Maritime accident risk prediction integrating weather data using machine learning

Abstract

The study explores the capability of various machine learning (ML) models in maritime accident risk prediction. Data from 1981 to 2021 from the Norwegian Maritime Authorities (NMA) was analysed together with the data of 51 different weather-related variables, which were collected from Visual Crossing for each accident recorded in the NMA dataset. The findings reveal an increased predictive ability of ML models when relevant weather data is introduced. The results show that the Light Gradient Boosted Trees with Early Stopping perform the best, with a five-fold cross validation accuracy of 70.23% when weather data was included, compared to 64.86% without. Furthermore, the study revealed that the leading weather variables for accident prediction are wind, sea level pressure, visibility, and moon phase. The most effective multi-classification ML algorithm can be deployed for improving maritime safety resilience through vulnerability assessment and preparedness.