIF analysis using machine learning. It makes new methodological contributions, such as 1) the development of a human fatigue investigation model to identify significant RIFs leading to human fatigue based on historical accident and incident data; 2) the combination of least absolute shrinkage and selection operator (LASSO) and bayesian network (BN) to formulate a new machine learning model to rationalise the investigation of human fatigue in maritime accidents and incidents; 3) provision of insightful implications to guide the survey of fatigue's contribution to maritime accidents and incidents without the support of psychological data. The results show the importance of RIFs and their interdependencies for human fatigue in maritime accidents. It takes advantage of available knowledge and machine learning to open a new direction for fatigue management, which will benefit the maritime fatigue investigation and provide insights into other high-risk sectors suffering from human fatigue (e.g. nuclear and offshore).