Machine Learning With Multi-Source Data to Predict and Explain Marine Pilot Occupational Accidents

Marine pilot occupational accidents during transfer to/from ships are the primary concern of the International Marine Pilots ' Association (IMPA) and industry professionals. There are multiple transfer methods for marine pilots, with the most common being the pilot boat. To reach the mother ship bridge, the following stages must be safely completed: car transfer, walking on the pier, pier to pilot boat, pilot transfer by boat, cutter to pilot ladder, mother ship freeboard climbing, and ship deck to the bridge. Each stage has its own risk. Previous accident records and expert opinions are commonly used to conduct a risk analysis and take preventive actions. However, the reports vary in scope and are often complex, making qualitative analysis a timeintensive task. To overcome this challenge, this study aggregates 500 reports to create a multi-source dataset describing instances of undesired events. A ML (machine learning) approach is used to predict and explain marine pilot occupational accidents. Analyzing the importance of factors distinguishing between accidents, incidents, and non-compliance, we conclude that workplace factors are more dangerous than environmental factors. The findings of this study provide a foundation for developing a unified accident reporting system for predicting accidents on a wider scale.