A Prediction Model for Maritime Accidents and Associated Risks in Arctic Routes Using Deep Learning

Climate change led to a global effort to seek for new trade routes in the Arctic, significantly altering the maritime sector and increasing maritime activities. Attempt to search and navigate on new routes brought likelihood to a rise in ship accidents, therefore presenting substantial associated risks. Study employed deep learning techniques to predict accidents and outcomes, focusing on the likelihood of occurrences. Accident dataset covering 2005-2017, including variables such as vessel length, age, tonnage, and weather conditions. Dataset was divided into 70% training and 30% testing, using a k-fold cross-validation approach on 511 input-output combinations with 1000 trials each. Results demonstrate that ship tonnage, length, and age are crucial predictors. Highest F1 score (0.89) and lowest standard deviation were achieved using all features. Removing features like minimum daily temperature significantly reduced model performance, reliability improved when combined with weather forecasts. Model can aid planning and management of Arctic maritime operations by predicting associated risks and optimizing insurance premiums. Future research should incorporate additional data sources, test the model under diverse maritime conditions, and focus on specific ship types to develop specialized mitigation strategies. Implementing Polar Code regulations into model predictions can expand the model's applicability and offer insights for policymakers.