Machine learning methods for predicting marine port accidents: a case study in container terminal
Üstün Atak, Yasin Arslanoğlu
Abstract
Rapid changes in voyage orders and increased container throughput could lead to undesirable situations such as incidents or accidents in maritime ports. As the demand for maritime transport grows, historical accident reports and data-driven approaches could help to achieve safer and quicker door-to-door transportation. In this scope, the cargo operation data retrieved from the terminal operating system and accident reports are analysed using machine learning classification methods for two sample maritime container terminals located in Turkey. The calculation of accident prediction is studied with features such as vessel capacity, weather information, and cargo handling time. As a validation process, the second container terminal data is used for predicting operation-related accidents. The findings show that XGBoost, LightGBM, and KNN algorithms performed accident prediction with precision metrics of 0.98 for Terminal B and over 0.99–1 for Terminal A amongst the other machine learning classification methods for one-day intervals.