Automated machine learning for maritime accident analysis: A case study

Abstract

The maritime fishery sector continues to be one of the highest-risk sectors due to the vulnerability and hostility of the marine environment, but there are always ways of reducing these risks. Accidents are often caused by more than one factor contributing to a complex interaction. Identifying the root causes and their interactions is essential to preventing and understanding these accidents. This study presents an analysis of maritime events in Morocco between 2014 and 2023 using automated machine learning (AutoML) techniques. These techniques offer a multitude of options for examining voluminous amounts of historical data concerning marine events through advanced algorithms that facilitate the incorporation of predictive analysis into decision-making, operational and policy processes with the aim of increasing safety at sea. Multiple classification ML algorithms were trained such as Decision Tree, Xgboost and Random Forest with the aim of increasing the accuracy and robustness of detection. The ensemble model, which merges six separate models to generate a final prediction, proved to be the best performer in terms of optimal predictive accuracy.