Predicting seaport disruptions from natural hazards using automated machine learning

Seaports play a vital role in global trade, and disruptions caused by natural hazards pose significant risks to operational continuity and economic stability. This study addresses a critical gap by leveraging automated machine learning (AutoML) models to predict seaport disruptions induced by natural hazards. Using data from 2010 to 2019, 24 AutoML models were evaluated across seven categories, with neural networks demonstrating the highest predictive accuracy (94.95 %) and support vector machines the lowest (76.32 %). At the individual level, wide neural networks exhibited the best performance achieving the highest accuracy of 96.44 %. Feature ranking identified "severity" and "recovery" as the most influential factors in disruption, emphasizing the importance of both the magnitude of the hazard and the efficiency of recovery efforts. The Area Under the Curve (AUC) results show strong model performance, particularly in predicting severe disruptions, with most models achieving near-perfect scores for major impact classes. These findings offer valuable insights for enhancing seaport resilience and preparedness. Despite limitations in data availability, the study establishes a strong foundation for future research focused on improving operational resilience and predictive capabilities through the incorporation of more granular data and the application of ensemble models.