An interpretable XGBoost-based approach for Arctic navigation risk assessment.

Area of Science:

• Maritime Safety
• Artificial Intelligence
• Arctic Navigation

Background:

• The Northern Sea Route (NSR) presents a shorter alternative for Europe-Asia shipping compared to southern routes.
• Accelerating global warming and melting Arctic ice are increasing the commercial viability and traffic in the NSR.
• The harsh Arctic environment poses significant navigation risks, necessitating robust safety assessments.

Purpose of the Study:

• To develop and validate models for assessing Arctic navigation risk using actual environmental data and expert judgments.
• To improve the accuracy and reliability of Arctic shipping safety assessments.
• To apply advanced artificial intelligence techniques for enhanced navigation risk evaluation.

Main Methods:

• Generation of a structured dataset using actual Arctic navigation environment data and expert judgments.
• Development of risk assessment models using extreme gradient boosting (XGBoost) and alternative machine learning methods.
• Validation of models using cross-validation and interpretation using Feature Importance (FI) and Shapley Additive Explanations (SHAP).

Main Results:

• XGBoost models demonstrated superior performance over alternative methods, evidenced by lower mean absolute errors and root mean squared errors.
• The XGBoost models effectively learned and reproduced expert judgments regarding Arctic navigation risk.
• Feature Importance (FI) and SHAP analyses provided insights into the relationships between input data and risk predictions.

Conclusions:

• The study validates the effectiveness of XGBoost models in assessing Arctic navigation risk, outperforming conventional methods.
• The integration of AI techniques like XGBoost, FI, and SHAP significantly enhances the quality and robustness of Arctic shipping safety assessments.
• The validated AI-driven approach contributes to improving the overall safety of navigation in the Arctic region.