Random forest regressor-based Bayesian belief network approach for predicting self-heating and explosion risk of coal cargo in maritime transportation

Abstract Coal cargo self-heating and explosion risk poses significant challenges in maritime transportation since consequences may cause severe damage to the vessel, port and ship crew or marine environment. This paper presents a machine learning tool to predict a probabilistic failure framework integrating a Random forest regressor (RFR) and a Bayesian belief network (BBN) approach for self-heating and explosion risks of coal cargo on bulk carrier ships. The proposed approach systematically quantifies uncertainties associated with human and technical failures contributing to coal self-heating and explosion accidents. While RFR provides the ability to learn non-linear relationships from historical data, BBN enables dynamic probabilistic risk assessment by capturing causal dependencies between contributing factors. The outcomes of research highlight critical risk pathways and key failures in the system, providing practical insights for maritime safety regulations and operational safety. Also, an integrated conceptual framework offers a robust machine learning tool for risk assessment and management in coal cargo transportation and storage on bulk carrier ships.